Geographic Redundancy Configuration

Change SSH Keys - GR Deployment

It is required to modify SSH keys once GR installation/migration is complete.

Before you begin

Make sure diagnostics is clean and there is no alarm/warning.

Procedure

Step 1

On Site 1's Cluster Manager execute the following steps:

  1. Generate new keys.

    /var/qps/install/current/scripts/bin/support/manage_sshkey.sh --create

  2. Update keys on CPS VMs and Installer VM (Cluster Manager).

    /var/qps/install/current/scripts/bin/support/manage_sshkey.sh --update

  3. Backup new SSH keys.

    cd /var/qps/auth; tar -cvf current_ssh_keys.tar current/

Step 2

Transfer /var/qps/auth/current_ssh_keys.tar to Site 2 Cluster Manager.

Step 3

On Site 2 execute the following steps.

  1. Restore SSH keys.

    mkdir -p /var/qps/auth/temp/{qns,root};

  2. Copy tar file to /var/qps/auth/ ; tar -xvf current_ssh_keys.tar.

    cp /var/qps/auth/current/root/* /var/qps/auth/temp/root/ ; cp /var/qps/auth/current/qns/* /var/qps/auth/temp/qns/ ;

  3. Update keys on CPS VMs and Installer VM (Cluster Manager).

    /var/qps/install/current/scripts/bin/support/manage_sshkey.sh --update

MongoDB Authentication - Post Installation

On OpenStack Setup

Disable MongoDB Authentication (CPS downtime is involved)

  1. Modify MongoDB authentication configuration using PATCH API on all the sites (GR setup excluding Arbiter). For more information on PATCH API, refer to CPS Installation Guide for OpenStack.

    Wait for the process to complete.

  2. For arbiter, update the configuration using /etc/facter/facts.d/mongo_auth.txt on all the arbiter sites. 

    db_authentication_enabled=FALSE

    Run /var/qps/install/current/scripts/upgrade/reinit.sh scripts on all the Arbiter sites.

  3. Execute disable MongoDB authentication script only on one site:

    /var/qps/install/current/scripts/modules/mongo_auth_upgrade.py

  4. Restart all the JAVA processes on all the sites.

Enable MongoDB Authentication (CPS downtime is involved)

  1. Modify MongoDB authentication configuration using PATCH API on all the sites (GR setup excluding Arbiter). For more information on PATCH API, refer to CPS Installation Guide for OpenStack.

    Wait for the process to complete.

  2. For arbiter, update the configuration using /etc/facter/facts.d/mongo_auth.txt on all the arbiter sites. 

    db_authentication_enabled=TRUE
db_authentication_admin_passwd=XXXX
db_authentication_readonly_passwd=YYYY

    where, XXXXX and YYYYY are encrypted passwords. For encrypted passwords, you need to SSH to a Cluster Manager and execute the following command:

    /var/qps/bin/support/mongo/encrypt_passwd.sh <Password>

    Run /var/qps/install/current/scripts/upgrade/reinit.sh scripts on all sites including arbiter.

  3. Verify Key file on Session Mgr VMs on all the sites.

    ls /root/.dbkey
/root/.dbkey

  4. Execute enable MongoDB authentication script only on one site:

    /var/qps/install/current/scripts/modules/mongo_auth_upgrade.py

  5. Restart all the JAVA processes on all the sites.

Change MongoDB User Password (CPS downtime is involved)

  1. Modify MongoDB authentication configuration using PATCH API on all the sites (GR setup excluding Arbiter). For more information on PATCH API, refer to CPS Installation Guide for OpenStack.

    Wait for the process to complete.

  2. For arbiter, update the configuration using /etc/facter/facts.d/mongo_auth.txt on all the arbiter sites. 

    db_authentication_enabled=TRUE
db_authentication_admin_passwd=XXXX
db_authentication_readonly_passwd=YYYY

    where, XXXXX and YYYYY are encrypted passwords. For encrypted passwords, you need to SSH to a Cluster Manager and execute the following command:

    /var/qps/bin/support/mongo/encrypt_passwd.sh <Password>

    Run /var/qps/install/current/scripts/upgrade/reinit.sh scripts on all the Arbiter sites.

  3. Update the password only on one site.

    /var/qps/install/current/scripts/modules/mongo_change_password.py <old password>

  4. Restart all the JAVA processes on all the sites.

On VMware Setup

Disable MongoDB Authentication (CPS downtime is involved)

  1. Modify mongo authentication configuration in Configuration.csv file on all the sites (GR setup excluding Arbiter).

    db_authentication_enabled,FALSE

    Update the configuration using /var/qps/install/current/scripts/import/import_deploy.sh script.

  2. For arbiter, update the configuration using /etc/facter/facts.d/mongo_auth.txt on all the arbiter sites. 

    db_authentication_enabled=FALSE

  3. Run /var/qps/install/current/scripts/upgrade/reinit.sh scripts on all the Arbiter sites.

  4. Execute disable MongoDB authentication script only on one site:

    /var/qps/install/current/scripts/modules/mongo_auth_upgrade.py

  5. Restart all the JAVA processes on all the sites.

Enable MongoDB Authentication (CPS downtime is involved)

  1. Modify mongo authentication configuration in Configuration.csv file on all the sites (GR setup excluding Arbiter).

    db_authentication_enabled,TRUE
db_authentication_admin_passwd,XXXX
db_authentication_readonly_passwd,YYYY

    where, XXXXX and YYYYY are encrypted passwords. For encrypted passwords, you need to SSH to a Cluster Manager and execute the following command:

    /var/qps/bin/support/mongo/encrypt_passwd.sh <Password>

    Update the configuration using /var/qps/install/current/scripts/import/import_deploy.sh script.

  2. For arbiter, update the configuration using /etc/facter/facts.d/mongo_auth.txt on all the arbiter sites. 

    db_authentication_enabled=TRUE
db_authentication_admin_passwd=XXXX
db_authentication_readonly_passwd=YYYY

    where, XXXXX and YYYYY are encrypted passwords. For encrypted passwords, you need to SSH to a Cluster Manager and execute the following command:

    /var/qps/bin/support/mongo/encrypt_passwd.sh <Password>

  3. Run /var/qps/install/current/scripts/upgrade/reinit.sh scripts on all the sites including arbiter.

  4. Verify Key file on Session Mgr VMs on all the sites.

    ls /root/.dbkey
/root/.dbkey

  5. Execute enable MongoDB authentication script only on one site:

    /var/qps/install/current/scripts/modules/mongo_auth_upgrade.py

  6. Restart all the JAVA processes on all the sites.

Change MongoDB User Password (CPS downtime is involved)

  1. Modify mongo authentication configuration in Configuration.csv file on all the sites (GR setup excluding Arbiter).

    db_authentication_enabled,TRUE
db_authentication_admin_passwd,XXXX
db_authentication_readonly_passwd,YYYY

    where, XXXXX and YYYYY are encrypted passwords. For encrypted passwords, you need to SSH to a Cluster Manager and execute the following command:

    /var/qps/bin/support/mongo/encrypt_passwd.sh <Password>

    Update the configuration using /var/qps/install/current/scripts/import/import_deploy.sh script.

  2. For arbiter, update the configuration using /etc/facter/facts.d/mongo_auth.txt on all the arbiter sites. 

    db_authentication_enabled=TRUE
db_authentication_admin_passwd=XXXX
db_authentication_readonly_passwd=YYYY

    where, XXXXX and YYYYY are encrypted passwords. For encrypted passwords, you need to SSH to a Cluster Manager and execute the following command:

    /var/qps/bin/support/mongo/encrypt_passwd.sh <Password>

  3. Run /var/qps/install/current/scripts/upgrade/reinit.sh scripts on all the sites including arbiter.

  4. Update the password only on one site.

    /var/qps/install/current/scripts/modules/mongo_change_password.py <old password>

  5. Restart all the JAVA processes on all the sites.

Database Migration Utilities

The database migration utilities can be used to migrate a customer from Active/Standby Geographic Redundancy (GR) environment to Active/Active Geographic Redundancy environment. Currently, the migration utilities support doing remote database lookup based on NetworkId (i.e. MSISDN, IMSI, and so on). The user needs to split the SPR and balance databases from Active/Standby GR model i.e. one for each site in Active/Active GR model.

The workflow for splitting the databases is as follows:

  • Dump the mongoDB data from active site of active/standby system using mongodump command.

  • Run the Split Script on SPR and balance database files collected using mongodump command.

  • Restore the mongo database for each site with mongorestore command using files collected from running Split Script.

After the database splitting is done, you can audit the data by running the Audit Script on each set of site-specific database files separately.

The Split Script is a python script to split SPR and balance database into two site specific parts. The file split.csv is the input file which should have the Network Id regex strings for each site. The Audit Script is a tool to do auditing on the split database files to check for any missing/orphaned records.

To extract the database migration utility, execute the following command:

tar -zxvf /mnt/iso/app/install/xxx.tar.gz -C /tmp/release-train-directory

where, xxx is the release train version.

This command will extract release train into /tmp/release-train-directory.

Split Script

The split script first splits the SPR database into two site-specific SPR databases based on the network_id_key field. Then it loops through the balance database to check which site each balance record correlates to based on the subscriberId field and puts the balance record into one of two site-specific balance databases. If there is no match, then it is considered as Orphaned balance record and added to nositebal.json.

Here are the usage details of the split script:

Usage

python split.py split.csv > output.txt

Prerequisite

The prerequisite to run the script is python-pymongo module. To install python-pymongo on CPS VMs, run the command yum install python-pymongo.

System Requirements

  • RAM: Minimum 1 GB of free memory. The script is memory-intensive and it needs at least 1 GB of RAM to work smoothly.

  • vCPUs: Minimum 4 vCPUs. The script is CPU intensive.

  • Persistent Storage: Free storage is required which should be at least as much as the Active/Standby database file sizes. SSD storage type is preferred (for faster runtimes) but not required.

Input Files

  • The command line argument split.csv is a CSV file that will have network ID regex strings listed per site. The format of each line is site-name, one or more comma-separated regex strings. The regex format is python regex.

    Here is an example of a split.csv file where the networkId regex strings are in the MSISDN Prefix format (i.e. "Starts With" type in Policy Builder configuration).

    site1,5699[0-9]*,5697[0-9]*,5695[0-9]*,5693[0-9]*,5691[0-9]*

    site2,569[86420][0-9]*

    Here is another example where the networkId strings are in the suffix format (i.e. "Ends With" type in Policy Builder configuration).

    site1,^.*[0-4]$

    site2,^.*[5-9]$

    Important

    Since this is a CSV file, using "," in regex strings would result in unexpected behavior, so avoid using "," in regex strings.

  • The script looks for the file subscriber.bson and one or more account.bson files in current directory. The account.bson files could be in nested folders to support a sharded balance database. The balance database could be compressed or uncompressed (the script does not look into the compressed fields).

Output Files

  • site1-balance-mgmt_account.bson

  • site1_spr_subscriber.bson

  • site2-balance-mgmt_account.bson

  • site2_spr_subscriber.bson

In addition there will be following error/debug output files:

  • errorbal.json

  • errorspr.json

  • nositebal.json

  • nositespr.json

Here is the output from a sample run of the split script. 
$ time python split.py split.csv > output.txt
real8m44.015s
user8m0.236s
sys0m35.270s 

$ more output.txt
Found the following subscriber file
./spr/spr/subscriber.bson
Found the following balance files
./balance_mgmt/balance_mgmt/account.bson
./balance_mgmt_1/balance_mgmt_1/account.bson
./balance_mgmt_2/balance_mgmt_2/account.bson
./balance_mgmt_3/balance_mgmt_3/account.bson
./balance_mgmt_4/balance_mgmt_4/account.bson
./balance_mgmt_5/balance_mgmt_5/account.bson
Site1 regex strings: 5699[0-9]*|5697[0-9]*|5695[0-9]*|5693[0-9]*|5691[0-9]*
Site2 regex strings: 569[86420][0-9]* 

Started processing subscriber file

….

….

<snip>

Audit Script

The audit script first goes through the balance database and retrieves a list of IDs. Then it loops through each record in SPR database and tries to match the network_id_key or _id with the ID list from balance database. If there is no match, they are tagged with the counter for Subscribers missing balance records.

Here are the usage details for the audit script:

Usage

python audit.py > output.txt

Prerequisite

The prerequisite to run the script is python-pymongo module. To install python-pymongo on CPS VMs, run the command yum install python-pymongo.

System Requirements

  • RAM: Minimum 1 GB of free memory. The script is memory-intensive and it needs at least 1 GB of RAM to work smoothly.

  • vCPUs: Minimum 4 vCPUs. The script is CPU intensive.

Input Files

The script looks for the file subscriber.bson and one or more account.bson files in current directory. The account.bson files could be in nested folders to support a sharded balance database. The balance database could be compressed or uncompressed (the script does not look into the compressed fields).

Output Files

sprbalmissing.bson

Sample console output from script before splitting SPR and balance databases. 

Total subscriber exceptions: 0
Total subscriber errors: 0
Total subscriber empty records: 1
Total subscriber records: 6743644
Total subscriber matched records: 6733102
Total subscriber missing records: 10541

After running the script on site-specific databases after the split, the user gets the following:

Site1: 
Total subscriber exceptions: 0
Total subscriber errors: 0
Total subscriber empty records: 1
Total subscriber records: 4137817
Total subscriber matched records: 4131978
Total subscriber missing records: 5839
Site2: 
Total subscriber exceptions: 0
Total subscriber errors: 0
Total subscriber empty records: 1
Total subscriber records: 2605826
Total subscriber matched records: 2601124
Total subscriber missing records: 4702

Recovery Procedures

This section covers the following recovery cases in Geographic Redundancy environment:

  • Site recovery after entire site fails.

  • Individual virtual machines recovery.

  • Databases and replica set members recovery.

Site Recovery Procedures

Manual Recovery

When a site fails, it is assumed that other (secondary or standby) site is now operational and has become primary.

Here are the steps to recover the failed site manually:

Procedure

Step 1

Confirm that the databases are in primary/secondary state on running site.

Step 2

Reset the member priorities of failed site so that when the site recovers, these members do not become primary.

  1. Log on to current primary member and reset the priorities by executing the following commands:

    Note

     

    To modify priorities, you must update the members array in the replica configuration object. The array index begins with 0. The array index value is different than the value of the replica set member's members[n]._id field in the array. 

    
ssh <current primary replica set member>
mongo --port <port>
conf=rs.conf()
###here, note the output and note array index value of members for which we want to reset the priorities. 
#Assuming that array index value of members of failed members are 1 and 2,
conf.members[1].priority=2
conf.members[2].priority=3
conf.members[3].priority=5
conf.members[4].priority=4
rs.reconfig(conf)
#Ensure the changed priorities are reflected.
exit

Step 3

Re-configure gateways to make sure that no traffic is sent to failed site during recovery stage.

Step 4

Power on the VMs in the following sequence:

  1. Cluster Manager

  2. pcrfclients

    Stop the following two scripts using monit on both pcrfclients to avoid automatic switchover of databases or automatic stopping of load balancer processes:

    mon_db_for_call_model

    mon_db_for_lb_failover

  3. Session managers

  4. Policy Server (QNS)

  5. Load balancers

Step 5

Synchronize the timestamps between the sites for all VMs by executing the following command from pcrfclient01 of current secondary (recovering) site:

/var/qps/bin/support/sync_times.sh gr

Important

 
The script should be executed only when policy director (lbs) time has been synced (NTP).

Step 6

Confirm that the databases on the failed site completely recovers and become secondary. If they do not become secondary, refer to Database Replica Members Recovery Procedures.

Step 7

After the databases are confirmed as recovered and are secondary, reset these database's priorities using set_priority.sh script from Cluster Manager so that they become primary.

Step 8

If possible, run sample calls and test if recovered site is fully functional or not.

Step 9

Reconfigure the gateways to send the traffic to recovered site.

Step 10

Start mon_db_for_call_model and mon_db_for_lb_failover scripts on both pcrfclients.

Step 11

Monitor the recovered site for stability.

Automatic Recovery

CPS allows you to automatically recover a failed site.

In a scenario where a member fails to recover automatically, use the procedures described in Manual Recovery.

For VMware

For VMware (CSV based installations), execute automated_site_recovery.py script on a failed site. The script recovers the failed replica members that are in RECOVERING or FATAL state. The script is located on Cluster Manager at /var/qps/bin/support/gr_mon/automated_site_recovery.py. The script starts the QNS processes on the Load Balancer VMs, resets the priorities of the replica set, and starts the DB monitor script. However, the script does not alter the state of the VIPs.

If you provide the replica set name as an input parameter, the script recovers the failed member of that replica set.

python /var/qps/bin/support/gr_mon/automated_site_recovery.py --setname <setname>

For example, python /var/qps/bin/support/gr_mon/automated_site_recovery.py --setname set01

If you do not provide any input parameter to the script, the script searches for all replica members from all sets and determines if any of the replica members are in RECOVERING or FATAL state. If yes, the script recovers the members of that replica set.

You can also execute the script with -force. The -force option recovers the replica members that are in RECOVERING, FATAL and STARTUP/STARTUP2 state as well. The script starts the QNS processes on the Load Balancer VMs in the course of recovering the DB Member. However, the script does not alter the state of the VIPs. The -force option must only be used when any database replica member does not come out of STARTUP/STARTUP2 state automatically.

For example: python /var/qps/bin/support/gr_mon/automated_site_recovery.py --setname set01 --force

During recovering of a failed site, if some replica set members do not recover, then such errors are logged in the log file located at /var/log/broadhop/scripts/automated_site_recovery.log.

For Open Stack

The following APIs are used to trigger a recovery script for a failed site.

The logs are located at /var/log/orchestration-api-server.log on the Cluster Manager VM.

/api/site/recover/start

This API is used to trigger a recovery script for a failed site. The API must only be used during a planned maintenance phase. Cluster and database processes may get reset during this process and traffic is affected. This API must only be used when the cluster is in a failed state.

  • Endpoint and Resource: http://<Cluster Manager IP>:8458/api/site/recover/start

    Note

    If HTTPS is enabled, the Endpoint and Resource URL changes from HTTP to HTTPS. For more information, see chapter Installation in CPS Installation Guide for OpenStack.

  • Header: Content-Type: application/json

  • Method: POST

  • Payload: YAML with force and setName fields

    force: true/false

    setName: All replica sets or specific replica set

  • Response: 200 OK: success; 400 Bad Request: The input parameters are malformed or invalid.

  • Example: 

    
   "force": "true "
   "setName": "set01"

/api/system

This API is used to view the status of a recovery process.

  • Endpoint and Resource: http://<Cluster Manager IP>:8458/api/system

    Note

    If HTTPS is enabled, the Endpoint and Resource URL changes from HTTP to HTTPS. For more information, see chapter Installation in CPS Installation Guide for OpenStack.

  • Header: Content-Type: application/json

  • Method: GET

  • Payload: No payload

  • Response: 200 OK: success; 500: Script config not found

  • Example:

    Recovery is currently underway

    "state":"recovering"

    or

    Problem with the recovery

    "state":"error_recovering"

Individual Virtual Machines Recovery

During recovery, the CPS VMs should come UP automatically on reboot without intervention. However, there are scenarios when the VMs will not recover, may be they are unstable or have become corrupt.

The following options exist to recover these CPS VMs:

  • Reinitialize a VM — If a VM is stable but configurations are modified or have become corrupt and one is willing to restore the VM (reset all configurations, else, the configurations can be corrected manually). In that case, execute /etc/init.d/vm-init-client from the VM. Note that if the IP addresses are changed, this command would not recover the same.

  • Redeploy a VM — If current VM is not recoverable, the operator may run the command deploy.sh <vm-name> from the cluster manager. This command will recreate the VM with latest saved configurations.

Database Replica Members Recovery Procedures

CPS database replica members can be recovered automatically or manually.

Automatic Recovery

A replica member holds a copy of the operations log (oplog) that is usually in synchronization with the oplog of the primary member. When the failed member recovers, it starts syncing from previous instance of oplog and recovers. If the member is session_cache whose data is on /tmpfs and if it is recovering from a reboot, the data and oplog has been lost. Therefore, the member resynchronizes all the data from primary's data files first, and then from the primary's oplog.

Verification: Execute diagnostics.sh and verify REPLICA STATE and LAST SYNC status.

  • If REPLICA STATE does not come up as SECONDARY, and stuck into RECOVERING state for longer duration, then follow Manual Recovery. (Refer Verification Step 1)

  • Also, if REPLICA STATE comes up as SECONDARY but does not catch up with PRIMARY and also you can see that replica lag is increasing (in LAST SYNC). (Refer Verification Step 2)

Verification Step 1

Execute diagnostics.sh script to verify if all the members are healthy.

diagnostics.sh --get_replica_status <sitename> 
CPS Diagnostics GR Multi-Node Environment
---------------------------
Checking replica sets...
|--------------------------------------------------------------------------------------------------------------------------------|
| Mongo:3.2.10                         MONGODB REPLICA-SETS STATUS INFORMATION OF site1               Date : 2017-02-20 16:35:30 |
|--------------------------------------------------------------------------------------------------------------------------------|
|  SET NAME  -  PORT  : IP ADDRESS       - REPLICA STATE -           HOST NAME             -  HEALTH  -   LAST SYNC   - PRIORITY |
|--------------------------------------------------------------------------------------------------------------------------------|
| BALANCE:set10                                                                                                                  |
|  Member-1  -  27718 : 172.20.18.54     -  ARBITER      -  arbiter-site3                  - ON-LINE  -   --------    -    0     |
|  Member-2  -  27718 : 172.20.17.40     -  RECOVERING   -  sessionmgr01-site1             - ON-LINE  -   10 sec      -    2     |
|  Member-3  -  27718 : 172.20.17.38     -  RECOVERING   -  sessionmgr02-site1             - ON-LINE  -   10 sec      -    3     |
|  Member-4  -  27718 : 172.20.19.29     -  PRIMARY      -  sessionmgr01-site2             - ON-LINE  -    --------   -    5     |
|  Member-5  -  27718 : 172.20.19.27     -  SECONDARY    -  sessionmgr02-site2             - ON-LINE  -    0 sec      -    4     |
|--------------------------------------------------------------------------------------------------------------------------------|

Note

If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

Also, you can login to mongo on that member and check its actual status.
Verification Step 2

  1. Execute diagnostics.sh script to verify if all the members are healthy.

    diagnostics.sh --get_replica_status <sitename> 
    CPS Diagnostics GR Multi-Node Environment
---------------------------
Checking replica sets...
|--------------------------------------------------------------------------------------------------------------------------------|
| Mongo:3.2.10                         MONGODB REPLICA-SETS STATUS INFORMATION OF site1               Date : 2017-02-20 16:35:30 |
|--------------------------------------------------------------------------------------------------------------------------------|
|  SET NAME  -  PORT  : IP ADDRESS       - REPLICA STATE -           HOST NAME             -  HEALTH  -   LAST SYNC   - PRIORITY |
|--------------------------------------------------------------------------------------------------------------------------------|
| BALANCE:set10                                                                                                                  |
|  Member-1  -  27718 : 172.20.18.54     -  ARBITER      -  arbiter-site3                  - ON-LINE  -   --------    -    0     |
|  Member-2  -  27718 : 172.20.17.40     -  SECONDARY    -  sessionmgr01-site1             - ON-LINE  -    10 sec     -    2     |
|  Member-3  -  27718 : 172.20.17.38     -  SECONDARY    -  sessionmgr02-site1             - ON-LINE  -    10 sec     -    3     |
|  Member-4  -  27718 : 172.20.19.29     -  PRIMARY      -  sessionmgr01-site2             - ON-LINE  -    --------   -    5     |
|  Member-5  -  27718 : 172.20.19.27     -  SECONDARY    -  sessionmgr02-site2             - ON-LINE  -    0 sec      -    4     |
|--------------------------------------------------------------------------------------------------------------------------------|

    Note

    If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

    Also, you can login to mongo on that member and check its actual status.

  2. Execute the following command from mongo CLI (PRIMARY member) to verify if replica lag is increasing: 

    mongo sessionmgr01-site2:27718

set10:PRIMARY> rs.printSlaveReplicationInfo()
    If it is observed that the lag is increasing, then run rs.status to check for any exception. Refer to Manual Recovery to fix this issue. 
    mongo —-host sessionmgr01-site2 port 27718

set10:PRIMARY> rs.status()
{
"set" : "set10",
"date" : ISODate("2017-02-15T07:21:35.367Z"),
"myState" : 2,
"term" : NumberLong(-1),
"heartbeatIntervalMillis" : NumberLong(2000),
"members" : [
{
"_id" : 0,
"name" : "arbiter-site3:27718",
"health" : 1,
"state" : 7,
"stateStr" : "ARBITER",
"uptime" : 28,
"lastHeartbeat" : ISODate("2017-02-15T07:21:34.612Z"),
"lastHeartbeatRecv" : ISODate("2017-02-15T07:21:34.615Z"),
"pingMs" : NumberLong(150),
"configVersion" : 2566261
},
{
"_id" : 1,
"name" : "sessionmgr01-site1:27718",
"health" : 1,
"state" : 2,
"stateStr" : "SECONDARY",
"uptime" : 28,
"optime" : Timestamp(1487066061, 491),
"optimeDate" : ISODate("2017-02-14T09:54:21Z"),
"lastHeartbeat" : ISODate("2017-02-15T07:21:34.813Z"),
"lastHeartbeatRecv" : ISODate("2017-02-15T07:21:34.164Z"),
"pingMs" : NumberLong(0),
"lastHeartbeatMessage" : "could not find member to sync from",
"configVersion" : 2566261
},
{
"_id" : 2,
"name" : "sessionmgr02-site1:27718",
"health" : 1,
"state" : 2,
"stateStr" : "SECONDARY",
"uptime" : 30,
"optime" : Timestamp(1487066061, 491),
"optimeDate" : ISODate("2017-02-14T09:54:21Z"),
"infoMessage" : "could not find member to sync from",
"configVersion" : 2566261,
"self" : true
},
{
"_id" : 3,
"name" : "sessionmgr01-site2:27718",
"health" : 1,
"state" : 1,
"stateStr" : "PRIMARY",
"uptime" : 28,
"optime" : Timestamp(1487145333, 99),
"optimeDate" : ISODate("2017-02-15T07:55:33Z"),
"lastHeartbeat" : ISODate("2017-02-15T07:21:34.612Z"),
"lastHeartbeatRecv" : ISODate("2017-02-15T07:21:34.603Z"),
"pingMs" : NumberLong(150),
"electionTime" : Timestamp(1487066067, 1),
"electionDate" : ISODate("2017-02-14T09:54:27Z"),
"configVersion" : 2566261
},
{
"_id" : 4,
"name" : "sessionmgr02-site2:27718",
"health" : 1,
"state" : 2,
"stateStr" : "SECONDARY",
"uptime" : 28,
"optime" : Timestamp(1487145333, 95),
"optimeDate" : ISODate("2017-02-15T07:55:33Z"),
"lastHeartbeat" : ISODate("2017-02-15T07:21:34.613Z"),
"lastHeartbeatRecv" : ISODate("2017-02-15T07:21:34.599Z"),
"pingMs" : NumberLong(150),
"syncingTo" : "sessionmgr01-site2:27718",
"configVersion" : 2566261
}
],
"ok" : 1
}

In a scenario where a member fails to recover automatically, the operator should use procedures described in Manual Recovery.

Manual Recovery

Before performing recovery steps, refer to the following guidelines:

  • Perform the recovery steps in a maintenance window on any production system. These recovery steps require restarts of the application.

  • If a failure impacts the system for a long period (for example, data center, power or hardware failure) the database instance must be resynchronized manually as the oplog will have rolled over. Full resynchronizations of the database are considered events that operation teams should execute during maintenance windows with personnel monitoring the status of the platform.

  • In Geo Redundancy, replica sets are used for different databases. The replication happens based on oplog. The oplog is a data structure that mongo maintains internally at Primary where the data operations logs are maintained. The secondaries fetch from the oplog entries from the primary asynchronously and apply those operations on themselves to achieve synchronization. If a secondary goes down for a long time, due to the limited size of oplog, there is a chance that some of the logs in oplog will be overwritten by new entries. In that case, when secondary comes up, it is unable to synchronize with the primary as it does not see a timestamp from where it had gone down.

Therefore, manual resynchronization is required which is termed as initial-sync in MongoDB. In this scenario, mongod process is stopped on concerned secondary, all the data in data directory is deleted and mongod process is started. The secondary Session Manager first copies all the data from primary and then copies the oplog.


Note

These procedures are only for manually recovering the databases. Based on the system status (all VMs down, traffic on other site, LBs down or up, all session cache down or only one down etc.), execution of some pre- and post- tests may be required. For example, if only one session manager is to be recovered, and we have primary database and traffic on current site, we must not reset the database priorities.

Similarly, if all of the CPS databases and load balancers are down, monit processes corresponding to mon_db_for_lb_failover and mon_db_for_call_model scripts should be stopped. These scripts monitor load balancers and if LB processes or LB itself are down, they make local instances of databases secondary. Also, if local databases are secondary, these scripts shut down load balancer process. All these scripts refer to corresponding configurations in /etc/broadhop. Also, post recovery, user can run some sample calls on recovered site to make sure that system is stable, and then finally migrate traffic back to original Primary.

This following sections provide the detailed steps to recover a MongoDB when the database replica set member does not recover by itself:

Recovery Using Repair Option

The repair option can be used when few members have not recovered due to VM reboot or abrupt VM shutdown or some other problem.

Procedure

Step 1

Execute the diagnostics script (on pcrfclient01/02) to know which replica set or respective member has failed.

For Site1:

#diagnostics.sh --get_replica_status site1

For Site2:

#diagnostics.sh --get_replica_status site2

Note

 

If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

Also, you can login to mongo on that member and check its actual status.

Step 2

Log onto session manager VM and check if mongod process is running or not.

#ps -ef | grep 27720

Note

 

Port number can be different.

Step 3

If process is running, then shut down the process and try to repair the database.

  1. To stop the process.

    /usr/bin/systemctl stop sessionmgr-<port#>

  2. To repair database.

    /usr/bin/systemctl repair sessionmgr-<port#>

    Sometimes the repair process takes time to recover. Check the mongo log to find the status:

    #tailf /var/log/mongodb-<port#>.log

  3. If the repair process is completed successfully, then start mongo process.

    /usr/bin/systemctl start sessionmgr-<port#>

Step 4

Execute the diagnostics script again (on pcrfclient01/02) to know if replica set member has recovered.

For Site1:

#diagnostics.sh --get_replica_status site1

For Site2:

#diagnostics.sh --get_replica_status site2

Note

 

If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

Also, you can login to mongo on that member and check its actual status.

Step 5

To recover other failed members, follow the recovery steps from Step 1 to Step 4.

If the secondary member is still in RECOVERING state, refer Recovery Using Remove/Add Members Option.

Recovery Using Remove/Add Members Option
Remove Specific Members

Caution

Before removing the particular member from the replica set, make sure that you have identified correct member.

Sometimes a member lags behind significantly due to failure or network issues, and is unable to resync. In such cases, remove that member from replica set and add it again so that it can resync from start and come up.

Procedure

Step 1

Log in to Cluster Manager.

Step 2

Execute the diagnostic script to know which replica set or respective member needs to be removed.

For Site1:

#diagnostics.sh --get_replica_status site1

For Site2:

#diagnostics.sh --get_replica_status site2

Note

 

If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

Also, you can login to mongo on that member and check its actual status.

Step 3

Execute build_set.sh with the port option to remove particular member from replica set. Script prompts to enter <VM>:<port> where member resides.

cd /var/qps/bin/support/mongo/

For session database:

#./build_set.sh --session --remove-members

For SPR database:

#./build_set.sh --spr --remove-members

Step 4

Execute the diagnostic script again to verify if that particular member is removed.

For Site1:

#diagnostics.sh --get_replica_status site1

For Site2:

#diagnostics.sh --get_replica_status site2

Note

 

If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

Also, you can login to mongo on that member and check its actual status.
Add Members

To add the earlier removed members to replica set, perform the following steps:

Procedure

Step 1

Log in to Cluster Manager.

Step 2

Execute the diagnostic script to know which replica set member is not in configuration or failed member.

For Site1:

diagnostics.sh --get_replica_status site1

For Site2:

diagnostics.sh --get_replica_status site2

Note

 

If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

Also, you can login to mongo on that member and check its actual status.

Step 3

Update /etc/broadhop/mongoConfig.cfg file. Execute build_etc.sh to accept the changes done in /etc/broadhop/mongoConfig.cfg file and wait for AIDO server to create the replica-set.

cd /var/qps/bin/support/mongo/

To verify the replica-set has been created, run the following command for session database:

#./build_set.sh --session

OR

diagnostics.sh --get_replica_status

To verify whether the replica-set are created, run the following command for SPR database:

#./build_set.sh --spr

OR

diagnostics.sh --get_replica_status

Step 4

Set priority using set_priority.sh command. The following are example commands: 

cd /var/qps/bin/support/mongo/; ./set_priority.sh --db session
cd /var/qps/bin/support/mongo/; ./set_priority.sh --db spr
cd /var/qps/bin/support/mongo/; ./set_priority.sh --db admin
cd /var/qps/bin/support/mongo/; ./set_priority.sh --db balance

Step 5

Execute the diagnostic script from Cluster Manager to know if member(s) are added successfully into the replica set.

For Site1:

#diagnostics.sh --get_replica_status site1

For Site2:

#diagnostics.sh --get_replica_status site2

Note

 

If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

Also, you can login to mongo on that member and check its actual status.
Recovery for High TPS

When the HA/GR setup is running with high TPS and if replica members are having high latency between them then some replica members can go to RECOVERING state and will not recover from that state unless some commands are executed to recover those replica members. We can use the manual/automated recovery procedure to recover the replica members which are in RECOVERING state.

Automated Recovery

There can be three different scenarios of setup possible for recovery of replica members:

  1. Case 1: Two members of replica set are in RECOVERING state

  2. Case 2: With all replica members except primary are in RECOVERING state

  3. Case 3: Some replica members are in RECOVERING state


Note

Automation script recovers only those replica members which are in RECOVERING state.

Procedure

Step 1

Before executing automated recovery script (high_tps_db_recovery.sh <replica_setname>), go to current primary member (site-1) and reset the priorities by executing the following commands:

Note

 

To modify priorities, you must update the members array in the replica configuration object. The array index begins with 0. The array index value is different than the value of the replica set member's members[n]._id field in the array. 

ssh <primary member>
mongo --port <port>
conf=rs.conf()
conf.members[1].priority=2
conf.members[2].priority=3
conf.members[3].priority=5
conf.members[4].priority=4
rs.reconfig(conf)
exit

Step 2

Execute the following command script to recover the member:

high_tps_db_recovery.sh <replica_setname>

For Example:

high_tps_db_recovery.sh SPR-SET1

Step 3

Execute diagnostics.sh command to check whether the RECOVERING member has recovered.

diagnostics.sh --get_replica_status

Note

 

If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

Also, you can login to mongo on that member and check its actual status.
After the replica set member is recovered, the state will change to SECONDARY and all the process logs are stored in a log file.

Note

If you are unable to recover the replica set member from RECOVERING state using automated recovery script, refer to Manual Recovery.

Manual Recovery
Procedure

Step 1

Before recovery, on all concerned replica-set VMs, perform the following steps:

  1. Edit sshd_config file.

    vi /etc/ssh/sshd_config

  2. Add the following entry at the end of sshd_config file. The below value (130) should be based on number of files that we have under secondary's data directory. It should be close to the number of files there.

    MaxStartups 130

  3. Restart sshd service by executing the following command:

    service sshd restart

  4. Execute diagnostics.sh --get_replica_status command to know which members are down.

    Based on the status of system and subject to above note, check if you need to reset member priorities. For example, if site-1 is Primary and site-2 is Secondary and if site-1 has gone down, we need to login to new Primary and reset the replica members priorities in such a way that when site-1 comes UP again, it would not become Primary automatically.

    Note

     

    If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

    Also, you can login to mongo on that member and check its actual status.

    For this purpose, perform the following steps:

    1. Go to current primary member (site-1) and reset the priorities by executing the following commands:

      Note

       

      To modify priorities, you must update the members array in the replica configuration object. The array index begins with 0. The array index value is different than the value of the replica set member's members[n]._id field in the array. 

      ssh <primary member>
mongo --port <port>
conf=rs.conf()
conf.members[1].priority=2
conf.members[2].priority=3
conf.members[3].priority=5
conf.members[4].priority=4
rs.reconfig(conf)
exit

    2. Also on the failed replica set site there are chances that monitoring scripts would have stopped the load balancers, and shifted all the databases primaries to other site. Stop the monitoring on the failed site pcrfclient01 and pcrfclient02 both by executing the following commands: 

      monit stop mon_db_for_lb_failover 
monit stop mon_db_for_callmodel

      At this point the operator should maintain two consoles: one for executing commands to recover the secondary member and another to manage the source secondary. The source Secondary is the Secondary that is nearest in terms of latency from the recovering Secondary.

      Also, note down the port and data directory for these members. Typically these are the same, and only the host name will be different.

Step 2

Recover the member:

  1. Go to recovering Secondary and execute the following commands: 

    ssh <recovering Secondary>
ps -eaf | grep mongo
/usr/bin/systemctl stop sessionmgr-<port>
cd <member data directory>
\rm -rf *
cp /var/qps/bin/support/gr_mon/fastcopy.sh

  2. Go to nearest working available secondary and execute the following commands: 

    ssh <source Secondary> mongo --port <mongo port>
#lock this secondary from writes 
db.fsyncLock()
exit
ps -eaf | grep mongo 
cd <data directory>
tar -cvf _tmp.tar _tmp

    Any errors can be ignored.

    tar -cvf rollback.tar rollback

  3. Go to recovering Secondary and execute the following commands: 

    cd <data directory>
./fastcopy.sh <nearest working secondary>
<secondary data directory> 
ps -eaf | grep scp | wc -l

  4. After the count is one, start secondary by executing the following commands: 

    tar -xvf _tmp.tar
tar -xvf rollback.tar
/usr/bin/systemctl start sessionmgr-<port>

  5. Go to nearest secondary from where you are recovering and execute the following commands: 

    mongo --port <port>
db.fsyncUnlock()
db.printSlaveReplicationInfo()

    Exit the database by executing exit command.

    Monitor the lag for some time (close to 3 to 4 minutes). Initially the lag will be small, later it will increase and then decrease. This is because secondary is catching up with primary oplog and also calculating the lag. As the secondary has just restarted, it takes sometime to calculate real lag, but MongoDB shows intermittent values, hence, we see the lag initially increasing. On the other hand, the secondary is also catching up on synchronization and eventually the lag would reduce to one second or less. The member should become secondary soon.

    On similar lines, recover another secondary, preferably from the just recovered one if it is closest in terms of ping connectivity.

    Once all the secondaries are recovered, we need to reset the priorities and then restart the stopped load balancers.

  6. Connect to primary of concerned replica set: 

    ssh <primary member>
mongo --port <port>
conf=rs.conf()

    Based on the output, carefully identify the correct members and their ids:

    Note

     

    To modify priorities, you must update the members array in the replica configuration object. The array index begins with 0. The array index value is different than the value of the replica set member's members[n]._id field in the array. 

    
conf.members[1].priority=5
conf.members[2].priority=4
conf.members[3].priority=3
conf.members[3].priority=2
rs.reconfig(conf)
exit

Step 3

Log in to lb01 and lb02 and start monit for all qns processes.

Step 4

Check the status of qns processes on each of load balancers VMs by executing the following command:

monit status qnsXX

Step 5

Now login to pcrfclient01 and 02 and start the monit scripts that we had stopped earlier. 

monit start mon_db_for_lb_failover
monit start mon_db_for_callmodel

Now reset the sshd connection on all virtual machines. Comment out the MaxStartup line in /etc/ssh/sshd_conf file and restart sshd using service sshd restart command.

Rebuild Replica Set

There could be a situation when all replica set members go into recovery mode and none of members are either in primary or secondary state.

Procedure

Step 1

Stop CPS processes.

Step 2

Log in to Cluster Manager.

Step 3

Execute the diagnostic script to know which replica set (all members) have failed.

For Site1:

#diagnostics.sh --get_replica_status site1

For Site2:

#diagnostics.sh --get_replica_status site2

The output displays which replica set members of replica set set01 for session data are in bad shape.

Note

 

If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

Also, you can login to mongo on that member and check its actual status.

Step 4

Build session replica sets. Add the member information in /etc/broadhop/mongoConfig.cfg file and run build_etc.sh script and wait for AIDO server to create the replica-set.

To verify the replica-set has been created, run the following command:

# ./build_set.sh --session

OR

diagnostics.sh --get_replica_status

Step 5

Set priority using set_priority.sh command. The following are example commands: 

cd /var/qps/bin/support/mongo/; ./set_priority.sh --db session
cd /var/qps/bin/support/mongo/; ./set_priority.sh --db spr
cd /var/qps/bin/support/mongo/; ./set_priority.sh --db admin
cd /var/qps/bin/support/mongo/; ./set_priority.sh --db balance

Step 6

To recover other failed set, follow the recovery steps from Step 1 toStep 4.

Step 7

Restart CPS.

restartall.sh

Caution

 

Executing restartall.sh will cause messages to be dropped.

Add New Members to the Replica Set
Procedure

Step 1

Log in to Cluster Manager.

Step 2

Execute the diagnostic script to know which replica-set member is not in configuration or failed member.

For Site1:

#diagnostics.sh --get_replica_status site1

For Site2:

#diagnostics.sh --get_replica_status site2

Note

 

If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

Also, you can login to mongo on that member and check its actual status.

Step 3

Update the member information in the mongoConfig.cfg file.

Example: The following is the example for adding two members in balance replica-set. 

cd /etc/broadhop
vi mongoConfig.cfg
Before Update After Update

[BALANCE-SET1] SETNAME=set03

ARBITER1=pcrfclient01-prim-site-1:37718

ARBITER_DATA_PATH=/data/sessions.3

PRIMARY-MEMBERS

MEMBER1=sessionmgr01-site-1:27718

SECONDARY-MEMBERS

MEMBER1=sessionmgr01-site-2:27718

DATA_PATH=/data/sessions.3

[BALANCE-SET1-END]

[BALANCE-SET1] SETNAME=set03

ARBITER1=pcrfclient01-prim-site-1:37718

ARBITER_DATA_PATH=/data/sessions.3

PRIMARY-MEMBERS

MEMBER1=sessionmgr01-site-1:27718

MEMBER2=sessionmgr02-site-1:27718

SECONDARY-MEMBERS

MEMBER1=sessionmgr01-site-2:27718

MEMBER2=sessionmgr02-site-2:27718

DATA_PATH=/data/sessions.3

[BALANCE-SET1-END]

Run build_etc.sh to accept the changes done in /etc/broadhop/mongoConfig.cfg file and wait for AIDO server to add the new replica-set

cd /var/qps/bin/support/mongo/

Example: To verify the replica-set members has been added to balance database, run the following command: 

./build_set.sh --balance

OR

diagnostics.sh --get_replica_status

Step 4

Execute the diagnostic script to know if member/s are added successfully into the replica-set.

For Site1:

#diagnostics.sh --get_replica_status site1

For Site2:

#diagnostics.sh --get_replica_status site2

Note

 

If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

Also, you can login to mongo on that member and check its actual status.

Additional Session Replication Set on GR Active/Active Site

The objective of this section is to add one additional session replication set in GR Active/Active site.

The steps mentioned in this section needs to be executed from primary site Cluster Manager.


Note

The steps in this section assume that the engineer performing the steps has SSH and VMware vCenter access to the production PCRF System. No impact to traffic is foreseen during the implementation of the steps although there might be a slight impact on response time during rebalance CLI.


Note

In case there is CPS blade expansion, Max Timer T P S parameter in Policy Buider must to be tuned to handle internal RAR TPS. For more information, contact your Cisco Account representative.

Before you begin

You should run all the sanity checks prior and after executing the steps in this section.

Procedure

Step 1

Run diagnostics.sh to verify that the system is in healthy state.

Step 2

Log in to primary Cluster Manager using SSH.

Step 3

Take the backup of /etc/broadhop/mongoConfig.cfg file.

cp /etc/broadhop/mongoConfig.cfg /etc/broadhop/mongoConfig.cfg.date.BACKUP

Step 4

Take the backup of admin database from Cluster Manager. 

[root@cm-a ~]# mkdir admin
[root@cm-a ~]# cd admin
[root@cm-a admin]# mongodump -h sessionmgr01 --port 27721
connected to: sessionmgr01:27721
2016-09-23T16:31:13.962-0300 all dbs
** Truncated output **

Step 5

Edit /etc/broadhop/mongoConfig.cfg file using vi editor. Find the section for session replication set. Add the new session replication set members.

Note

 

Server name and ports are specific to each customer deployment. Make sure that new session replication set has unique values.

Session set number must be incremented.

Make sure the port used in MEMBER1, MEMBER2, MEMBER3, MEMBER4, and so on are same. 

#SITE1_START
[SESSION-SET2]
SETNAME=set10
OPLOG_SIZE=1024
ARBITER1=pcrfclient01a:27727
ARBITER_DATA_PATH=/var/data/sessions.1/set10
MEMBER1=sessionmgr01a:27727
MEMBER2=sessionmgr02a:27727
MEMBER3=sessionmgr01b:27727
MEMBER4=sessionmgr02b:27727
DATA_PATH=/var/data/sessions.1/set10
[SESSION-SET2-END]

#SITE2_START
[SESSION-SET5]
SETNAME=set11
OPLOG_SIZE=1024
ARBITER1=pcrfclient01b:47727
ARBITER_DATA_PATH=/var/data/sessions.1/set11
MEMBER1=sessionmgr01b:37727
MEMBER2=sessionmgr02b:37727
MEMBER3=sessionmgr01a:37727
MEMBER4=sessionmgr02a:37727
DATA_PATH=/var/data/sessions.1/set11
[SESSION-SET5-END]

Run build_etc.sh to accept the changes done in mongoConfig.cfg file and wait for AIDO server to create the additonal replica-set.

Note

 

Verify that the /etc/hosts file on the both sites is correctly configured with alias.

Site1 /etc/hosts file should have the following content: 

x.x.x.a sessionmgr01 sessionmgr01a
x.x.x.b sessionmgr02 sessionmgr02a
y.y.y.a psessionmgr01 sessionmgr01b
y.y.y.b psessionmgr02 sessionmgr02b

Site2 /etc/hosts file should have the following content: 

y.y.y.a sessionmgr01 sessionmgr01b
y.y.y.b sessionmgr02 sessionmgr02b
x.x.x.a psessionmgr01 sessionmgr01a
x.x.x.b psessionmgr02 sessionmgr02a

Step 6

SSH to Cluster-A/Site1 Cluster Manager. Add the new session replication set information in /etc/broadhop/mongoConfig.cfg file. Run build_etc.sh to accept the changes and create new session replication set from Cluster Manager.

To verify the replica-set has been created, run the following command:

build_set.sh --session

OR

diagnostics.sh --get_replica_status

Step 7

Set priority using set_priority.sh command. The following are example commands: 

cd /var/qps/bin/support/mongo/; ./set_priority.sh --db session
cd /var/qps/bin/support/mongo/; ./set_priority.sh --db spr
cd /var/qps/bin/support/mongo/; ./set_priority.sh --db admin
cd /var/qps/bin/support/mongo/; ./set_priority.sh --db balance

Step 8

Add shard to Cluster-B/Site2. Add the new session replication set information in /etc/broadhop/mongoConfig.cfg file. Run build_etc.sh to accept the changes and create new session replication set from Cluster Manager.

To verify the replica-set has been created, run the following command:

build_set.sh --session

OR

diagnostics.sh --get_replica_status

Step 9

Set priority using set_priority.sh command. The following are example commands: 

cd /var/qps/bin/support/mongo/; ./set_priority.sh --db session
cd /var/qps/bin/support/mongo/; ./set_priority.sh --db spr
cd /var/qps/bin/support/mongo/; ./set_priority.sh --db admin
cd /var/qps/bin/support/mongo/; ./set_priority.sh --db balance

Step 10

Copy mongoConfig.cfg file to all the nodes using copytoall.sh from Cluster Manager.

copytoall.sh /etc/broadhop/mongoConfig.cfg /etc/broadhop/mongoConfig.cfg 

Copying '/var/qps/config/mobile/etc/broadhop/mongoConfig.cfg' 
to '/etc/broadhop/mongoConfig.cfg' on all VMs 
lb01 
mongoConfig.cfg                                                                                                                                                                                           100% 4659     4.6KB/s   00:00    
lb02
mongoConfig.cfg                                                                                                                                                                                           100% 4659     4.6KB/s   00:00    
sessionmgr01
** Truncated output **

Step 11

Transfer the modified mongoConfig.cfg file to Site2 (Cluster-B).

scp /etc/broadhop/mongoConfig.cfg cm-b:/etc/broadhop/mongoConfig.cfg 

root@cm-b's password: 
mongoConfig.cfg                                                                                                                                                                                           100% 4659     100% 4659     4.6KB/s   00:00

Step 12

SSH Cluster-B (Cluster Manager). Run build_etc.sh to make sure modified mongoConfig.cfg file is restored after the reboot.

/var/qps/install/current/scripts/build/build_etc.sh 

Building /etc/broadhop...
Copying to /var/qps/images/etc.tar.gz...
Creating MD5 Checksum...

Step 13

Copy mongoConfig.cfg file from Cluster-B (Cluster Manager) to all the nodes using copytoall.sh from Cluster Manager.

copytoall.sh /etc/broadhop/mongoConfig.cfg /etc/broadhop/mongoConfig.cfg 

Copying '/var/qps/config/mobile/etc/broadhop/mongoConfig.cfg' 
to '/etc/broadhop/mongoConfig.cfg' on all VMs
lb01
mongoConfig.cfg                                                                                                                                                                                           100% 4659     100% 4659     4.6KB/s   00:00    
lb02
mongoConfig.cfg                                                                                                                                                                                           100% 4659     100% 4659     4.6KB/s   00:00        
** Truncated output **

Step 14

(Applicable for HA and Active/Standby GR only) Adding shards default option. Login to OSGi mode and add the shards as follows:

telnet qns01 9091 
Trying XXX.XXX.XXX.XXX...
Connected to qns01.
Escape character is '^]'.

addshard seed1[,seed2] port db-index siteid [backup]

osgi> addshard sessionmgr01,sessionmgr02 27727 1 Site1

osgi> addshard sessionmgr01,sessionmgr02 27727 2 Site1

osgi> addshard sessionmgr01,sessionmgr02 27727 3 Site1

osgi> addshard sessionmgr01,sessionmgr02 27727 4 Site1

osgi> addshard sessionmgr01,sessionmgr02 37727 1 Site1

osgi> addshard sessionmgr01,sessionmgr02 37727 2 Site1

osgi> addshard sessionmgr01,sessionmgr02 37727 3 Site1

osgi> addshard sessionmgr01,sessionmgr02 37727 4 Site1

osgi> rebalance

osgi> migrate 

Migrate ...
All versions up to date - migration starting

Step 15

Verify that the sessions have been created in the newly created replication set and are balanced.

session_cache_ops.sh --count site2

session_cache_ops.sh --count site1

Sample output: 
Session cache operation script
Thu Jul 28 16:55:21 EDT 2016
------------------------------------------------------
Session Replica-set SESSION-SET4
------------------------------------------------------
Session Database          : Session Count
------------------------------------------------------
 session_cache            : 1765
 session_cache_2          : 1777
 session_cache_3          : 1755
 session_cache_4          : 1750
------------------------------------------------------
 No of Sessions in SET4   : 7047
------------------------------------------------------

------------------------------------------------------
Session Replica-set SESSION-SET5
------------------------------------------------------
Session Database          : Session Count
------------------------------------------------------
 session_cache            : 1772
 session_cache_2          : 1811
 session_cache_3          : 1738
 session_cache_4          : 1714
------------------------------------------------------
 No of Sessions in SET5   : 7035
------------------------------------------------------

Step 16

(Applicable for Active/Active GR only) Add shards with Site option. Login to OSGi mode and add the shards as follows:

Note

 
This process is adding shards for Active/Active GR setup which has Site options enabled. To enable Geo-HA feature, refer to Active/Active Geo HA - Multi-Session Cache Port Support. 
telnet qns01 9091
Trying XXX.XXX.XXX.XXX...
Connected to qns01.
Escape character is '^]'.

Run listsitelookup if you are unsure about the site names. Similar information can be obtained from /etc/broadhop/qns.conf file (-DGeoSiteName=Site1).

Note

 

In listsitelookup configuration ‘LookupValues’ must be unique per PrimarySiteId. In case if they are not unique, you must enable FTS (Full Table Scan) which will impact the performance. 

osgi> listsitelookup

                  Id           PrimarySiteId         SecondarySiteId            LookupValues
                   1                   Site1                   Site2     pcef-gx-1.cisco.com
                   1                   Site1                   Site2     pcef-gy-1.cisco.com
                   1                   Site1                   Site2    ocs1.server.cisco.com
                   2                   Site2                   Site1    pcef2-gx-1.cisco.com
                   2                   Site2                   Site1    pcef2-gy-1.cisco.com
                   2                   Site2                   Site1    ocs1.server.cisco.com

Note

 

Do not run addshard command on multiple sites in parallel. Wait for the command to finish on one site and then proceed to second site.

Adding shard to Site1. Run the following command from the qns of Site1: 

osgi> addshard sessionmgr01,sessionmgr02 27727 1 Site1

osgi> addshard sessionmgr01,sessionmgr02 27727 2 Site1

osgi> addshard sessionmgr01,sessionmgr02 27727 3 Site1

osgi> addshard sessionmgr01,sessionmgr02 27727 4 Site1

Adding shards to Site2. Run the following command from the qns of Site2: 

osgi> addshard sessionmgr01,sessionmgr02 37727 1 Site2

osgi> addshard sessionmgr01,sessionmgr02 37727 2 Site2

osgi> addshard sessionmgr01,sessionmgr02 37727 3 Site2

osgi> addshard sessionmgr01,sessionmgr02 37727 4 Site2

Run osgi> rebalance Site1 command from Site1 qns.

Run osgi> rebalance Site2 command from Site2 qns.

Run the following command from the Site1 qns: 

osgi> migrate Site1
Migrate ...
All versions up to date - migration starting

Run the following command from the Site2 qns: 

osgi> migrate Site2
Migrate ...
All versions up to date - migration starting

Step 17

Verify that the sessions have been created in the newly created replication set and are balanced.

session_cache_ops.sh --count site2

session_cache_ops.sh --count site1

Sample output: 
Session cache operation script
Thu Jul 28 16:55:21 EDT 2016
------------------------------------------------------
Session Replica-set SESSION-SET4
------------------------------------------------------
Session Database          : Session Count
------------------------------------------------------
 session_cache            : 1765
 session_cache_2          : 1777
 session_cache_3          : 1755
 session_cache_4          : 1750
------------------------------------------------------
 No of Sessions in SET4   : 7047
------------------------------------------------------

------------------------------------------------------
Session Replica-set SESSION-SET5
------------------------------------------------------
Session Database          : Session Count
------------------------------------------------------
 session_cache            : 1772
 session_cache_2          : 1811
 session_cache_3          : 1738
 session_cache_4          : 1714
------------------------------------------------------
 No of Sessions in SET5   : 7035
------------------------------------------------------

Step 18

Secondary Key Ring Configuration: This step only applies If you are adding additional session replication set to a new session manager server. Assuming that existing setup has the secondary key rings configured for existing session Replication servers.

Refer to the section Secondary Key Ring Configuration in CPS Installation Guide for VMware.

Step 19

Configure session replication set priority from Cluster Manager.

cd /var/qps/bin/support/mongo/; ./set_priority.sh --db session

Step 20

Verify whether the replica set status and priority is set correctly by running the following command from Cluster Manager:

diagnostics.sh --get_replica_status 
|-------------------------------------------------------------------------------------------------------------|
| SESSION:set10                                                                                                    |
|  Member-1  -  27727 : 192.168.116.33   -  ARBITER      -  pcrfclient01a    - ON-LINE  -   --------    -    0     |
|  Member-2  -  27727 : 192.168.116.71   -  PRIMARY      -  sessionmgr01a    - ON-LINE  -   --------    -    5     |
|  Member-3  -  27727 : 192.168.116.24   -  SECONDARY    -  sessionmgr02a    - ON-LINE  -    0 sec      -    4     |
|  Member-4  -  27727 : 192.168.116.70   -  SECONDARY    -  sessionmgr01b    - ON-LINE  -    0 sec      -    3     |
|  Member-5  -  27727 : 192.168.116.39   -  SECONDARY    -  sessionmgr02b    - ON-LINE  -    0 sec      -    2     |
|---------------------------------------------------------------------------------------------------------------------|

Note

 

If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

Also, you can login to mongo on that member and check its actual status.

Step 21

Run diagnostics.sh to verify whether the priority for new replication set has been configured or not.

Step 22

Add session geo tag in MongoDBs. Repeat these steps for both session replication sets.

For more information, refer to Session Query Restricted to Local Site during Failover for more details.

Site1 running log: This procedure only applies if customer have local site tagging enabled.

Note

 
To modify priorities, you must update the members array in the replica configuration object. The array index begins with 0. The array index value is different than the value of the replica set member's members[n]._id field in the array. 

mongo sessionmgr01:27727
MongoDB shell version: 2.6.3
connecting to: sessionmgr01:27727/test

set10:PRIMARY> conf = rs.conf();
{
	"_id" : "set10",
	"version" : 2,
	"members" : [
		{
			"_id" : 0,
			"host" : "pcrfclient01a:27727",
			"arbiterOnly" : true
		},
		{
			"_id" : 1,
			"host" : "sessionmgr01a:27727",
			"priority" : 5
		},
		{
			"_id" : 2,
			"host" : "sessionmgr02a:27727",
			"priority" : 4
		},
		{
			"_id" : 3,
			"host" : "sessionmgr01b:27727",
			"priority" : 3
		},
		{
			"_id" : 4,
			"host" : "sessionmgr02b:27727",
			"priority" : 2
		}
	],
	"settings" : {
		"heartbeatTimeoutSecs" : 1
	}
}
set10:PRIMARY> conf.members[1].tags = { "sessionLocalGeoSiteTag": "Site1" }
{ "sessionLocalGeoSiteTag" : "Site1" }
set10:PRIMARY> conf.members[2].tags = { "sessionLocalGeoSiteTag": "Site1"}
{ "sessionLocalGeoSiteTag" : "Site1" }
set10:PRIMARY> conf.members[3].tags = { "sessionLocalGeoSiteTag": "Site2"}
{ "sessionLocalGeoSiteTag" : "Site2" }
set10:PRIMARY> conf.members[4].tags = { "sessionLocalGeoSiteTag": "Site2"}
{ "sessionLocalGeoSiteTag" : "Site2" }
set10:PRIMARY> rs.reconfig(conf);
{ "ok" : 1 }
set10:PRIMARY> rs.conf();
{
	"_id" : "set10",
	"version" : 3,
	"members" : [
		{
			"_id" : 0,
			"host" : "pcrfclient01a:27727",
			"arbiterOnly" : true
		},
		{
			"_id" : 1,
			"host" : "sessionmgr01a:27727",
			"priority" : 5,
			"tags" : {
				"sessionLocalGeoSiteTag" : "Site1"
			}
		},
		{
			"_id" : 2,
			"host" : "sessionmgr02a:27727",
			"priority" : 4,
			"tags" : {
				"sessionLocalGeoSiteTag" : "Site1"
			}
		},
		{
			"_id" : 3,
			"host" : "sessionmgr01b:27727",
			"priority" : 3,
			"tags" : {
				"sessionLocalGeoSiteTag" : "Site2"
			}
		},
		{
			"_id" : 4,
			"host" : "sessionmgr02b:27727",
			"priority" : 2,
			"tags" : {
				"sessionLocalGeoSiteTag" : "Site2"
			}
		}
	],
	"settings" : {
		"heartbeatTimeoutSecs" : 1
	}
}
set10:PRIMARY>

Site2 TAG configuration:

Note

 
To modify priorities, you must update the members array in the replica configuration object. The array index begins with 0. The array index value is different than the value of the replica set member's members[n]._id field in the array. 

mongo sessionmgr01b:37727
MongoDB shell version: 2.6.3
connecting to: sessionmgr01b:37727/test
set11:PRIMARY> conf = rs.conf();
{
	"_id" : "set11",
	"version" : 2,
	"members" : [
		{
			"_id" : 0,
			"host" : "pcrfclient01b:47727",
			"arbiterOnly" : true
		},
		{
			"_id" : 1,
			"host" : "sessionmgr01b:37727",
			"priority" : 5
		},
		{
			"_id" : 2,
			"host" : "sessionmgr02b:37727",
			"priority" : 4
		},
		{
			"_id" : 3,
			"host" : "sessionmgr01a:37727",
			"priority" : 3
		},
		{
			"_id" : 4,
			"host" : "sessionmgr02a:37727",
			"priority" : 2
		}
	],
	"settings" : {
		"heartbeatTimeoutSecs" : 1
	}
}
set11:PRIMARY> conf.members[1].tags = { "sessionLocalGeoSiteTag": "Site2"}
{ "sessionLocalGeoSiteTag" : "Site2" }
set11:PRIMARY> conf.members[2].tags = { "sessionLocalGeoSiteTag": "Site2"}
{ "sessionLocalGeoSiteTag" : "Site2" }
set11:PRIMARY> conf.members[3].tags = { "sessionLocalGeoSiteTag": "Site1"}
{ "sessionLocalGeoSiteTag" : "Site1" }
set11:PRIMARY> conf.members[4].tags = { "sessionLocalGeoSiteTag": "Site1"}
{ "sessionLocalGeoSiteTag" : "Site1" }
set11:PRIMARY> rs.reconfig(conf);
{ "ok" : 1 }
set11:PRIMARY> rs.conf();
{
	"_id" : "set11",
	"version" : 3,
	"members" : [
		{
			"_id" : 0,
			"host" : "pcrfclient01b:47727",
			"arbiterOnly" : true
		},
		{
			"_id" : 1,
			"host" : "sessionmgr01b:37727",
			"priority" : 5,
			"tags" : {
				"sessionLocalGeoSiteTag" : "Site2"
			}
		},
		{
			"_id" : 2,
			"host" : "sessionmgr02b:37727",
			"priority" : 4,
			"tags" : {
				"sessionLocalGeoSiteTag" : "Site2"
			}
		},
		{
			"_id" : 3,
			"host" : "sessionmgr01a:37727",
			"priority" : 3,
			"tags" : {
				"sessionLocalGeoSiteTag" : "Site1"
			}
		},
		{
			"_id" : 4,
			"host" : "sessionmgr02a:37727",
			"priority" : 2,
			"tags" : {
				"sessionLocalGeoSiteTag" : "Site1"
			}
		}
	],
	"settings" : {
		"heartbeatTimeoutSecs" : 1
	}
}
set11:PRIMARY>

Step 23

Run diagnostics.sh to verify that the system is in healthy state.

Rollback Additional Session Replication Set


Caution
Removing session replication set from running Production system can impact in session loss hence it is not recommended. But if there are no other options due to any circumstances, follow these instructions.

Procedure

Step 1

Run diagnostics.sh from OAM (pcrfclient) or Cluster Manager to verify the system is in healthy state.

Step 2

Restore ADMIN database from the backup. Restore sharding database only.

mongorestore --drop --objcheck --host sessionmgr01 --port 27721 --db sharding sharding

Step 3

Run restartall.sh to restart the system.

Note

 

If it is a GR site, run restartall.sh on both sites before proceeding to the next step.

Caution

 

Executing restartall.sh will cause messages to be dropped.

Step 4

Drop the newly created session replication set. In this example, remove set15.

build_set.sh --session --remove-replica-set --setname set15

Step 5

Verify sharding errors are not reported by qns nodes. Login to pcrfclient01 of Site-1 and Site-2.

tailf /var/log/broadhop/consolidated-qns.log

Ignore the following errors: 

2016-10-05 11:45:01,446 [pool-3-thread-1] WARN  c.b.c.m.dao.impl.ShardInterface.? - Unexpected error
java.lang.NullPointerException: null
at com.broadhop.cache.mongodb.dao.impl.ShardInterface$MonitorShards.monitorSessionTypeStatisticsCounter(ShardInterface.java:496) ~[com.broadhop.policy.geoha.cache_8.1.1.r090988.jar:na]
at com.broadhop.cache.mongodb.dao.impl.ShardInterface$MonitorShards.run(ShardInterface.java:407) ~[com.broadhop.policy.geoha.cache_8.1.1.r090988.jar:na]
at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511) [na:1.8.0_45]
at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308) [na:1.8.0_45]
at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180) [na:1.8.0_45]
at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294) [na:1.8.0_45]
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142) [na:1.8.0_45]
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617) [na:1.8.0_45]
at java.lang.Thread.run(Thread.java:745) [na:1.8.0_45]

Step 6

Run diagnostics.sh from OAM (pcrfclient) or Cluster Manager to verify the system is in healthy state.

Step 7

Edit mongoConfig.cfg file and remove the entries related to set15 from Site-1 (Cluster-A).

Step 8

Copy mongoConfig.cfg file to all the nodes using copytoall.sh script from Cluster Manager.

copytoall.sh /etc/broadhop/mongoConfig.cfg /etc/broadhop/mongoConfig.cfg 
Copying '/var/qps/config/mobile/etc/broadhop/mongoConfig.cfg' to '/etc/broadhop/mongoConfig.cfg' on all VMs
lb01
mongoConfig.cfg 
100%  4659  4.6KB/s  00:00    
lb02
mongoConfig.cfg
100%  4659  4.6KB/s  00:00    
sessionmgr01 
<output truncated>

Step 9

Transfer the modified mongoConfig.cfg file to Site2 ( Cluster-B ). 

scp /etc/broadhop/mongoConfig.cfg cm-b:/etc/broadhop/mongoConfig.cfg
root@cm-b's password: 
mongoConfig.cfg
100%  4659  4.6KB/s  00:00    
[root@cm-a ~]#

Step 10

SSH to Cluster-B Cluster Manager. Run build_etc.sh to make sure modified mongoConfig.cfg file is restored after reboot. 

/var/qps/install/current/scripts/build/build_etc.sh 
Building /etc/broadhop...
Copying to /var/qps/images/etc.tar.gz...
Creating MD5 Checksum...

Step 11

Copy mongoConfig.cfg file from Cluster-B Cluster Manager to all the nodes using copytoall.sh from Cluster Manager.

copytoall.sh /etc/broadhop/mongoConfig.cfg /etc/broadhop/mongoConfig.cfg 
Copying '/var/qps/config/mobile/etc/broadhop/mongoConfig.cfg' to '/etc/broadhop/mongoConfig.cfg' on all VMs
lb01
mongoConfig.cfg
100%  4659  4.6KB/s  00:00    
lb02
mongoConfig.cfg
100%  4659  4.6KB/s  00:00    
<output truncated>

Network Latency Tuning Parameters

In GR, if the network latency between two sites is more than the threshold value, -DringSocketTimeOut, -DshardPingerTimeoutMs and -balancePingerTimeoutMs parameters need to be configured with the appropriate values.

To get the values that must be configured in -DringSocketTimeOut, -DshardPingerTimeoutMs and -balancePingerTimeoutMs, check the latency using ping command for the sessionmgr which hosts the shard.

Example:

If the network latency between two sites is 150 ms, the value must be configured as 50 + 150 (network latency in ms) = 200 ms.

The parameters need to be added in /etc/broadhop/qns.conf on both sites: 


	-DringSocketTimeOut=200
 -DshardPingerTimeoutMs=200
 -DbalancePingerTimeoutMs=200

If the parameters are not configured, default values will be considered: 


	-DringSocketTimeOut=50
 -DshardPingerTimeoutMs=75
 -DbalancePingerTimeoutMs=75

In addition to the above parameters, the following parameters need to updated if network latency is found to be 150 ms. These values need to be increased so that QNS processes can come up and get connected to Mongo Database on Session Managers without having timeouts. 

-Dmongo.client.thread.maxWaitTime=1700
	-Dmongo.client.thread.maxWaitTime.balance=1700
	-Dmongo.client.thread.maxWaitTime.remoteBalance=1700
	-Dmongo.client.thread.maxWaitTime.remoteSpr=1700
	-Dmongo.client.thread.maxWaitTime.cdrrep=1700
	-Dmongo.client.thread.maxWaitTime.cdr=1700

Remote SPR Lookup based on IMSI/MSISDN Prefix

Prerequisites

Policy Builder configuration on both the sites should be the same.

Configuration

Procedure

Step 1

Configure the Lookup key field in Policy Builder under 'Domain'. It can be IMSI, MSISDN, Session User Name, and so on. An example configuration is given below:

Figure 1. Remote Db Lookup Key


Step 2

Configure remote databases in Policy Builder under USuM Configuration.

Consider there are two sites: Site1 (Primary) and Site2 (Secondary). In Policy Builder there will be two clusters for Site1 and Site2 in case of Active/Active model.

Under 'Cluster-Site2', create USuM Configuration and add remote databases to be accessed when Site1 is not available.

Here is an example configuration:

Figure 2. Remote Database Configuration
Table 1. Remote Database Configuration Parameters

Parameter

Description

Name

Unique name to identify the remote database.

Note

 

This is needed to see the correct sites's subscriber in Control Center, when multiple SPR is configured.

Match Type

Pattern match type.

It can be Starts With, Ends With, Equals, Regex match type.

Match Value

Key/regex to be used in pattern matching.

Connection per host

Number of connections that can be created per host.

Db Read Preference

Database read preferences.

Primary/Secondary/Tertiary/Quaternary Database Host, Port

Connection parameter to access database. This should be accessible from Site2 irrespective of Site1 is UP or DOWN.

Important

 

The host names must exactly be the same host name used when the corresponding replica-set is created in Mongo. Only the data holding members need to be configured (and not the arbiters).

For more information on Remote Database Configuration parameters, refer to the CPS Mobile Configuration Guide for this release.

Remote Balance Lookup based on IMSI/MSISDN Prefix

Prerequisites

Policy Builder configuration on both the sites should be the same.

Configuration

Procedure

Step 1

Configure the Lookup key field in policy builder under Domain. It can be IMSI, MSISDN, Session User name and so on. An example configuration is given:

Figure 3. Lookup Key


Step 2

Configure remote databases in policy builder under Balance Configuration.

Consider there are two sites: Site1 (Primary) and Site2 (Secondary). So in Policy Builder there will be two clusters for Site1 and Site2.

Under 'Cluster-Site2', create Balance Configuration and add remote databases to be accessed when Site1 is not available.

An example configuration is given:

Figure 4. Example Configuration


Table 2. Remote Database Configuration Parameters

Parameter

Description

Name

Unique name to identify the remote database.

Match Type

Match type to be matched for the remote database to be selected for lookup.

Connection per host

Number of connections that can be created per host.

Db Read Preference

Database read preferences.

Primary Ip Address, Secondary Ip Address, Port

Connection parameter to access database. This should be accessible from Site2 irrespective of Site1 is UP or DOWN.

For more information on Balance Configuration parameters, refer to the CPS Mobile Configuration Guide for this release.

SPR Provisioning

CPS supports multiple SPR and multiple balance databases in different ways based on deployments. SPR provisioning can be done either based on end point/listen port or using API router configuration.

SPR Location Identification based on End Point/Listen Port

Prerequisites

Policy Builder configuration on both the sites should be the same.

Configuration

Consider there are two sites: Site1 (Primary) and Site2 (Secondary).

Procedure

Add new entry on Site2 in haproxy.cfg (/etc/haproxy/haproxy.cfg) file listening on port 8081 (or any other free port can be used) with custom header “RemoteSprDbName”. Same configuration to be done on both load balancers. 

listen pcrf_a_proxy lbvip01:8081
        mode http
        reqadd RemoteSprDbName:\ SPR_SITE1
        balance roundrobin
        option httpclose
        option abortonclose
        option httpchk GET /ua/soap/KeepAlive
        server qns01_A qns01:8080 check inter 30s
        server qns02_A qns02:8080 check inter 30s
        server qns03_A qns03:8080 check inter 30s
        server qns04_A qns04:8080 check inter 30s
# If there are more qns add all entries here

Where,

RemoteSprDbName is the custom header.

SPR_SITE1 is the remote database name configured in Step 2.

API Router Configuration

The following are the three use cases where a user must configure API router:

  • Multiple SPR/Multiple Balance

  • Common SPR/Multiple Balance (Hash Based)

  • Common SPR Database and Multiple Balance Database based on SPR AVP

Use Cases

Multiple SPR/Multiple Balance
Use Case

This will be useful when there are multiple active sites and each has their own separate SPR and balance database.

Logic

  • When API router receives the request it will extract the Network Id (MSISDN) from the request.

  • It will iterate through the API router criteria table configured in Api Router Configuration in Policy Builder and find SPR, Balance database name by network Id.

  • API router will make unified API call adding SPR, balance database name in http header.

  • SPR and balance module will use this SPR, balance database name and make queries to appropriate databases.

Common SPR/Multiple Balance (Hash Based)
Use Case

This will be useful when there is common SPR across multiple sites, and latency between site is less. Also this will evenly distribute the data across all balance databases.

Logic

  • When API router receives the create subscriber request:

    • It first generates hash value using network Id (in range 0 to n-1, where n is from qns.conf parameter -DmaxHash=2)

    • It will add generated hash value in SPR AVP (with code _balanceKeyHash).

    • For example, if maxHash is 2 and generated hash value is 1, then SPR AVP will be _balanceKeyHash=1

  • When API router receives the request other than create subscriber.

    • It will query subscriber and get hash value from subscriber AVP (_balanceKeyHash).

  • Once hash value is available, it will iterate through the API router criteria table configured in Api Router Configuration in Policy Builder and find balance database name by hash.

  • API router will make unified API call adding balance database name in http header.

  • Balance module will use this balance database name and make query to appropriate balance database.

Common SPR Database and Multiple Balance Database based on SPR AVP
Use Case

This will be useful when there is common SPR across multiple sites, but each has separate balance in each site. We can add region AVP in each subscriber to read from local database.

Logic

  • When API router receives the request:

    • It will query subscriber and get subscriber AVP from subscriber. AVP name is configurable.

    • Once AVP value is available, it will iterate through the API router criteria table configured in Api Router Configuration in Policy Builder and find balance database name by AVP.

    • API router will make unified API call adding balance database name in http header.

  • Balance module will use this balance database name and make query to appropriate balance database.

HTTP Endpoint

By default, API router is exposed on /apirouter and the Unified API endpoint is exposed on /ua/soap. The default URLs are as follows:

Table 3. Default URLs

Setup

Unified API

API Router

HA

https://lbvip01:8443/ua/soap

https://lbvip01:8443/apirouter

Figure 5. API Router Configuration


Some customer may have configured the URL at multiple places and do not want to change URL. To change the endpoint so that the API router uses /ua/soap, add the following parameters in /etc/broadhop/qns.conf. This makes API router act as unified API. 

-DapirouterContextPath=/ua/soap
-Dapi.ua.context.path=/ua/backend
-Dua.context.path=/ua/backend

Note

The api.ua.context.path and ua.context.path parameters must be the same.

Figure 6. Unified API Router Configuration


New URLs are as follows:

Table 4. New URLs

Setup

Unified API

API Router

HA

https://lbvip01:8443/ua/backend

https://lbvip01:8443/ua/soap

Configuration

By default, API router configuration feature is not installed. To install this feature, put the following entries in feature files.

Table 5. Feature File Changes

Feature File

Feature Entry

/etc/broadhop/pcrf/feature

com.broadhop.apirouter.service.feature

/etc/broadhop/pb/feature

com.broadhop.client.feature.apirouter


Note

Change in feature file requires to run buildall.sh, reinit.sh from Cluster Manager for the features to get installed.

Policy Builder Configuration
Domain

  1. Remote Db lookup key field:

    This retriever fetches the value that can be used in patten match to get remote SPR, balance database name.

    This field retriever can be:

    • MSISDN retriever, IMSI retriever or whatever is networkid in subscriber to support multiple SPR, multiple balance.

    • NetworkIdHash retriever to support common SPR, multiple balance based on hashing.

    • Subscriber retriever AVP to support common SPR, multiple balance based on subscriberAVP.

API Router Configuration

  1. Enable Multi-Credential Management: This check box is used to add the support for the multi-credential management for the API router to handle the following Unified API requests: CreateSubscriber, DeleteSubscriber, GetSubscriber, CreateBalance, DeleteBalance, ChangeCredentialUsername.


    Note

    AddCredential, AddCredentials, DeleteCredential, and DeleteCredentials API requests will be added in the future to complete the multi-credential handling.

  2. Enable Backup Cache Lookup For Primary Key: This check box is used to turn on/off the step to iterate over the know databases to lookup the subscriber primary key if the secondary key cache does not find the record.


    Note

    A CreateSubscriberRequest does not look into the cache or iterate over the databases because it inserts a new record in the database and must build the cache entries from the request.

  3. Filter Type: Type of filter to be used.

    • NetworkId — To configure multiple SPR, multiple balance.

    • NetworkIdHash — To configure common SPR, multiple balance based on hashing.

    • SubscriberAVP — To configure common SPR, multiple balance based on susbcriberAVP.

      AVP Name can be changed by adding flag -DbalanceKeyAvpName=avpName. Refer to Configurable Flags.

  4. Router Criteria: The following is the list of criteria to consider for pattern matching.

    Table 6. Router Criteria

    Criteria

    Description

    Match Type

    Pattern match type.

    It can be Starts With, Ends With, Equals, Regex match type.

    Match Value

    Key/regex to be used in pattern matching.

    Remote SPR DB name

    If criteria match, use this database name as SPR database name.

    This database name should match with the remote database name configured in USuMConfiguration > Remote databases > Name.

    Remote Balance DB name

    If criteria match, use this database name as Balance database name.

    This database name should match with the remote database name configured in BalanceConfiguration > Remote databases > Name.

Balance Configuration (remote databases)

The following parameters can be configured under Remote Database for Balance Configuration.

Table 7. Remote Database Parameters

Parameter

Description

Name

Balance database name.

Match Type

Pattern match type.

It can be Starts With, Ends With, Equals, Regex match type.

Match Value

Key/regex to be used in pattern matching.

Connection per host

Balance database mongo connection per host.

Db Read preference

Balance database read preference.

Primary Ip Address

Balance database primary member IP address.

Secondary Ip Address

Balance database secondary member IP address.

Port

Balance database primary member IP address.

Following fields needs to be configured if hot standby for balance database is needed

Backup DB Host

Backup balance database primary member IP address.

Backup DB Secondary Host

Backup balance database secondary member IP address.

Backup DB Port

Backup balance database primary member IP address.

USuM Configuration (remote databases)

The following parameters can be configured under Remote Database for USuM Configuration.


Important

The host names must exactly be the same host name used when the corresponding replica-set is created in Mongo. Only the data holding members need to be configured (and not the arbiters).

Table 8. Remote Database Parameters

Parameter

Description

Name

SPR database name.

Match Type

Pattern match type.

It can be Starts With, Ends With, Equals, Regex match type.

Match Value

Key/regex to be used in pattern matching.

Connection Per host

SPR database mongo connection per host.

Db Read preference

SPR database read preference.

Primary Database Host

SPR database primary member host name.

Secondary Database Host

SPR database secondary member host name.

Tertiary Database Host

SPR database tertiary member host name.

Quaternary Database Host

SPR database quaternary member host name.

Port

SPR database primary member port number.

Configurable Flags

The following flags are configurable in /etc/broadhop/qns.conf file:

Table 9. Configurable Flags

Flag

Description

Default Value

balanceKeyAvpName

Subscriber AVP name to be used for subscriber based multiple balance databases.

_balanceKey

balanceKeyHashAvpName

Internal AVP name being used for hash based multiple balance databases.

_balanceKeyHash

maxHash

Maximum value of hash to generate.

No default value

ua.context.path

Unified API context path.

/ua/soap

apirouterContextPath

API router context path.

/apirouter

Configuration Examples
Multiple SPR/Multiple Balance

Domain Configuration

The Remote Db Lookup Key Field can be MSISDN, IMSI, and so on which is used as network ID in SPR.

Figure 7. Domain Configuration


USuM Configuration
Figure 8. USuM Configuration
Balance Configuration
Figure 9. Balance Configuration


API Router Configuration
Figure 10. API Router Configuration


Common SPR/Multiple Balance (Hash Based)
Domain Configuration
Figure 11. Domain Configuration


USuM Configuration
Figure 12. USuM Configuration
Balance Configuration
Figure 13. Balance Configuration


API Router Configuration
Figure 14. API Router Configuration


Common SPR Database and Multiple Balance Database based on SPR AVP
Domain Configuration
Figure 15. Domain Configuration


USuM Configuration
Figure 16. USuM Configuration
Balance Configuration
Figure 17. Balance Configuration


API Router Configuration
Figure 18. API Router Configuration


Rebalance

For hash-based balance and common SPR, rebalance is supported. It means old balance data can be rebalanced to new balance databases without need of re-provisioning.

Rebalance can be done by executing the following OSGi commands:

  • rebalanceByHash — Rebalance with same balance shards (shards here means internal balance shards).

  • rebalanceByHash [oldShardCount] [newShardCount] — Rebalance to change (increase/decrease) balance shards.

Rebalance with same balance shard

This is applicable only for hash based balance databases. To add new database to existing database, perform the following steps:

Procedure

Step 1

Log in to Control Center and note down few subscriber which has balance.

Step 2

Change Policy Builder configuration: API Router, Balance, Domain, so on and publish the modified configuration.

Step 3

Add parameter maxHash in qns.conf file.

  1. Value depends on number of databases.

    For example, if there are two balance databases configured in Policy Builder, set value to 2.

    -DmaxHash=2

Step 4

Add context path parameter ua.context.path and apirouterContextPath in qns.conf file. This is needed for Control Center to call via API router.

-DapirouterContextPath=/ua/soap

-Dua.context.path=/ua/backEnd

Step 5

Execute copytoall.sh and restart Policy Server (QNS) processes.

Step 6

Login to OSGi console on qns01.

telnet qns01 9091

Step 7

Execute rebalanceByHash command.

rebalanceByHash

Step 8

Log in to Control Center and verify subscriber still has balance noted in Step 1.

Rebalance to Change Number of Balance Shards

This is applicable only for hash-based balance databases.

To increase the number of balance shards, perform the following steps:

  1. Login to Control Center and note down com.cisco.balance.dbs few subscribers who have balance.

  2. In the qns.conf file, add or edit com.cisco.balance.dbs.

    1. Value will be new shard number.

      Example — If you are increasing balance shards from 4 to 8, value should be set to 8.

      -Dcom.cisco.balance.dbs=8

  3. Run copytoall.sh and restart qns processes.

  4. Login to OSGi console on qns01.

    telnet qns01 9091

  5. Run rebalanceByHash command.

    rebalanceByHash <old shard number> <new shard number>

    Example — If you are increasing balance shards from 4 to 8, old shard number is 4 and new shard number is 8.

    rebalanceByHash 4 8

  6. Login to Control Center and verify subscriber still has balance noted in Step 1.

To decrease the number of balance shards, perform the following steps:

  1. Login to Control Center and note down few subscribers who have balance.

  2. Login to OSGi console on qns01.

    telnet qns01 9091

  3. Run rebalanceByHash command.

    rebalanceByHash <old shard number> <new shard number>

    Example — If you are decreasing balance shards from 6 to 4, old shard number is 6 and new shard number is 4.

    rebalanceByHash 6 4

  4. In the qns.conf file, add or edit com.cisco.balance.dbs.

    1. Value will be new shard number

      Example — If you are decreasing balance shards from 6 to 4, value should be set to 4.

      -Dcom.cisco.balance.dbs=4

  5. Run copytoall.sh and restart qns processes.

  6. Login to Control Center and verify subscriber still has balance noted in Step1.

Configurations to Handle Traffic Switchover

When Policy Server (QNS) is Down

The following configurations are needed to enable qns_hb.sh script. This script stops Policy Server (QNS) processes from lb01/lb02 when all Policy Server (QNS) are down (that is, qns01,02..n).


Note

To understand traffic switchover, refer to Load Balancer VIP Outage.

Procedure

Step 1

To enable script, add the following configuration in /var/qps/config/deploy/csv/Configuration.csv file:

mon_qns_lb,true,

For more information on how to add the parameter in Configuration.csv file, refer to CPS Installation Guide for VMware for 9.1.0 and later releases.

Note

 

Any database that is not replicated across sites should not be configured for monitoring by mon_db_for_callmodel.sh script.

Step 2

To disable script, add the following configuration in /var/qps/config/deploy/csv/Configuration.csv file or remove mon_qns_lb tag from this CSV file:

mon_qns_lb,,

Step 3

Import CSV to JSON by executing the following command:

/var/qps/install/current/scripts/import/import_deploy.sh

Step 4

Execute the following command to validate the imported data:

cd /var/qps/install/current/scripts/deployer/support/

python jvalidate.py

Note

 

The above script validates the parameters in the Excel/csv file against the ESX servers to make sure ESX server can support the configuration and deploy VMs.

Note

 

If you are deploying the VMs using the --nossh feature, jvalidate will not work as jvalidate needs SSH login credentials to get the memory, CPU and network details from the ESXi. Since --nonssh feature doesn’t have the SSH login credentials, the script execution fails for getting the details. You can login to the vCenter where the ESXi’s are mapped and check the required details to plan the VM design.

Step 5

Reinitiate lb01 and lb02 by executing the following command:

/etc/init.d/vm-init

For more information on configuring GR features, refer to CPS Mobile Configuration Guide.

When Replicated (inter-site) Database is not Primary on a Site


Note

To understand traffic switchover, refer to Load Balancer VIP Outage.

Procedure

Step 1

Add the list of databases that needs to be monitored in mon_db_for_callmodel.conf file (/etc/broadhop/mon_db_for_callmodel.conf) in Cluster Manager.

Important

 

Contact your Cisco Technical Representative for more details.

Add the following content in the configuration file (mon_db_for_callmodel.conf):

Note

 

The following is an example and needs to be changed based on your requirement: 

#this file contains set names that are available in mongoConfig.cfg. Add set names one below other.
#Refer to README in the scripts folder.
SESSION-SET1
SESSION-SET2
BALANCE-SET1
SPR-SET1

Step 2

To enable switch-off of UAPI traffic when replicated (inter-site) configured databases are not primary on this site, add STOP_UAPI=1 in /etc/broadhop/mon_db_for_callmodel.conf file.

Note

 

To disable switch-off of UAPI traffic when replicated (inter-site) configured databases are not primary on this site, add STOP_UAPI=0 (if this parameter is not defined, ) in /etc/broadhop/mon_db_for_callmodel.conf file.

When we recover GR site, we have to manually start UAPI interface (if it is disabled) by executing the following command as a root user on lb01 and lb02:

echo "enable frontend https-api" | socat stdio /tmp/haproxy

Step 3

To configure the percentage value for session replica sets to be monitored from the configured session replica set list, set the PERCENTAGE_SESS_DB_FAILURE parameter in the /etc/broadhop/mon_db_for_callmodel.conf configuration file.

Use an integer from 1 to 100.

Note

 

PERCENTAGE_SESS_DB_FAILURE parameter should be configured such that it results in integer value and not float for the number of sets user expects to be down to trigger the fail over. For example, if there are 4 session replica-sets and user wants failover to be triggered only when two or more replica-sets are down (>= 2 condition), user should configure this value as 50 (%). User at times wrongly interprets this parameter as "more than given number of replica-sets". For example in above case, user intends to have a failover if more than one replica-set (so, this is equivalent to "> 1" condition) are down and configures a value which is more than 25 (for example, 45) which does not work. The expected value should meet the condition of ">= desired number of replica-sets to be down for failover trigger".

Step 4

Rebuild etc directory on cluster by executing the following command:

/var/qps/install/current/scripts/build/build_etc.sh

When Virtual IP (VIP) is Down

You can configure CPS to monitor VIPs. If any of the configured VIPs goes down, the configured databases from the local site are made secondary. However, if the databases at a site go down, the VIP is not shut down.

For VMware

Specify the configuration in /etc/broadhop/mon_db_for_lb_failover.conf.


Note

The /etc/broadhop/mon_db_for_lb_failover.conf file contains the configuration for VIPs along with the database set names.

For OpenStack

  • During fresh install, apply the configuration using http://<cluman-ip>:8458/api/system/config/action/.

  • Once system is deployed, use PATCH API http://<cluman-ip>:8458/api/system/config/application-config to apply the configuration that enables monitoring of VIPs.

The new configuration vipMonitorForLb with the following format is created under applicationConfig: 

vipMonitorForLb:
 vipName:
 - lbvip01
 - lbvip02

Where, vipName is the array of VIPs to be monitored.

In case of any issues, check the API log file /var/log/orchestration-api-server.log and the /var/log/broadhop/scripts directory (after system configuration) for any errors.

Configuring Session Database Percentage Failure

You can configure the percentage value of session replica sets to be monitored from the configured session replica set list.

For VMware

Specify the configuration in /etc/broadhop/mon_db_for_callmodel.conf by modifying the PERCENTAGE_SESS_DB_FAILURE parameter. Use an integer from 1 to 100.

For example, PERCENTAGE_SESS_DB_FAILURE=50


Note

PERCENTAGE_SESS_DB_FAILURE parameter should be configured such that it results in integer value and not float for the number of sets user expects to be down to trigger the fail over. For example, if there are 4 session replica-sets and user wants failover to be triggered only when two or more replica-sets are down (>= 2 condition), user should configure this value as 50 (%). User at times wrongly interprets this parameter as "more than given number of replica-sets". For example in above case, user intends to have a failover if more than one replica-set (so, this is equivalent to "> 1" condition) are down and configures a value which is more than 25 (for example, 45) which does not work. The expected value should meet the condition of ">= desired number of replica-sets to be down for failover trigger".

For OpenStack

  • During fresh install, apply the configuration using http://<cluman-ip>:8458/api/system/config/action.

  • Once system is deployed, use PATCH API http://<cluman-ip>:8458/api/system/config/application-config to apply the configuration.

The new configuration dbMonitorForQns with the following format is created under applicationConfig: 

dbMonitorForQns:
stopUapi: "false"
percentageSessDBFailure: 50
setName:
- SESSION-SET1
- SESSION-SET2
- SESSION-SET3
- SESSION-SET4

In case of any issues, check the API log file /var/log/broadhop/scripts/mon_db_for_callmodel_$DATE.log.

Remote Databases Tuning Parameters

If remote databases are configured for balance or SPR, respective mongo connection parameters are required to be added in /etc/broadhop/qns.conf file.

Remote SPR 

-DdbSocketTimeout.remoteSpr=1200
-DdbConnectTimeout.remoteSpr=600
-Dmongo.connections.per.host.remoteSpr=10
-Dmongo.threads.allowed.to.wait.for.connection.remoteSpr=10
-Dmongo.client.thread.maxWaitTime.remoteSpr=1200

Remote Balance 

-DdbSocketTimeout.remoteBalance=1200
-DdbConnectTimeout.remoteBalance=600
-Dmongo.connections.per.host.remoteBalance=10
-Dmongo.threads.allowed.to.wait.for.connection.remoteBalance=10
-Dmongo.client.thread.maxWaitTime.remoteBalance=1200

SPR Query from Standby Restricted to Local Site only (Geo Aware Query)

Procedure

Step 1

Add new entry for Site1 and Site 2 in /etc/broadhop/pcrf/qns.conf file on pcrfclient01.

-DsprLocalGeoSiteTag=Site1 ====> in Site1 qns.conf file

-DsprLocalGeoSiteTag=Site2 ====> in Site2 qns.conf file

Step 2

Execute syncconfig.sh. To reflect the above change, CPS needs to be restarted.

Step 3

Add the site tag information in mongoConfig.cfg file to make sure that the site tags aren't removed post ISSM.

Here is a sample site tag information added in mongoConfig.cfg file. 

[SPR-SET1]
SETNAME=set04
OPLOG_SIZE=3072
ARBITER=S1-CA-PRI-arbitervip:27720
ARBITER_DATA_PATH=/var/data/sessions.4
PRIMARY_TAG=Site1
PRIMARY-MEMBERS
MEMBER1=S1-CA-PRI-sessionmgr01:27720
MEMBER2=S1-CA-PRI-sessionmgr02:27720
SECONDARY_TAG=Site2
SECONDARY-MEMBERS
MEMBER1=S2-CA-SEC-sessionmgr01:27720
MEMBER2=S2-CA-SEC-sessionmgr02:27720
DATA_PATH=/var/data/sessions.4
[SPR-SET1-END]

Step 4

Add tag to SPR MongoDBs.

  1. Run diagnostics.sh command on pcrfclient01 and find SPR database primary member and port number.

    diagnostics.sh --get_replica_status

    Note

     

    If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

    Also, you can login to mongo on that member and check its actual status.

  2. Login to SPR database using the primary replica set hostname and port number.

    For example, mongo --host sessionmgr01 --port 27720

  3. Get the replica members by executing the following command:

    Execute rs.conf() from any one member.

    SAMPLE OUTPUT 

    set04:PRIMARY> rs.conf(); 
{ 
        "_id" : "set04", 
        "version" : 319396, 
        "members" : [ 
                { 
                        "_id" : 0, 
                        "host" : "pcrfclient-arbiter-site3:37720", 
                        "arbiterOnly" : true 
                }, 
                { 
                        "_id" : 1, 
                        "host" : "sessionmgr01-site1:27720", 
                        "priority" : 2                         
                }, 
                { 
                        "_id" : 2, 
                        "host" : "sessionmgr02-site1:27720", 
                        "priority" : 2 
                }, 
                { 
                        "_id" : 3, 
                        "host" : "sessionmgr05-site1:27720", 
                        "priority" : 2 
                }, 
                { 
                        "_id" : 4, 
                        "host" : "sessionmgr06-site1:27720", 
                        "votes" : 0, 
                        "priority" : 2 
                }, 
                { 
                        "_id" : 5, 
                        "host" : "sessionmgr01-site2:27720" 
                }, 
                { 
                        "_id" : 6, 
                        "host" : "sessionmgr02-site2:27720"                  
                }, 
                { 
                        "_id" : 7, 
                        "host" : "sessionmgr05-site2:27720"                         
                }, 
                { 
                        "_id" : 8, 
                        "host" : "sessionmgr06-site2:27720", 
                        "votes" : 0 
                } 
        ], 
        "settings" : { 
                "heartbeatTimeoutSecs" : 1 
        } 
}

  4. Now from the list, find out the members of Site1 and Site2 to be tagged (excluding the arbiter). After finding member, execute the following command from Primary member to tag members.

    Note

     

    To modify priorities, you must update the members array in the replica configuration object. The array index begins with 0. The array index value is different than the value of the replica set member's members[n]._id field in the array. 

    
conf = rs.conf(); 
conf.members[1].tags = { "sprLocalGeoSiteTag": "Site1" } 
conf.members[2].tags = { "sprLocalGeoSiteTag": "Site1"} 
conf.members[3].tags = { "sprLocalGeoSiteTag": "Site1"} 
conf.members[4].tags = { "sprLocalGeoSiteTag": "Site1"} 
conf.members[5].tags = { "sprLocalGeoSiteTag": "Site2"} 
conf.members[6].tags = { "sprLocalGeoSiteTag": "Site2"} 
conf.members[7].tags = { "sprLocalGeoSiteTag": "Site2"} 
conf.members[8].tags = { "sprLocalGeoSiteTag": "Site2"} 
rs.reconfig(conf);

    Note

     
    This is a sample output. Configuration, members and tag can be different as per your environment.

    conf.members[1] means member with _id = 1 in output of rs.conf().

    After executing this command, primary member can be changed.

    Verify tags are properly set by log in on any member and executing the following command:

    rs.conf();

    SAMPLE OUTPUT 

    set04:PRIMARY> rs.conf(); 
{ 
        "_id" : "set04", 
        "version" : 319396, 
        "members" : [ 
                { 
                        "_id" : 0, 
                        "host" : "pcrfclient-arbiter-site3:37720", 
                        "arbiterOnly" : true 
                }, 
                { 
                        "_id" : 1, 
                        "host" : "sessionmgr01-site1:27720", 
                        "priority" : 2, 
                        "tags" : { 
                                "sprLocalGeoSiteTag" : "Site1" 
                        } 
                }, 
                { 
                        "_id" : 2, 
                        "host" : "sessionmgr02-site1:27720", 
                        "priority" : 2, 
                        "tags" : { 
                                "sprLocalGeoSiteTag" : "Site1" 
                        } 
                }, 
                { 
                        "_id" : 3, 
                        "host" : "sessionmgr05-site1:27720", 
                        "priority" : 2, 
                        "tags" : { 
                                "sprLocalGeoSiteTag" : "Site1" 
                        } 
                }, 
                { 
                        "_id" : 4, 
                        "host" : "sessionmgr06-site1:27720", 
                        "votes" : 0, 
                        "priority" : 2, 
                        "tags" : { 
                                "sprLocalGeoSiteTag" : "Site1" 
                        } 
                }, 
                { 
                        "_id" : 5, 
                        "host" : "sessionmgr01-site2:27720", 
                        "tags" : { 
                                "sprLocalGeoSiteTag" : "Site2" 
                        } 
                }, 
                { 
                        "_id" : 6, 
                        "host" : "sessionmgr02-site2:27720", 
                        "tags" : { 
                                "sprLocalGeoSiteTag" : "Site2" 
                        } 
                }, 
                { 
                        "_id" : 7, 
                        "host" : "sessionmgr05-site2:27720", 
                        "tags" : { 
                                "sprLocalGeoSiteTag" : "Site2" 
                        } 
                }, 
                { 
                        "_id" : 8, 
                        "host" : "sessionmgr06-site2:27720", 
                        "votes" : 0, 
                        "tags" : { 
                                "sprLocalGeoSiteTag" : "Site2" 
                        } 
                } 
        ], 
        "settings" : { 
                "heartbeatTimeoutSecs" : 1 
        } 
}

Step 5

Repeat Step 4 for all other sites. Tag names should be unique for each site.

This change overrides the read preference configured in USuM Configuration in Policy Builder.

Step 6

Execute rs.reconfig() command to make the changes persistent across replica-sets.

Balance Location Identification based on End Point/Listen Port

Prerequisites

Policy Builder configuration on both the sites should be the same.

Configuration

Consider there are two sites: Site1 (Primary) and Site2 (Secondary).

Procedure

Add new entry on Site2 in haproxy.cfg (/etc/haproxy/haproxy.cfg) file listening on port 8081 (or any other free port can be used) with custom header RemoteBalanceDbName. Same configuration needs to be done on both load balancers. 

listen pcrf_a_proxy lbvip01:8081
mode http
reqadd RemoteBalanceDbName:\ BAL_SITE1
balance roundrobin
option httpclose
option abortonclose
option httpchk GET /ua/soap/KeepAlive
server qns01_A qns01:8080 check inter 30s
server qns02_A qns02:8080 check inter 30s
server qns03_A qns03:8080 check inter 30s
server qns04_A qns04:8080 check inter 30s
# If there are more qns add all entries here

where,

RemoteBalanceDbName is the custom header.

BAL_SITE1 is the remote database name configured in Remote Balance Lookup based on IMSI/MSISDN Prefix.

Figure 19. Balance Site


Balance Query Restricted to Local Site

The following steps need to be performed for balance query from restricted to local site only (Geo aware query) during the database failover:

Consider there are two sites: Site1 and Site2


Note

If there are more than one balance databases, follow the below mentioned steps for all the databases.

The following steps are not needed to be performed for backup or hot standby databases.

Procedure

Step 1

Add new entry in Site1 qns.conf file (/etc/broadhop/qns.conf) on Cluster Manager.

-DbalanceLocalGeoSiteTag=Site1

  1. Run copytoall.sh to restart the qns processes.

Step 2

Add new entry on Site2 qns.conf file (/etc/broadhop/qns.conf) on Cluster Manager.

-DbalanceLocalGeoSiteTag=Site2

  1. Run copytoall.sh to restart the qns processes.

Step 3

Add the site tag information in mongoConfig.cfg file to make sure that the site tags aren't removed post ISSM.

Here is a sample site tag information added in mongoConfig.cfg file. 

[SPR-SET1]
SETNAME=set04
OPLOG_SIZE=3072
ARBITER=S1-CA-PRI-arbitervip:27720
ARBITER_DATA_PATH=/var/data/sessions.4
PRIMARY_TAG=Site1
PRIMARY-MEMBERS
MEMBER1=S1-CA-PRI-sessionmgr01:27720
MEMBER2=S1-CA-PRI-sessionmgr02:27720
SECONDARY_TAG=Site2
SECONDARY-MEMBERS
MEMBER1=S2-CA-SEC-sessionmgr01:27720
MEMBER2=S2-CA-SEC-sessionmgr02:27720
DATA_PATH=/var/data/sessions.4
[SPR-SET1-END]

Step 4

Add balance geo tag in MongoDBs.

  1. Run diagnostics.sh command on pcrfclient01 and find balance database primary member and port number.

    $ diagnostics.sh --get_replica_status

    Note

     

    If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

    Also, you can login to mongo on that member and check its actual status.

  2. Login to balance database using the primary replica set hostname and port number.

    For example, $ mongo --host sessionmgr01 --port 27720

  3. Get the balance database replica members information.

    Execute rs.conf() from any one member.

    SAMPLE OUTPUT 

    set04:PRIMARY> rs.conf(); 
{ 
        "_id" : "set04", 
        "version" : 319396, 
        "members" : [ 
                { 
                        "_id" : 0, 
                        "host" : "pcrfclient-arbiter-site3:27718", 
                        "arbiterOnly" : true 
                }, 
                { 
                        "_id" : 1, 
                        "host" : "sessionmgr01-site1:27718", 
                        "priority" : 4                         
   }, 
                { 
                        "_id" : 2, 
                        "host" : "sessionmgr02-site1:27718", 
                        "priority" : 3 
                },                 
                { 
                        "_id" : 3, 
                        "host" : "sessionmgr01-site2:27718", 
                        "priority" : 2 
                }, 
                { 
                        "_id" : 4, 
                        "host" : "sessionmgr02-site2:27718" 
                        "priority" : 1 
                }  
        ], 
        "settings" : { 
                "heartbeatTimeoutSecs" : 1 
        } 
}

  4. Now from the list, find out the members of Site1 and Site2 to be tagged (excluding the arbiter).

  5. After finding member, execute the following command from primary member to tag members. 

    conf = rs.conf(); 
conf.members[1].tags = { "balanceLocalGeoSiteTag": "Site1" } 
conf.members[2].tags = { "balanceLocalGeoSiteTag": "Site1"} 
conf.members[3].tags = { "balanceLocalGeoSiteTag": "Site2"} 
conf.members[4].tags = { "balanceLocalGeoSiteTag": "Site2"} 
rs.reconfig(conf);

    Note

     

    This is a sample configuration. Members and tag can be different according to your deployment.

    conf.members[1] means member with _id = 1 in output of rs.conf().

    After tagging the members, primary member may get changed if all the members have same priority.

    To verify that the tags are properly set, log in to any member and execute rs.conf() command.

    SAMPLE OUTPUT 

    set04:PRIMARY> rs.conf(); 
{ 
        "_id" : "set04", 
        "version" : 319396, 
        "members" : [ 
                { 
                        "_id" : 0, 
                        "host" : "pcrfclient-arbiter-site3:27718", 
                        "arbiterOnly" : true 
                }, 
                { 
                        "_id" : 1, 
                        "host" : "sessionmgr01-site1:27718", 
                        "priority" : 4, 
                        "tags" : { 
                                "balanceLocalGeoSiteTag" : "Site1" 
                        } 
                }, 
                { 
                        "_id" : 2, 
                        "host" : "sessionmgr02-site1:27718", 
                        "priority" : 3, 
                        "tags" : { 
                                "balanceLocalGeoSiteTag" : "Site1" 
                        } 
                }, 
                { 
                        "_id" : 3, 
                        "host" : "sessionmgr01-site2:27718", 
                        "priority" : 2, 
                        "tags" : { 
                                "balanceLocalGeoSiteTag" : "Site2" 
                        } 
                }, 
                { 
                        "_id" : 4, 
                        "host" : "sessionmgr01-site2:27718", 
                        "priority" : 1, 
                        "tags" : { 
                                "balanceLocalGeoSiteTag" : "Site2" 
                        } 
                } 
        ], 
        "settings" : { 
                "heartbeatTimeoutSecs" : 1 
        } 
}

Session Query Restricted to Local Site during Failover

The following steps need to be performed for session query from restricted to local site only (Geo aware query) during the database failover:

Consider there are two sites: Site1 and Site2


Note

If there are more than one session databases, follow the below mentioned steps for all the databases.

The following steps are not needed to be performed for backup or hot standby databases.

This geo tagging is applicable only during database failover time period. In normal case, session database query/update always happen on primary member.

Procedure

Step 1

Add new entry in Site1 qns.conf file (/etc/broadhop/qns.conf) on Cluster Manager.

-DsessionLocalGeoSiteTag=Site1

  1. Run copytoall.sh to restart the qns processes.

Step 2

Add new entry on Site2 qns.conf file (/etc/broadhop/qns.conf) on Cluster Manager.

-DsessionLocalGeoSiteTag=Site2

  1. Run copytoall.sh to restart the qns processes.

Step 3

Add the site tag information in mongoConfig.cfg file to make sure that the site tags aren't removed post ISSM.

Here is a sample site tag information added in mongoConfig.cfg file. 

[SPR-SET1]
SETNAME=set04
OPLOG_SIZE=3072
ARBITER=S1-CA-PRI-arbitervip:27720
ARBITER_DATA_PATH=/var/data/sessions.4
PRIMARY_TAG=Site1
PRIMARY-MEMBERS
MEMBER1=S1-CA-PRI-sessionmgr01:27720
MEMBER2=S1-CA-PRI-sessionmgr02:27720
SECONDARY_TAG=Site2
SECONDARY-MEMBERS
MEMBER1=S2-CA-SEC-sessionmgr01:27720
MEMBER2=S2-CA-SEC-sessionmgr02:27720
DATA_PATH=/var/data/sessions.4
[SPR-SET1-END]

Step 4

Add session geo tag in MongoDBs.

  1. First, get the session database replica members information.

    Execute rs.conf() from any one member.

    SAMPLE OUTPUT 

    set04:PRIMARY> rs.conf(); 
{ 
        "_id" : "set04", 
        "version" : 319396, 
        "members" : [ 
                { 
                        "_id" : 0, 
                        "host" : "pcrfclient-arbiter-site3:27717", 
                        "arbiterOnly" : true 
                }, 
                { 
                        "_id" : 1, 
                        "host" : "sessionmgr01-site1:27717", 
                        "priority" : 4                         
   }, 
                { 
                        "_id" : 2, 
                        "host" : "sessionmgr02-site1:27717", 
                        "priority" : 3 
                },                 
                { 
                        "_id" : 3, 
                        "host" : "sessionmgr01-site2:27717", 
                        "priority" : 2 
                }, 
                { 
                        "_id" : 4, 
                        "host" : "sessionmgr02-site2:27717" 
                        "priority" : 1 
                }  
        ], 
        "settings" : { 
                "heartbeatTimeoutSecs" : 1 
        } 
}

  2. Now from the list, find out the members of Site1 and Site2 to be tagged (excluding the arbiter).

  3. After finding member, execute the following command from Primary member to tag members.

    Note

     

    To modify priorities, you must update the members array in the replica configuration object. The array index begins with 0. The array index value is different than the value of the replica set member's members[n]._id field in the array. 

    
conf = rs.conf(); 
conf.members[1].tags = { "sessionLocalGeoSiteTag": "Site1" } 
conf.members[2].tags = { "sessionLocalGeoSiteTag": "Site1"} 
conf.members[3].tags = { "sessionLocalGeoSiteTag": "Site2"} 
conf.members[4].tags = { "sessionLocalGeoSiteTag": "Site2"} 
rs.reconfig(conf);

    THIS IS SAMPLE CONFIG, MEMBERS and TAG can be different according to your deployment.

    conf.members[1] means member with "_id" = "1" in output of rs.conf();

    After executing this command, primary member may get changed if all members have same priority.

    Verify that the tags are properly set by log in to on any member and executing rs.conf() command.

    SAMPLE OUTPUT 

    set04:PRIMARY> rs.conf(); 
{ 
        "_id" : "set04", 
        "version" : 319396, 
        "members" : [ 
                { 
                        "_id" : 0, 
                        "host" : "pcrfclient-arbiter-site3:27717", 
                        "arbiterOnly" : true 
                }, 
                { 
                        "_id" : 1, 
                        "host" : "sessionmgr01-site1:27717", 
                        "priority" : 4, 
                        "tags" : { 
                                "sessionLocalGeoSiteTag" : "Site1" 
                        } 
                }, 
                { 
                        "_id" : 2, 
                        "host" : "sessionmgr02-site1:27717", 
                        "priority" : 3, 
                        "tags" : { 
                                "sessionLocalGeoSiteTag" : "Site1" 
                        } 
                }, 
                { 
                        "_id" : 3, 
                        "host" : "sessionmgr01-site2:27717", 
                        "priority" : 2, 
                        "tags" : { 
                                "sessionLocalGeoSiteTag" : "Site2" 
                        } 
                }, 
                { 
                        "_id" : 4, 
                        "host" : "sessionmgr01-site2:27717", 
                        "priority" : 1, 
                        "tags" : { 
                                "sessionLocalGeoSiteTag" : "Site2" 
                        } 
                } 
        ], 
        "settings" : { 
                "heartbeatTimeoutSecs" : 1 
        } 
}

Publishing Configuration Changes When Primary Site becomes Unusable

Procedure

Step 1

Configure auto SVN sync across sites on GR secondary site.

  1. Configure this site's pcrfclient01 public IP in /var/qps/config/deploy/csv/AdditionalHosts.csv file of Cluster Manager (use same host name in 2.a).

    Example: public-secondary-pcrfclient01,,XX.XX.XX.XX,

    For OpenStack, add the new hosts in the AdditionalHosts section. For more information, refer to /api/system/config/additional-hosts section in CPS Installation Guide for OpenStack

  2. Recreate SVN repository on secondary site pcrfclient01/02.

    Note

     

    Take backup of SVN repository for rollback purpose.

    From pcrfclient01, execute the following commands:

    /bin/rm -fr /var/www/svn/repos

    /usr/bin/svnadmin create /var/www/svn/repos

    /etc/init.d/vm-init-client

    From pcrfclient02, execute the following commands:

    /bin/rm -fr /var/www/svn/repos

    /usr/bin/svnadmin create /var/www/svn/repos

    /etc/init.d/vm-init-client

  3. Login to pcrfclient01 and recover svn using /var/qps/bin/support/recover_svn_sync.sh script.

  4. Verify SVN status using diagnostics.sh script from Cluster Manager. Output should look like:

    diagnostics.sh --svn 
    CPS Diagnostics HA Multi-Node Environment
---------------------------
Checking svn sync status between pcrfclient01 and pcrfclient02...[PASS]

Step 2

Configure auto SVN sync across site on GR primary site.

For OpenStack, add the new hosts in the AdditionalHosts section. For more information, refer to /api/system/config/additional-hosts section in CPS Installation Guide for OpenStack

  1. Configure remote/secondary pcrfclient01 public IP in /var/qps/config/deploy/csv/AdditionalHosts.csv file of Cluster Manager.

    Example: public-secondary-pcrfclient01,,XX.XX.XX.XX,

  2. Configure svn_slave_list as svn_slave_list,pcrfclient02 public-secondary-pcrfclient01 in /var/qps/config/deploy/csv/Configuration.csv file of Cluster Manager (replace public-secondary-pcrfclient01 with the host name assigned in Step 2 a).

  3. Configure svn recovery to be every 10 minutes that is, auto_svn_recovery,enabled in /var/qps/config/deploy/csv/Configuration.csv file of Cluster Manager.

  4. Execute /var/qps/install/current/scripts/import/import_deploy.sh on Cluster Manager.

  5. Execute the following command to validate the imported data:

    cd /var/qps/install/current/scripts/deployer/support/

    python jvalidate.py

    Note

     

    The above script validates the parameters in the Excel/csv file against the ESX servers to make sure ESX server can support the configuration and deploy VMs.

    Note

     

    If you are deploying the VMs using the --nossh feature, jvalidate will not work as jvalidate needs SSH login credentials to get the memory, CPU and network details from the ESXi. Since --nonssh feature doesn’t have the SSH login credentials, the script execution fails for getting the details. You can login to the vCenter where the ESXi’s are mapped and check the required details to plan the VM design.

  6. Login to pcrfclient01 and re-initiate pcrfclient01 using /etc/init.d/vm-init-client command.

  7. Verify SVN status using diagnostics.sh script from Cluster Manager. Output should look like (Wait for maximum 10 mins as per cron interval for PASS status): 

    diagnostics.sh --svn
CPS Diagnostics HA Multi-Node Environment
---------------------------
Checking svn sync status between pcrfclient01 & pcrfclient02...[PASS]
Checking svn sync status between pcrfclient01 & remote-pcrfclient01...[PASS]
    Test scenario
    1. Both primary and secondary sites are up:

      • Login to primary site Policy Builder and update policy.

      • Verify primary site SVNs are in sync between pcrfclient01 and 02.

      • Verify primary site and secondary site SVN are in sync (this takes approx 10 mins as per cron interval).

    2. Both primary and secondary sites are up (This is not recommended).

      • Login to secondary site Policy Builder and update policy.

      • Verify secondary site SVNs are in sync between pcrfclient01 and 02.

      • Verify primary site and secondary site SVN are in sync (this takes approx 10 mins as per cron interval).

    3. Primary site up and secondary site down.

      • Login to primary site Policy Builder and update policy.

      • Verify primary site SVNs are in sync between pcrfclient01 and 02.

      • Recover secondary site. Verify primary site and secondary site SVN are in sync (this takes approx10 mins as per cron interval).

    4. Secondary site up and primary site down.

      • Login to secondary site Policy Builder and update policy.

      • Verify secondary site SVNs are sync between pcrfclient01 and 02.

      • Recover primary site. Verify primary site and secondary site SVN are in sync (this takes approx 10 mins as per cron interval).

Graceful Cluster Shutdown

Utility script /var/qps/bin/support/mongo/migrate_primary.sh is a part of the build and will be available in newly installed or upgraded setups.

This utility simply reads the cluster configuration file and mongoConfig.cfg file, and migrates Mongo Primary status from one cluster to another before doing an upgrade by setting priority '1' (or given in the command). After upgrade, utility script again migrates Mongo Primary status to original cluster by restoring the original priority.

Here is a help page of migrate_primary.sh (/var/qps/bin/support/mongo/migrate_primary.sh -help) utility: 

/var/qps/bin/support/mongo/migrate_primary.sh [--options]  
[--db-options] [--hosts-all|--hosts-files <host-file-name> --hosts <host list..> ] 
/var/qps/bin/support/mongo/migrate_primary.sh --restore <restore-filename>  
 
  --hosts      CPS Host list which are upgrading (like sessionmgr01, lb01, pcrfclient01) 
  --hosts-file File name which contains CPS upgrading hosts 
  --hosts-all  Upgrading all hosts from this cluster 
  --restore    Restore priority back 
 
DB Options - you can provide multiples DBs 
  --all         All databases in the configuration 
  --spr         All SPR databases in the configuration 
  --session     All Session databases in the configuration 
  --balance     All Balance databases in the configuration 
  --admin       All Admin databases in the configuration 
  --report      All Report databases in the configuration 
  --portal      All Portal databases in the configuration 
  --audit       All Aduit databases in the configuration 
 
Options: 
  --setpriority <priority-num> Set specific priority (default is 1) 
  --noprompt                  Do not ask verfiication & set priority, without y/n prompt 
  --prompt                    Prompt before setting priority (Default) 
  --nonzeroprioritychk        Validate all upgrading members have non-zero priority 
  --zeroprioritychk           Validate all upgrading members have zero priority 
  --debug                     For debug messages (default is non-debug) 
  --force                     Set priority if replica set is not healthy or member down case 
  --h [ --help ]              Display this help and exit 
 
Description: 
Reconfiguring Mongo DB priorities while doing an upgrade of a set of session managers, 
the Mongo DBs that exist on that session manager need to be moved to priority 1   
(provided priority) so that they will never be elected as the primary at time of upgrade. 
 
Examples: 
  /var/qps/bin/support/mongo/migrate_primary.sh --all --hosts sessionmgr01 
  /var/qps/bin/support/mongo/migrate_primary.sh --session --hosts-all 
  /var/qps/bin/support/mongo/migrate_primary.sh --noprompt --spr --hosts sessionmgr01 sessionmgr02 
  /var/qps/bin/support/mongo/migrate_primary.sh --setpriority 1 --all --hosts sessionmgr01

Note

When you execute migrate_primary.sh --restore command, ignore the following error message if encountered: 

[WARN] Failed to set priority to <priority number> to <hostname> (<set number>), due to "errmsg" : 
"replSetReconfig should only be run on PRIMARY, but my state is SECONDARY; use the "force" argument to override",

Active/Active Geo HA - Multi-Session Cache Port Support

CPS supports communication with multiple session cache databases and process the Gx and Rx messages in Active/Active Geo HA model.

The criteria to select which Session Cache for Gx Requests for a given session should be based on configurable criteria, for example, origin realm and/or host. Wildcards are also supported. For Rx requests, CPS uses secondaryKey lookup to load session.

When session cache database is not available, backup database is used.

By default, Geo HA feature is not installed and is not enabled. To install and enable the Geo HA, perform the following steps:


Note

To configure Active/Active Geo HA using APIs, refer to Active-Active Geo HA Support section in CPS Installation Guide for OpenStack.

Install Geo HA

Procedure

Step 1

Edit feature file on cluster manager: /etc/broadhop/pcrf/features

Step 2

Remove policy feature entry from feature file.

com.broadhop.policy.feature

Step 3

Add policy Geo HA feature entry in feature file.

com.broadhop.policy.geoha.feature

Step 4

Execute the following commands:

Note

 

${CPS version} - Input the CPS version in the following commands.

/var/qps/install/${CPS version}/scripts/build/build_all.sh

/var/qps/install/${CPS version}/scripts/upgrade/reinit.sh

Example:

/var/qps/install/9.0.0/scripts/build/build_all.sh

/var/qps/install/9.0.0//scripts/upgrade/reinit.sh

Enable Geo HA

Procedure

Step 1

Set GeoHA flag to true in qns.conf file to enable Geo HA feature.

-DisGeoHAEnabled=true

Step 2

Remove -DclusterFailureDetectionMS parameter from /etc/broadhop/qns.conf file.

Note

 

  • This parameter is also used to acquire the lock information to perform the RAR handling and writing reporting data.

  • When clusterFailureDetectionMS parameter is configured then the application interacts with the environment in Active-Standby mode.

  • Only the application running on the SITE where ADMIN replica-set primary member is available will be able to acquire the locks.

  • If clusterFailureDetectionMS parameter is not removed from qns.conf file, the application fails to acquire the locks running on all other clusters whose ADMIN replica-set is not primary and will affects the system behavior.

Step 3

Verify other site Policy Director (lb) IP addresses are present in /etc/hosts file.

If entries are missing, modify AdditionalHost.csv to add entries in /etc/hosts file. Remote load balancer should be accessible from load balancer by host name.

Configuration

Consider there are two sites Site1, Site2. Both are active sites and Site1 failovers to Site2.

Session database is replicated across site. Session database of the site can be selected based on realm or host or local information.

Procedure

Step 1

Configure the lookup type in qns.conf file. Possible values can be realm/host/local.

For example, -DgeoHASessionLookupType=realm

Note

 

When session lookup type is set to “local”, local session database is used for read/write session irrespective of site lookup configuration. For “local” session lookup type, site lookup configuration is not required. Even if it is configured, it is not used. However, user still needs to add site and shards.

For “local” session lookup type, add the entry for “diameter” under Lookaside Key Prefixes under Cluster configuration (if it is not already configured) in Policy Builder.

For local session capacity planning, refer to Local Session Affinity - Capacity Planning.

Note

 

For memory and performance impact when session lookup is configured as local, refer to Memory and Performance Impact.

Note

 

In case there is an error in configuring geoHASessionLookupType, CPS behaves incorrectly and drop messages. The following logs come continuously if there is an error in the configuration:

GeoHA is enabled, unknown lookuptype is configured: <>. Possible values are…

Step 2

Clean up the following data from the database if any data exists.

  1. (Session database) Clean the sessions:

    session_cache_ops.sh --remove

  2. Run diagnostics.sh command on pcrfclient01 and find session database primary member and port number.

    diagnostics.sh --get_replica_status

    Note

     

    If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

    Also, you can login to mongo on that member and check its actual status.

  3. Login to session database using the primary replica set hostname and port number.

    For example, mongo --host sessionmgr01 --port 27720

  4. (Admin database) Remove shard entries from shards collection:

    use sharding

    db.shards.remove({});

    db.buckets.drop(); ===> This collection is not used in GeoHA any more, so deleting.

  5. (Admin database) Clear endpoints:

    use queueing

    db.endpoints.remove({});

    use diameter

    db.endpoints.remove({});

  6. Exit the database:

    exit

Step 3

Enable dictionary reload flag (Only for GR) in /etc/broadhop/qns.conf file.

-DenableReloadDictionary=true

Step 4

Update the following parameters in /etc/broadhop/qns.conf file as per site IDs.

Examples:

-DGeoSiteName=Site1

-DSiteId=Site1

-DRemoteSiteId=Site2

Step 5

Add Sites - configure/add all physical sites.

  1. Login to qns OSGi console. All the following commands are to be executed from OSGi console.

    telnet qns01 9091

  2. Run addsite command for each primary (active) site with its secondary site ID.

    addsite <SiteId> <SecondarySiteId>

    where,

    <SiteId> is the primary site ID.

    <SecondarySiteId> is the secondary site ID.

    Example: addsite Site1 Site2

    This primary and secondary site IDs should be in sync with following entries in /etc/broadhop/qns.conf file.

    Examples:

    -DGeoSiteName=Site1 ===> (this should be <SiteId> from the above addsite command

    -DSiteId=Site1 ===> (this should be <SiteId> from the above addsite command)

    -DRemoteSiteId=Site2 ===> (this should be <SecondarySiteId> from above addsite command)

  3. Configure Site to realm or host mapping.

    User needs to configure all the realms for all the interfaces (such as, Gx, Rx, and so on) here:

    addsitelookup <SiteId> <LookupValue>

    where,

    <SiteId> is the primary site ID.

    <LookupValue> is the realm/host value.

    If geoHASessionLookupType is configured as realm in Step 1.

    Provide lookup value as realm as follows:

    Example: If you have multiple Gx and Gy clients connected to CPS and the details for realms of clients are as follows: 

    Client-1: pcef-1-Gx.cisco.com
Client-2: pcef-1-Gy.cisco.com
Client-3: pcef-2-Gx.cisco.com
Client-4: pcef-2-Gy.cisco.com

    For example, Client-1 and Client-2 are connected to Site1. Client-3 and Client-4 are connected to the Site2 then, 

    addsitelookup Site1 pcef-1-Gx.cisco.com
addsitelookup Site1 pcef-1-Gy.cisco.com
addsitelookup Site2 pcef-2-Gx.cisco.com
addsitelookup Site2 pcef-2-Gy.cisco.com

    If geoHASessionLookupType is configured as host in Step 1.

    Provide lookup value as host as follows:

    Example: If you have multiple Gx and Gy clients connected to CPS and the details for hostnames of clients are as follows: 

    Client-1: pcef-1-Gx
Client-2: pcef-1-Gy
Client-3: pcef-2-Gx
Client-4: pcef-2-Gy

    For example, Client-1 and Client-2 are connected to Site1. Client-3 and Client-4 are are connected to the Site2 then, 

    addsitelookup Site1 pcef-1-Gx
addsitelookup Site1 pcef-1-Gy
addsitelookup Site2 pcef-2-Gx
addsitelookup Site2 pcef-2-Gy

    Note

     

    The pattern matching is supported for site lookup mapping. In case the incoming host/realm does not match any of the values configured under LookupValues, request is dropped with the following exception in log:

    GeoHASiteMappingNotFound - No realm/host to site mapping matched for: <incoming value>

    Note

     

    If you configure the same realm/host look up pattern for both sites, application picks only one site look up (first one in the list for same realms/hosts) to process the incoming requests. If the configured realms/host look-ups entries are same for both sites and you send calls to any site, the application will always try to use one site look up (first one in the list for same realms/hosts) to process the sessions.

    So, Cisco recommends to configure identical realms/hosts names for the sites, and not configure same realms/hosts for the both sites.

    Other commands:

    • listsitelookup: To see all the configured site lookup mapping.

    • deletesitelookup <SiteId> <LookupValue>: To delete specific site lookup mapping.

  4. Add Shards: Configure/add shards for the site.

    addshard <Seed1>[,<Seed2>] <Port> <Index> <SiteId> [<BackupDb>]

    where,

    <SiteId> is the primary site of the shard. This maps the shard with site.

    [<BackupDb>] is an optional parameter.

    Example: addshard sessionmgr01,sessionmgr02 27717 1 Site1

    Note

     

    By default, there may not be any default shards added when Geo HA is enabled. So add the shards starting from index 1.

    To configure hot standby feature, use addshard command with backup database parameter:

    addshard <Seed1>[,<Seed2>] <Port> <Index> <SiteId> [<BackupDb>]

    Examples:

    addshard sessionmgr09,sessionmgr10 27717 1 Site1 backup

    addshard sessionmgr09,sessionmgr10 27717 2 Site1 backup

  5. Rebalance the shards by executing the following command:

    rebalance <SiteId>

    Example: rebalance Site1

  6. Migrate the shards by executing the following command:

    migrate <SiteId>

    Example: migrate Site1

Step 6

(Optional) The following parameter should be updated in /etc/broadhop/qns.conf file for SP Wi-Fi based deployments:

-DisRadiusBasedGeoHAEnabled=true

Note

 

For SP Wi-Fi based deployments, lookup value can be configured as NAS IP of ASR1K or ASR9K in Step 5.

Note

 

RADIUS-based policy control is no longer supported in CPS 14.0.0 and later releases as 3GPP Gx Diameter interface has become the industry-standard policy control interface.

Memory and Performance Impact

Performance Impact

In in-service migration (ISSU), if sessions are stored in remote database, there are additional queries on memcache and session databases. This leads to performance impact till all the sessions are migrated to local.

In fresh installation, there is no performance impact.

Memory Impact

As additional primary keys are stored in the cache ring, it takes additional memory for storing primary keys in memcache.

Failover Impact

In case of failover, Policy Server (QNS) queries local session database first and if sessions are not found in local, it queries memcache to find the remote shard and loads session from remote database.

As there is one additional local query, it leads to some performance impact during/after failover period till all the existing sessions are migrated. There is no impact for new sessions.

GR Configuration with Session Replication Across Sites

Login to pcrfclient01/02 to create/update rings from Policy Server (QNS) OSGi console. Assuming there are two session cache replica-sets. By default, Ring-1 Set-1 get configured automatically and remaining rings need to be configured manually.

Configure cluster-1 (Ring-1) 

osgi> setSkRingSet 1 1 <hostname_primary_site_sessionmgr01>:11211,
<hostname_primary_site_sessionmgr02>:11211
Ring updated
osgi> setSkRingSet 1 2 <hostname_primary_site_sessionmgr03>:11211,
<hostname_primary_site_sessionmgr04>:11211
Ring updated

Configure cluster-2 (Ring-2) 

telnet qns01 9091
osgi> createSkRing 2
Successfully added ring with ID: 2
osgi> setSkRingSet 2 1 <hostname_secondary_site_sessionmgr01>:11211,
<hostname_secondary_site_sessionmgr02>:11211
osgi> setSkRingSet 2 2 <hostname_secondary_site_sessionmgr03>:11211,
<hostname_secondary_site_sessionmgr04>:11211

An example configuration is given below:

  • Configure cluster-1 (Ring-1): 

    osgi> setSkRingSet 1 1 L2-CA-PRI-sessionmgr01:11211, L2-CA-PRI-sessionmgr02:11211
Ring updated
osgi> setSkRingSet 1 2 L2-CA-PRI-sessionmgr03:11211, L2-CA-PRI-sessionmgr04:11211
Ring updated

  • Configure cluster-2 (Ring-2): 

    telnet qns01 9091
osgi> createSkRing 2
Successfully added ring with ID: 2
osgi> setSkRingSet 2 1 L2-CA-SEC-sessionmgr01:11211, L2-CA-SEC-sessionmgr02:11211
osgi> setSkRingSet 2 2 L2-CA-SEC-sessionmgr03:11211, L2-CA-SEC-sessionmgr04:11211

Local Session Affinity - Capacity Planning

Consider there are two sites Site-1 and Site-2. Site-1 failovers to Site-2 and vice-versa.

With Local session affinity, both sites store new sessions to their local database in normal and failover condition. In normal conditions both sites need two session shards as shown in Site1 and Site2 - Normal Conditions.

Figure 20. Site1 and Site2 - Normal Conditions


Site-1 writes sessions to Shard1 and Shard2, while Site-2 writes to Shard3 and Shard4.

Now consider, Site-2 goes down and Site-1 receives traffic for both the sites. Post failover, for new sessions, Site-1 will use Shard-1 and Shard-2 and it will not use the Shard-3 and Shard-4.

Site-1 needs to have extra session shards to handle the traffic from Site-1 and Site2 (after failover). Same for Site-2 (refer to Failover Scenario with additional shards).

Figure 21. Failover Scenario


Site-1 writes sessions to Shard1, Shard2, Shard5, Shard6 and Site-2 writes to Shard3, Shard4, Shard7, Shard8. Now when failover occurs both sites have additional local shards to accommodate sessions from another site.

Limitation

In this case, for profile refresh scenario, there is no 'smart' search (for example, IMSI-based match in given set of shards). In case a session for given profile is not found in concerned site's all session shards, search would be extended to all shards in all the sites.

For SP Wi-Fi deployments: Portal call flows are currently not supported in SP Wi-Fi Active-Active with session replication deployments. Currently, only ASR1K and ASR9K devices are supported for Active-Active deployments.


Note

RADIUS-based policy control is no longer supported in CPS 14.0.0 and later releases as 3GPP Gx Diameter interface has become the industry-standard policy control interface.

Handling RAR Switching

When Gx session is created on Site2 and SPR update or Rx call comes on Site1, CPS sends Gx RAR from Site1 and in response PCEF sends RAA and next CCR request to Site1.

This leads to cross-site call switches from Site2 to Site1. If there are lot of call switches, Site1 may get overloaded.

By default, cross-site switching is enabled in CPS. To prevent cross-site switching, user needs to configure -DRemoteGeoSiteName parameter in /etc/broadhop/qns.conf file. This parameter enables cross-site communication for outbound messages like for RAR if we do not have DRA outside policy director (lb) and want to avoid RAR Switches.

Parameter Name: -DRemoteGeoSiteName=<sitename>

where, <sitename> is remote site name to be provided, and only to be added if we want to disable Gx-RAR switches from PCRF. It should match with -DRemoteSiteId.

Example:

-DRemoteSiteId=clusterA_SBY

-DRemoteGeoSiteName=clusterA_SBY

Prerequisite: Both remote site and local site policy server (QNS) should be able to communicate to load balancer on same interfaces. To change interface, flag -Dcom.broadhop.q.if=<enter replication interface name> can be used.

After configuring -DRemoteGeoSiteName parameter in qns.conf file, execute the following commands from Cluster Manager:

/var/qps/bin/control/copytoall.sh

/var/qps/bin/control/restartall.sh


Caution

Executing restartall.sh will cause messages to be dropped.

If Redis IPC is used, make sure remote/peer policy director (lb) VM information is configured on local site for RAR switching to work. For more information, refer to Policy Director (lb) VM Information on Local Site.

Configure Cross-site Broadcast Messaging

The cross-site broadcast message configuration is required when there are separate sessions (no replication for sessions DB) but common subscriber profile and subscriber provisioning event needs to be done on single site. In this case, the profile updates for subscriber sessions on remote sites need to be broadcasted to respective sites so that the corresponding RARs go from the remote sites to their respective diameter peers.

Procedure

Edit /etc/broadhop/qns.conf file and add the following line:

-DclusterPeers=failover:(tcp://<remote-site-lb01>:<activemq port>,tcp://<remote-site-lb02>:<activemq port>)?updateURIsSupported=false!<remote-site-cluster-name>.default

where,

  • <remote-site-lb01> is the IP address of the remote site lb01.

  • <activemq port> is the port on which activemq is listening. Default is 61616.

  • <remote-site-lb02> is the IP address of the remote site lb02.

  • <remote-site-cluster-name> is the cluster name of remote site. To get the cluster name of remote site, check the parameter value of -Dcom.broadhop.run.clusterId in /etc/broadhop/qns.conf file on remote site.

Example:

-DclusterPeers=failover:(tcp://107.250.248.144:61616,tcp://107.250.248.145:61616)?updateURIsSupported=false!Cluster-Site-2.default

Example

The following example considers three sites (SITE-A, SITE-B and SITE-C) to configure cluster broadcast messaging between them.

Note

Separator between two site configurations is colon (;).

  • SITE-A configuration: Edit /etc/broadhop/qns.conf file and add the following lines:

    -Dcom.broadhop.run.clusterId=Cluster-Site-A

    -DclusterPeers=failover:(tcp://105.250.250.150:61616,tcp://105.250.250.151:61616)?

    updateURIsSupported=false!Cluster-SITE-B.default;failover:(tcp://105.250.250.160:61616,

    tcp://105.250.250.161:61616)?updateURIsSupported=false!Cluster-SITE-C.default

  • SITE-B configuration: Edit /etc/broadhop/qns.conf file and add the following lines:

    -Dcom.broadhop.run.clusterId=Cluster-Site-B

    -DclusterPeers=failover:(tcp://105.250.250.140:61616,tcp://105.250.250.141:61616)?

    updateURIsSupported=false!Cluster-SITE-A.default;failover:(tcp://105.250.250.160:61616,

    tcp://105.250.250.161:61616)?updateURIsSupported=false!Cluster-SITE-C.default

  • SITE-C configuration: Edit /etc/broadhop/qns.conf file and add the following lines:

    -Dcom.broadhop.run.clusterId=Cluster-Site-C

    -DclusterPeers=failover:(tcp://105.250.250.140:61616,tcp://105.250.250.141:61616)?

    updateURIsSupported=false!Cluster-SITE-A.default;failover:(tcp://105.250.250.150:61616,

    tcp://105.250.250.151:61616)?updateURIsSupported=false!Cluster-SITE-B.default

Configure Redundant Arbiter (arbitervip) between pcrfclient01 and pcrfclient02

After the upgrade is complete, if the user wants a redundant arbiter (ArbiterVIP) between pcrfclient01 and pcrfclient02, perform the following steps:

Currently, this is only supported for HA setups.

Procedure

Step 1

Update the AdditionalHosts.csv and VLANs.csv files with the redundant arbiter information:

  • Update AdditionalHosts.csv:

    Assign one internal IP for Virtual IP (arbitervip).

    Syntax:

    <alias for Virtual IP>,<alias for Virtual IP>,<IP for Virtual IP>

    For example,

    arbitervip,arbitervip,< IP for Virtual IP>

  • Update VLANs.csv:

    Add a new column Pcrfclient VIP Alias in the VLANs.csv file to configure the redundant arbiter name for the pcrfclient VMs:

    Figure 22. VLANs.csv


    Execute the following command to import csv files into the Cluster Manager VM:

    /var/qps/install/current/scripts/import/import_deploy.sh

    This script converts the data to JSON format and outputs it to /var/qps/config/deploy/json/.

Step 2

SSH to the pcrfclient01 and pcrflclient02 VMs and run the following command to create arbitervip:

/etc/init.d/vm-init-client

Step 3

Synchronize /etc/hosts files across VMs by running the following command the Cluster Manager VM:

/var/qps/bin/update/synchosts.sh

Moving Arbiter from pcrfclient01 to Redundant Arbiter (arbitervip)

In this section we are considering the impacts to a session database replica set when the arbiter is moved from the pcrfclient01 VM to a redundant arbiter (arbitervip). The same steps need to be performed for SPR/balance/report/audit/admin databases.

Procedure

Step 1

Make sure all the members are up and no member is down,

Step 2

Shutdown the gr-arbiter node. Shutdown causes all the arbiter member to go into Unknown stage.

   #  on arbiter node run shutdown -P -h now

Step 3

Check with diagnostic --get_r that only arbiter members are in unknown state.

Step 4

Copy the mongoConfig.cfg files to all the nodes. For example, use the following command:

SSHUSER_PREFERROOT=true copytoall.sh /etc/broadhop/mongoConfig.cfg /etc/broadhop/mongoConfig.cfg

Step 5

Remove pcrfclient01 from replica set (set01 is an example in this step) by executing the following command from Cluster Manager:

To find the replica set from where you want to remove pcrfclient01, refer to your /etc/broadhop/mongoConfig.cfg file.

build_set.sh --session --remove-members --setname set01

This command asks for member name and port number. You can find the port number from your /etc/broadhop/mongoConfig.cfg file. 

Member:Port   --------> pcrfclient01:27717
pcrfclient01:27717
Do you really want to remove [yes(y)/no(n)]: y

Step 6

Verify whether the replica set member has been deleted by executing the following command from Cluster Manager: 

diagnostics.sh --get_replica_status
 
|----------------------------------------------------------------------------------------------|
| SESSION:set01                                                                                |
|  Member-1  -  27717 : 221.168.1.5  -  PRIMARY   -  sessionmgr01 - ON-LINE  -  --------  -  1 |
|  Member-2  -  27717 : 221.168.1.6  -  SECONDARY -  sessionmgr02 - ON-LINE  -   0 sec    -  1 |
|----------------------------------------------------------------------------------------------|

The output of diagnostics.sh --get_replica_status should not display pcrfclient01 as the member of replica set (set01 in this case).

Note

 

If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

Also, you can login to mongo on that member and check its actual status.

Step 7

Change arbiter member from pcrfclient01 to redundant arbiter (arbitervip) in the /etc/broadhop/mongoConfig.cfg file by executing the following command from Cluster Manager: 

vi /etc/broadhop/mongoConfig.cfg
[SESSION-SET1]
SETNAME=set01
OPLOG_SIZE=1024
ARBITER1=pcrfclient01:27717             <-- change pcrfclient01 to arbitervip
ARBITER_DATA_PATH=/var/data/sessions.1
MEMBER1=sessionmgr01:27717
MEMBER2=sessionmgr02:27717
DATA_PATH=/var/data/sessions.1
[SESSION-SET1-END]

Step 8

Edit the /etc/broadhop/mongoConfig.cfg file add a new replica set member. Run build_etc.sh to accept the updated configuration and wait for AIDO server to create the replica-set. `

To verify the replica-set has been added, run the following command:

build_set.sh --session

Step 9

Verify whether the replica set member is UP by executing the following command from Cluster Manager:

diagnostics.sh --get_replica_status 
|----------------------------------------------------------------------------------------|
| SESSION:set01                                                                                   |
|  Member-1  -  27717 : 221.168.1.5   -  PRIMARY    -  sessionmgr01  - ON-LINE  -  --------  -  1 |
|  Member-2  -  27717 : 221.168.1.6   -  SECONDARY  -  sessionmgr02  - ON-LINE  -   0 sec    -  1 |
|  Member-3  -  27717 : 221.168.1.9   -  ARBITER    -  arbitervip    - ON-LINE  -  --------  -  1 |
|-------------------------------------------------------------------------------------------------|

The output of diagnostics.sh --get_replica_status should now display arbitervip as the member of replica set (set01 in this case).

Note

 

If a member is shown in an unknown state, it is likely that the member is not accessible from one of other members, mostly an arbiter. In that case, you must go to that member and check its connectivity with other members.

Also, you can login to mongo on that member and check its actual status.
After ISSU, if arbitervip is not moving to upgraded site, see Troubleshooting section in the CPS Migration and Upgrade Guide.