Recovery procedure when contrail database is down for greater than gc_grace_seconds - Juniper/contrail-controller GitHub Wiki

When contrail-database has been down (either due to network partitioning or process stopped) for greater than gc_grace_seconds (defaulted to 10 days), the contrail-database init.d script will not start it. On issuing the service contrail-database start command, user will see a message similar to

Cassandra has been down for at least 777600 seconds, not starting

In this scenario, if contrail-database is brought online without following the below procedure, then it can result in inconsistent configuration database.

To recover from this situation the following steps need to be followed:

Remove the cassandra node.

For cassandra 1.2.x, the steps are at https://docs.datastax.com/en/cassandra/1.2/cassandra/operations/ops_remove_node_t.html.

For cassandra 2.1.x, the steps are at https://docs.datastax.com/en/cassandra/2.1/cassandra/operations/opsRemoveNode.html

Note: The nodetool removenode command mentioned in the steps above needs to run on the other cassandra nodes since cassandra is already stopped on the node to be removed.

If this is the analytics DB, clear the analytics data (the one that gets corrupted more often).

rm -rf /var/lib/cassandra/commitlog/*

rm -rf /var/lib/cassandra/ContrailAnalyticsCql/*

Delete the cassandra cluster status file

rm -f /var/log/cassandra/status-up

Start contrail-database

service contrail-database start