Cassandra SOLR Rolling Upgrade

问题:

We have a cluster of 12 nodes, 6 DSE-SOLR and 6 DSE-Cassandra. When upgrading from 3.0 to 3.1 we noticed that requests through the SOLR interface were broken until all nodes had been upgraded.

• Is this limitation still present when upgrading from 3.1 to 3.2?
• Are there any gotchas to note when making the upgrade?
• In the upgrade path docs it says to enable the old gossip protocol until all nodes have been upgraded, is this per DC or for the entire cluster?

回答1:

Russ,

What errors are you getting when running a query during an upgrade? Is it one of the following? http://www.datastax.com/docs/datastax_enterprise3.1/upgrade/version_3_1_0 located in the solr section of expected errors?

Is this limitation still present when upgrading from 3.1 to 3.2?

There isn't one that I'm aware of.

Are there any gotchas to note when making the upgrade? Just the 2 mentioned here : http://www.datastax.com/docs/datastax_enterprise3.2/upgrade/version_3_2_0#dse-upgrade-version3-2-0

1. Enabling the old gossip protocol
2. Altering the dse_system keyspace to use the Everywhere replication strategy

In the upgrade path docs it says to enable the old gossip protocol until all nodes have been upgraded, is this per DC or for the entire cluster?

Cluster wide, if you do not temporarily enable the old gossip protocol, the new nodes on 3.2 will not be able to talk to the nodes on 3.1 and vice versa which can lead to some nodes not seeing the rest of the cluster. Remember to remove JVM_OPTS="$JVM_OPTS -Denable-old-dse-state=true" from your cassandra-env.sh on all your nodes once you've upgraded and do another rolling restart.

Ham