CODEWITHSUNDEEP
cassandra
V.HL
V.HL

Reputation: 80

Cassandra restart cluster to operate maintenance

Currently, I have a cluster Cassandra 2.1.13 with 5 nodes in only one datacenter. I m going to do maintenance operation in this cluster. I will do it node by node.

I imagine 3 scenarios below in different contexts :

Before shutdown the node, i execute nodetool cmd "disablegossip, disablehandoff and drain", to boost the node startup.

1 - if the node is down less than "max_hint_windows_in_ms", i can start the node without another action

2 - if the node is down more than "max_hint_windows_in_ms" and less than repair windows "gc_grace_seconds", i start the node with a full repair

3 - if the node is down more than repair windows, i replace the node to avoid resurection data.

Are theses scenarios correct ? If it was you what do you do ?

Upvotes: 2

Views: 419

Answers (1)

Chris Lohfink
Chris Lohfink

Reputation: 16400

You could increase your gc_grace_seconds before starting. If using post C* 3.0 id increase HH window as well, but for 2.1 thats not a good idea (more incase someone reads this in future).

A proper node shutdown will handle disabling and draining, but if you want to be sure, disable thrift and native transport before gossip/HH. Disabling gossip before the transport and your client application may still try to use that node as coordinator for requests which wont work after leaving cluster.

Upvotes: 2

Related Questions