Details
-
Improvement
-
Resolution: Unresolved
-
Major
-
None
-
7.1, 7.1.3, 7.2
-
None
-
0
Description
Context:
With 7.0+ comes "unsafe mode" fail over.
It is mentionned inside our documentation:
In order that users do not have to abandon a cluster in which a majority of nodes have failed or are unreachable, hard failover can be performed in unsafe mode. Unsafe failovers override the consensus protocol; and may be performed even if a majority of nodes in the cluster cannot be contacted. No unsafe failover should be attempted without a full understanding of the consequences and requirements, which are listed immediately below. |
As described above, in Consequences of Unsafe Failover, following unsafe failover, the failed-over nodes cannot be recovered; and will need to be entirely re-initialized, in order to be re-introduced into the cluster.
|
The consequences of unsafe failover in Couchbase Server Version 7.0+ are therefore significantly different from in previous versions: previously, the failed over nodes remained in the cluster, and could be recovered; but in 7.0+, they are removed from the cluster, are non-recoverable, and must be re-initialized before being added back to the cluster. |
QUESTION: what do we mean EXACTLY saying following unsafe mode fail over, "failed over nodes must be re-initialized before being added back to the cluster" ?
Should the nodes be completely uninstall and re install?
I ask this question because I tested:
- doing nothing (just adding the node back) => NOT OK => the node thinks it is part of the cluster and see other nodes in orange/red state (see attachement my_tests.docx for all detailed tests containing screen capture)
- doing a
couchbase-cli node-initsame result => NOT OK => no change compared to previous attempts after node is re introduced.
So I think it means uninstall + re install node but would like confirmation.
Can anybody confirm?
Thanks