Closed apanasevich closed 1 month ago
I've created the PR, but I'm not sure that the target branch is correct. It'll be great to see a fix in some of 4.5.*
versions
Hi @vietj
thank you for reviewing ang merging the PR to the main branch.
Should I create a new PR to fix vertion 4.5.*? It'll be great to get it fixed in production.
can you provide a PR for backport ?
can you provide a PR for backport ?
Sure. Will do it a little bit later
Hi @vietj
I've created the PR for a backport
Version
4.5.7
Context
My app is based on Vert.x with embedded Hazelcast cluster manager. When I run several instances of the app with HA mode enabled some instances cannot deploy HA verticles, and logs show that they have not attain quorum:
Quorum not attained. Deployment of verticle will be delayed until there's a quorum.
But actually they have.
The problem is that Hazelcast cluster could be in a state when all nodes have already joined at the moment when
HAManager
initialization starts. SoHAManager
s methodnodeAdded
will never be called. AlsocheckQuorum
method called oninit
see all nodes fromclusterManager
, butclusterMap
do not contain information from all nodes. They will put it in several milliseconds later of course, butcheckQuorum
method will be completed at that time and never be called again.Steps to reproduce
It just occurs in some cases.