[ https://issues.apache.org/jira/browse/IGNITE-12227?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17016126#comment-17016126 ]
Anton Kalashnikov commented on IGNITE-12227: -------------------------------------------- [~mmuzaf], Unfortunately, these changes found the bug in discovery and I had stuck on the investigation of this problem for a while. But looks like I've understood the issue and I have some solution. I believe I'll extract this problem to another ticket because it's not simple changes. And I hope in the nearest time I'll finish this task. > Default auto-adjust baseline enabled flag calculated incorrectly in some cases > ------------------------------------------------------------------------------ > > Key: IGNITE-12227 > URL: https://issues.apache.org/jira/browse/IGNITE-12227 > Project: Ignite > Issue Type: Bug > Reporter: Anton Kalashnikov > Assignee: Anton Kalashnikov > Priority: Blocker > Fix For: 2.8 > > Time Spent: 10m > Remaining Estimate: 0h > > baselineAutoAdjustEnabled can be been different on different nodes because of > the calculation of default value happening locally on each node and including > only local configuration. It issue can happen by the following reasons: > * If IGNITE_BASELINE_AUTO_ADJUST_ENABLED flag set to a different value on > different nodes it leads to cluster hanging due to baseline calculation > finishing with the unpredictable state on each node. > * if cluster in mixed mode(included in-memory and persistent nodes) sometimes > flag is set to a different value due to calculation doesn't consider remote > nodes configuration. > Possible solution(both points required): > * Get rid of IGNITE_BASELINE_AUTO_ADJUST_ENABLED and replace it by the > explicit call of IgniteCluster#baselineAutoAdjustEnabled where it > required(test only). > * Calculating default value on the first started node as early as > possible(instead of activation) and this value always should be set to > distributed metastorage(unlike it happening now). It means that instead of > awaiting activation, the default value would be calculated by the first > started node. -- This message was sent by Atlassian Jira (v8.3.4#803005)