You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by "Shalin Shekhar Mangar (JIRA)" <ji...@apache.org> on 2015/12/10 17:18:11 UTC

[jira] [Commented] (SOLR-8368) Investigate a leader using older versions than it's replicas has for leader election peer sync after a 'crash' shutdown.

    [ https://issues.apache.org/jira/browse/SOLR-8368?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15051166#comment-15051166 ] 

Shalin Shekhar Mangar commented on SOLR-8368:
---------------------------------------------

bq. This means that a non-leader in shard construction state will no longer do a replay. I don't know enough about how shard splits work to know if this is ok, but I think the impact is that replicas will take longer to come online since they will need to do an additional recovery once the sub-shard leader is active. If that's the case then the rest of the patch LGTM.

bq. These sub shards should not have any tlogs to replay anyway though. If on some crazy failure case they did, replaying tlogs after a crash should never be a bad thing to do.

bq. Would be good for Shalin Shekhar Mangar or someone that knows shard splitting like him to take a look at it though.

You are right Mark. We only need to skip log replay for the sub-shard leader (because it starts buffering updates coming from parent shard leader during core construction). Non-leaders in construction don't really need log replay (they shouldn't have anything in their tlogs at all). 

{code}
// disable recovery in case shard is in construction state (for shard splits)
{code}
This comment can be fixed to say "disable log replay in case shard is in construction state (for shard splits)".

> Investigate a leader using older versions than it's replicas has for leader election peer sync after a 'crash' shutdown.
> ------------------------------------------------------------------------------------------------------------------------
>
>                 Key: SOLR-8368
>                 URL: https://issues.apache.org/jira/browse/SOLR-8368
>             Project: Solr
>          Issue Type: Bug
>            Reporter: Mark Miller
>            Assignee: Mark Miller
>         Attachments: SOLR-8368.patch
>
>
> If we do it after like now, the correct leader may not be able to become leader.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org