[
https://issues.apache.org/jira/browse/ZOOKEEPER-3408?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16854641#comment-16854641
]
Dmitry Konstantinov commented on ZOOKEEPER-3408:
------------------------------------------------
Thank you for the link.
Let's say we have 5 nodes in ZK ensemble with IDs: 1, 2, 3, 4, 5.
1 - the current leader.
So, we are speaking about the following combination of multiple problems:
1) Network or ZK node delays happened for some 2 ZK nodes (let's say 4 and 5),
so last transactions are not propagated to them.
2) Then the current leader (1) is failed before an actual write of it's
transaction to disk and restarted.
2) Then during the leader re-election due to a network partition or target ZK
node delays none of the ZK nodes with the latest transaction info (nodes: 2,3)
were participated, so nodes 1, 4, 5 have elected a new leader which is not
aware about the last transactions.
Did I get your case correctly?
> Improve information about risks for forceSync config option
> -----------------------------------------------------------
>
> Key: ZOOKEEPER-3408
> URL: https://issues.apache.org/jira/browse/ZOOKEEPER-3408
> Project: ZooKeeper
> Issue Type: Improvement
> Components: documentation
> Reporter: Dmitry Konstantinov
> Priority: Minor
>
> [https://zookeeper.apache.org/doc/r3.5.5/zookeeperAdmin.html#Unsafe+Options]
> {quote}
> The following options can be useful, but be careful when you use them. The
> risk of each is explained along with the explanation of what the variable
> does.
> {quote}
> {quote}
> _forceSync_ : (Java system property: *zookeeper.forceSync*) Requires updates
> to be synced to media of the transaction log before finishing processing the
> update. If this option is set to no, ZooKeeper will not require updates to be
> synced to the media.
> {quote}
> The risks for this option are not very clear. Does it only the risk of loss
> some recent committed transactions if all Zookeeper instances from an
> ensemble are crashed and restarted at almost the same time or some other
> problems are also possible?
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
