[jira] [Comment Edited] (IGNITE-7003) Ability to disable WAL (Non recoverable case)

["Vladimir Ozerov (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/IGNITE-7003?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16317970#comment-16317970 ] 

Vladimir Ozerov edited comment on IGNITE-7003 at 1/9/18 8:09 AM:
-----------------------------------------------------------------

[~avinogradov], solution was adjusted to address more cases, including non-affinity server nodes, client disconnect, operation queue, stale and duplicate messages, coordinator failover. Logic was abstracted to separate shared cache manager ({{WalStateManager}}). File-based persistence was removed for now, we will adopt meta store later. SQL support was added through {{ALTER TABLE}} command. 

Outstanding issues:
1) Improve exchange handling - we should wait for dirty page mark synchronously, and then await checkpoint and send messages asynchronously
2) Clean up log() method usages
3) Add more concurrency tests (listed in {{WalModeChangeAbstractSelfTest}})
4) Cleanup benchmarks.


was (Author: vozerov):
[~avinogradov], solution was adjusted to address more cases, including non-affinity server nodes, client disconnect, operation queue, stale and duplicate messages, coordinator failover. Logic was abstracted to separate shared cache manager. File-based persistence was removed for now, we will adopt meta store later. SQL support was added through {{ALTER TABLE}} command. 

Outstanding issues:
1) Improve exchange handling - we should wait for dirty page mark synchronously, and then await checkpoint and send messages asynchronously
2) Clean up log() method usages
3) Add more concurrency tests
4) Cleanup benchmarks.

> Ability to disable WAL (Non recoverable case)
> ---------------------------------------------
>
>                 Key: IGNITE-7003
>                 URL: https://issues.apache.org/jira/browse/IGNITE-7003
>             Project: Ignite
>          Issue Type: Sub-task
>          Components: cache, persistence, sql
>            Reporter: Anton Vinogradov
>            Assignee: Anton Vinogradov
>             Fix For: 2.4
>
>
> 1) WAL should be disabled by custom discovery message 
> - without triggering exchange,
> - without triggering checkpoint (since it's non recoverable case)
> In case someone is trying to disable already disabled WAL he should be notified that WAL already disabled
> Only cachegroups containing one cache can de disabled
> 2) WAL should be prepared to be enabled by custom discovery message with 
> - triggering exchange, 
> - disabling cache proxies,
> - waiting for checkpoints at every node.
> In case someone is trying to enable already enablling WAL he should be notified that enabling in progress.
> 3) WAL should be enabled by custom discovery message 
> - without triggering exchange
> but
> - with enabling proxies 
> once all nodes finished their checkpoints.
> Failover:
> On any failure during loading (while WAL disabled or enabling) we should be able to reactivate cluster without affected caches content.
> Originating node fail should be covered (at WAL disabled and enabling cases)
> API:
> Public API should be located at IgniteCluster



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)