[
https://issues.apache.org/jira/browse/ASTERIXDB-3574?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17944612#comment-17944612
]
ASF subversion and git services commented on ASTERIXDB-3574:
------------------------------------------------------------
Commit 2667a63b72c12798803f6c4da5fc5bc406bbb216 in asterixdb's branch
refs/heads/master from Ritik Raj
[ https://gitbox.apache.org/repos/asf?p=asterixdb.git;h=2667a63b72 ]
[ASTERIXDB-3574][STO] Taking resource-level lock instead of global lock
- user model changes: no
- storage format changes: no
- interface changes: yes
Details:
Currently, all dataset and index metadata operations (`create`,
`register`, `open`, `close`, and `unregister`) in
`DatasetLifecycleManager` are synchronized using a global lock,
leading to contention and limiting concurrency.
To improve performance, the global lock is replaced with
resource-specific locks where applicable. This allows operations
on different datasets or indexes to proceed in parallel.
Ext-ref: MB-65695
Change-Id: I9e3b931b363f082f0a7c69a0454adfd37937bb60
Reviewed-on: https://asterix-gerrit.ics.uci.edu/c/asterixdb/+/19484
Reviewed-by: Murtadha Hubail <[email protected]>
Tested-by: Jenkins <[email protected]>
Integration-Tests: Jenkins <[email protected]>
> Enhance concurrency in DatasetLifecycleManager by replacing the global lock
> with resource-specific locks where applicable.
> --------------------------------------------------------------------------------------------------------------------------
>
> Key: ASTERIXDB-3574
> URL: https://issues.apache.org/jira/browse/ASTERIXDB-3574
> Project: Apache AsterixDB
> Issue Type: Bug
> Components: STO - Storage
> Affects Versions: 0.9.10
> Reporter: Ritik Raj
> Assignee: Ritik Raj
> Priority: Major
> Labels: triaged
> Fix For: 0.9.10
>
>
> Currently, all dataset and index metadata operations (create, register, open,
> close, and unregister) in DatasetLifecycleManager are synchronized using a
> global lock. This limits concurrency.
> To improve performance, the global lock can be *downgraded* to a
> resource-specific lock when operating on individual datasets or indexes. This
> will allow more concurrency while maintaining correctness.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
