[jira] [Commented] (IGNITE-9411) Lock: make sure lock timeouts works fine

["Ivan Pavlukhin (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/IGNITE-9411?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16597270#comment-16597270 ] 

Ivan Pavlukhin commented on IGNITE-9411:
----------------------------------------

Locks are handled similarly for following SQL statements: {{INSERT, MERGE, UPDATE, DELETE, SELECT ... FOR UPDATE}}. Here is the procedure:

_Check if row is already locked. If locked wait lock release. Otherwise place own lock. Locks are released upon transaction end._

Lock waiting should support timeout. If timeout exceeds transaction should be marked _rollback-only_. Timeout can be specified per statement or per transaction.

> Lock: make sure lock timeouts works fine
> ----------------------------------------
>
>                 Key: IGNITE-9411
>                 URL: https://issues.apache.org/jira/browse/IGNITE-9411
>             Project: Ignite
>          Issue Type: Task
>          Components: mvcc
>            Reporter: Vladimir Ozerov
>            Assignee: Ivan Pavlukhin
>            Priority: Major
>             Fix For: 2.7
>
>
> In SQL it is not uncommon that locks are taken in arbitrary order, what may lead to deadlocks. Fair deadlock detector is good solution in monolithic databases - just analyze dependency graph and kill one of conflicting transactions.
> We have a ticket to implement distributed deadlock detector in Ignite [1]. However, this solution is rather complex and may bring some overhead. 
> For now it is better to rely on some timeout (global or per-transaction), and rollback TX when it fails to lock certain entry for a long time. Probably we already have this feature in some form. Need to add verify that it works and add more tests if needed.
> [1] https://issues.apache.org/jira/browse/IGNITE-9322



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)