[
https://issues.apache.org/jira/browse/HDDS-12087?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wei-Chiu Chuang updated HDDS-12087:
-----------------------------------
Release Note:
A throttling mechanism has been added to the SCM to prevent it from becoming
unstable or crashing during large-scale block deletions. This
prevents excessive memory usage and long garbage collection pauses.
Impact:
This change improves SCM stability and reliability, making the Ozone cluster
more robust.
> TransactionToDNeCommitMap too large causes GC to pause for a long time
> ----------------------------------------------------------------------
>
> Key: HDDS-12087
> URL: https://issues.apache.org/jira/browse/HDDS-12087
> Project: Apache Ozone
> Issue Type: Sub-task
> Components: SCM
> Reporter: GuoHao
> Assignee: Aryan Gupta
> Priority: Critical
> Labels: pull-request-available
> Fix For: 2.1.0
>
>
> Description: When deleting the block status from the SCM record, there will
> be a `transactionToDNsCommitMap` structure record. We encountered this
> structure accumulating too much content and occupying too much memory,
> causing the SCM to have a long GC time of 327 seconds.
>
> GC log:
> {code:java}
> 2025-01-15 08:25:14,789 [JvmPauseMonitor0] ERROR
> org.apache.ratis.server.RaftServer: 127e9d82-790c-40c5-af90-050564a06a45: JVM
> pause detected 372.305s longer than the close-threshold 120s, shutting down
> ... {code}
>
> Solution: Before iterating the deleteBlocks table and sending a delete
> request to dn, if the content in the `transactionToDNsCommitMap` is found to
> be too large and exceeds a certain threshold, pause the iteration of the
> deleteBlocks table for a while.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
