[
https://issues.apache.org/jira/browse/HDDS-10590?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Hemant Kumar updated HDDS-10590:
--------------------------------
Description:
SstFilteringService marks the sstFilter flag for a snapshot and updates the
RocksDB,
[code|https://github.com/apache/ozone/blob/2f2234c7b61714404399ada8f31b3fb4772b613a/hadoop-ozone/ozone-manager/src/main/java/org/apache/hadoop/ozone/om/SstFilteringService.java#L123].
It may cause snapshot chain corruption (probably not after the fix:
https://github.com/apache/ozone/pull/6443) or data inconsistency due to race
condition because SstFilteringService updates snapshot info in parallel to
SnapshotPurge and SnapshotProperty which also update snapshot info. Even tho
SstFilteringService takes a lock before updating snapshotInfo, SnapshotPurge
and SnapshotProperty APIs don't take a lock and rely on OMStateMachince because
OMStateMachince is going to process each request sequentially.
In general, each update to snapshotInfoTable should go through API but that is
not possible for *SstFilteringService* because *SstFilteringService* runs on
each OM independently. Hence it is directly updating snapshotInfoTable.
So we need to introduce the lock for SnapshotPurge and SetSnapshotProperty APIs
unless there is another better way.
was:
SstFilteringService marks the sstFilter flag for a snapshot and updates the
RocksDB,
[code|https://github.com/apache/ozone/blob/2f2234c7b61714404399ada8f31b3fb4772b613a/hadoop-ozone/ozone-manager/src/main/java/org/apache/hadoop/ozone/om/SstFilteringService.java#L123].
It may cause snapshot chain corruption (probably not after the fix:
https://github.com/apache/ozone/pull/6443) or data inconsistency due to race
condition because SstFilteringService updates snapshot info in parallel to
SnapshotPurge and SnapshotProperty which also update snapshot info. Even tho
SstFilteringService takes a lock before updating snapshotInfo, SnapshotPurge
and SnapshotProperty APIs don't take a lock and rely on OMStateMachince because
OMStateMachince is going to process each request sequentially.
In my opinion, each update to snapshotInfoTable should go through API but that
is not possible for *SstFilteringService* because *SstFilteringService* runs on
each OM independently. Hence it is directly updating snapshotInfoTable.
So we need to introduce the lock for SnapshotPurge and SetSnapshotProperty APIs
unless there is another better way.
> SstFilteringService updating snapshotInfo directly
> --------------------------------------------------
>
> Key: HDDS-10590
> URL: https://issues.apache.org/jira/browse/HDDS-10590
> Project: Apache Ozone
> Issue Type: Sub-task
> Reporter: Hemant Kumar
> Priority: Major
>
> SstFilteringService marks the sstFilter flag for a snapshot and updates the
> RocksDB,
> [code|https://github.com/apache/ozone/blob/2f2234c7b61714404399ada8f31b3fb4772b613a/hadoop-ozone/ozone-manager/src/main/java/org/apache/hadoop/ozone/om/SstFilteringService.java#L123].
> It may cause snapshot chain corruption (probably not after the fix:
> https://github.com/apache/ozone/pull/6443) or data inconsistency due to race
> condition because SstFilteringService updates snapshot info in parallel to
> SnapshotPurge and SnapshotProperty which also update snapshot info. Even tho
> SstFilteringService takes a lock before updating snapshotInfo, SnapshotPurge
> and SnapshotProperty APIs don't take a lock and rely on OMStateMachince
> because OMStateMachince is going to process each request sequentially.
> In general, each update to snapshotInfoTable should go through API but that
> is not possible for *SstFilteringService* because *SstFilteringService* runs
> on each OM independently. Hence it is directly updating snapshotInfoTable.
> So we need to introduce the lock for SnapshotPurge and SetSnapshotProperty
> APIs unless there is another better way.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
