HDDS-13982. OmKeyInfo fileName possible inconsistency between cache and DB

[adoroszlai]

What changes were proposed in this pull request?

OmKeyInfo#fileName is not persisted to DB, but set during deserialization based on keyName. This happens even if the OmKeyInfo did not have fileName initially. Thus the fileName may be different for the instance added to the table cache and the one read from DB. To avoid that, FSO requests need to manually setFileName, which is error-prone.

This PR changes OmKeyInfo to set fileName whenever keyName is initialized (in build()) or updated.

https://issues.apache.org/jira/browse/HDDS-13982

How was this patch tested?

Updated unit and integration test case.

https://github.com/adoroszlai/ozone/actions/runs/19699233660

[adoroszlai]

@ChenSammi I believe this is a bug in hsync: order of keyName and fileName is mixed up.

Parameters:

ozone/hadoop-ozone/ozone-manager/src/main/java/org/apache/hadoop/ozone/om/request/file/OMFileRequest.java

Lines 469 to 471 in 40304b6

	public static void addOpenFileTableCacheEntry(
	OMMetadataManager omMetadataManager, String dbOpenFileName,
	@Nullable OmKeyInfo omFileInfo, String fileName, String keyName, long trxnLogIndex) {

Arguments:

ozone/hadoop-ozone/ozone-manager/src/main/java/org/apache/hadoop/ozone/om/request/key/OMKeyCommitRequestWithFSO.java

Lines 210 to 211 in 40304b6

	OMFileRequest.addOpenFileTableCacheEntry(omMetadataManager,
	dbOpenKeyToDeleteKey, openKeyToDelete, keyName, fileName, trxnLogIndex);

Should we be passing keyName instead of fileName?

[ChenSammi]

@adoroszlai , good catch! Yeah, keyName and fileName was accidentally passed in with a reversed order.

[adoroszlai]

@smengcl could you please review?

[chungen0126]

Thanks @adoroszlai for the fix.

Would It be better to create an OmKeyInfoWithFSO subclass that extends OmKeyInfo for FSO buckets. We could then move the fileName-related methods (e.g., getFileName & getPath) to OmKeyInfoWithFSO.

This approach would prevent from calling these methods during OBS operations.

[ChenSammi]

Thanks @adoroszlai for this improvement. The last patch LGTM.

[adoroszlai]

Thanks @ChenSammi, @chungen0126 for the review.

Would It be better to create an OmKeyInfoWithFSO subclass that extends OmKeyInfo for FSO buckets. We could then move the fileName-related methods (e.g., getFileName & getPath) to OmKeyInfoWithFSO.

I'll give it a try in a follow-up.