HDDS-10204. TypedTable.putWithBatch may leak if conversion of value throws

[adoroszlai]

What changes were proposed in this pull request?

TypedTable.putWithBatch defers closing CodecBuffer until the batch is committed. If value conversion (or putWithBatch itself) throws, the buffers are not added to the batch, so they need to be released eagerly.

Example value conversion error:

NullPointerException
	at org.apache.hadoop.ozone.protocol.proto.OzoneManagerProtocolProtos$S3Secret$Builder.setAwsSecret(OzoneManagerProtocolProtos.java)
	at org.apache.hadoop.ozone.om.helpers.S3SecretValue.getProtobuf(S3SecretValue.java:100)
	at org.apache.hadoop.hdds.utils.db.DelegatedCodec.toCodecBuffer(DelegatedCodec.java:85)
	at org.apache.hadoop.hdds.utils.db.Codec.toDirectCodecBuffer(Codec.java:73)
	at org.apache.hadoop.hdds.utils.db.TypedTable.putWithBatch(TypedTable.java:171)
	at org.apache.hadoop.ozone.om.OmMetadataManagerImpl$1.addWithBatch(OmMetadataManagerImpl.java:1949)
	at org.apache.hadoop.ozone.om.response.s3.security.OMSetSecretResponse.addToDBBatch(OMSetSecretResponse.java:80)

Key leak:

2024-01-24 09:49:09,454 [Finalizer] WARN  db.CodecBuffer (CodecBuffer.java:detectLeaks(278)) - LEAK 1: [0<=18<=54: 74 65 73 74 55 73 65 72 31 61 ...], refCnt=1, capacity=54 allocation:
org.apache.hadoop.hdds.utils.db.CodecBuffer.<init>(CodecBuffer.java:53)
...
org.apache.hadoop.hdds.utils.db.StringCodecBase.toCodecBuffer(StringCodecBase.java:174)
org.apache.hadoop.hdds.utils.db.StringCodec.toCodecBuffer(StringCodec.java:28)
org.apache.hadoop.hdds.utils.db.StringCodecBase.toCodecBuffer(StringCodecBase.java:42)
org.apache.hadoop.hdds.utils.db.Codec.toDirectCodecBuffer(Codec.java:73)
org.apache.hadoop.hdds.utils.db.TypedTable.putWithBatch(TypedTable.java:170)
org.apache.hadoop.ozone.om.OmMetadataManagerImpl$1.addWithBatch(OmMetadataManagerImpl.java:1949)
org.apache.hadoop.ozone.om.response.s3.security.OMSetSecretResponse.addToDBBatch(OMSetSecretResponse.java:80)

The root cause of this specific conversion problem was fixed in HDDS-10200. This PR is only for the CodecBuffer leak.

https://issues.apache.org/jira/browse/HDDS-10204

How was this patch tested?

TestMultiTenantVolume passed 10x100 (without the fix from HDDS-10200):
https://github.com/adoroszlai/ozone/actions/runs/7658045401

CI:
https://github.com/adoroszlai/ozone/actions/runs/7657732895

[szetszwo]

+1 the change looks good.

[adoroszlai]

Thanks @szetszwo for the review.