HDDS-10943. Freon DN Echo should skip writing payload to ratis log

[jojochuang]

What changes were proposed in this pull request?

HDDS-10943. Freon DN Echo should skip writing payload to ratis log

Please describe your PR in detail:
Echo request transaction payload is saved into ratis log. The payload isn't used and causes rapid ratis log rotation

What is the link to the Apache JIRA

https://issues.apache.org/jira/browse/HDDS-10943

How was this patch tested?

This is an optimization; no functional change. Plus, it doesn't look easy to write a unit test around ContainerStateMachine within the current framework.

[kerneltime]

Maybe the right solution is to have 2 payloads, one that gets written to the log which presents the size of the request proto but serialized and another the data payload which does not get replication via the log? Not sure what will get measure this way when the payload is dropped entirely.

[jojochuang]

@kerneltime i want to do something similar to WriteChunk where the payload got sent to follower too.

[kerneltime]

@kerneltime i want to do something similar to WriteChunk where the payload got sent to follower too.

Makes sense. Do you want this PR to still go forward?

[adoroszlai]

/pending conflict

[github-actions]

Marking this issue as un-mergeable as requested.

Please use /ready comment when it's resolved.

Please note that the PR will be closed after 21 days of inactivity from now. (But can be re-opened anytime later...)

conflict

[szetszwo]

@jojochuang , thanks for working on this! Sorry that I missed this pr earlier. The change looks good. Some minor suggestions

• It is better to not building the intermediate protos since it may lead to more buffer copying.
• Combine the short lines.

    } else if (proto.getCmdType() == Type.Echo) {
      final EchoRequestProto echo = proto.getEcho();
      // skipping the payload field so it doesn't get written to the ratis log
      protoBuilder.setEcho(EchoRequestProto.newBuilder(echo).clearPayload()).build();
    }

[jojochuang]

Hm I think this PR is wrong. It should do what's similar to WriteChunk, do not remove data at startTransaction() because that way, follower don't get the data. Instead, ContainerStateMachine should remove data in write() and put back data in read()