HDDS-11853. Avoid redundant UUID conversion in XceiverClientRatis hot path

[lpavanvenkat]

What changes were proposed in this pull request?

What and why
XceiverClientRatis.sendCommandAsync called RatisHelper.toDatanodeId(reply.getReplierId()) on every Ratis reply, which internally parses a UUID string via UUID.fromString — only to look up a DatanodeDetails object that is already available from the pipeline. This is redundant work on the hot write path.
Changes
In hadoop-hdds/client/src/main/java/org/apache/hadoop/hdds/scm/XceiverClientRatis.java:

• A Map<String, DatanodeDetails> field peerIdToDatanode is pre-built at construction time from pipeline.getNodes(), keyed by each datanode's UUID string.
• addDatanodetoReply(String, XceiverClientReply) is added to do a direct map lookup instead of calling RatisHelper.toDatanodeId (which called UUID.fromString) on every reply.
• sendCommandAsync is updated to call addDatanodetoReply(reply.getReplierId(), ...) directly, removing the intermediate UUID serverId variable.
  The map is always complete: Ozone Ratis pipelines have fixed membership — when a node fails, SCM closes the pipeline and creates a new one. The Ratis peer ID is always datanode.getUuidString() (RatisHelper.toRaftPeerIdString), the exact key used in the map. A miss is structurally impossible.
  No changes to RatisHelper, DatanodeID, or any other class.

What is the link to the Apache JIRA

https://issues.apache.org/jira/browse/HDDS-11853

How was this patch tested?

• Unit tests: mvn -pl :hdds-client test — 148 tests, all passed.
• Checkstyle: ./hadoop-ozone/dev-support/checks/checkstyle.sh — 0 violations across all 58 modules.
• Integration test: mvn -pl :ozone-integration-test test -Dtest=TestCommitInRatis — 2 tests passed.
• Verified via temporary LOG.info that every addDatanodetoReply call during TestCommitInRatis produced hit=true with zero fallback UUID parses.
  No UI changes. No Protobuf changes.

---

[ashishkumar50]

@lpavanvenkat please enable CI workflow in your fork branch.

[lpavanvenkat]

I enabled the CI workflow on my forked branch.

[adoroszlai]

Thanks @lpavanvenkat for the patch.

[adoroszlai]

The reply's list of datanodes is used only for watchForCommit commands, to mark failed nodes:

ozone/hadoop-hdds/client/src/main/java/org/apache/hadoop/hdds/scm/storage/BlockOutputStream.java

Lines 542 to 553 in 7f33fc5

	private void checkReply(XceiverClientReply reply) {
	if (reply == null) {
	return;
	}
	final List<DatanodeDetails> dnList = reply.getDatanodes();
	if (dnList.isEmpty()) {
	return;
	}
	LOG.warn("Failed to commit BlockId {} on {}. Failed nodes: {}",
	blockID, xceiverClient.getPipeline(), dnList);
	failedServers.addAll(dnList);

and

ozone/hadoop-hdds/client/src/main/java/org/apache/hadoop/hdds/scm/storage/BlockDataStreamOutput.java

Lines 383 to 393 in 7f33fc5

	XceiverClientReply reply = bufferFull ?
	commitWatcher.watchOnFirstIndex() :
	commitWatcher.watchOnLastIndex();
	if (reply != null) {
	List<DatanodeDetails> dnList = reply.getDatanodes();
	if (!dnList.isEmpty()) {
	Pipeline pipe = xceiverClient.getPipeline();
	LOG.warn("Failed to commit BlockId {} on {}. Failed nodes: {}",
	blockID, pipe, dnList);
	failedServers.addAll(dnList);

But here the node is added for all kinds of commands and regardless of its result.

Question: is it even necessary to add the datanode here? (I tried removing it and CI passed.)

CC @jojochuang @szetszwo