Chu Cheng Li created HDDS-17448:
-----------------------------------
Summary: gRPC OM client should retry a not-ready leader like the
Hadoop RPC client
Key: HDDS-17448
URL: https://issues.apache.org/jira/browse/HDDS-17448
Project: Apache Ozone
Issue Type: Sub-task
Reporter: Chu Cheng Li
h3. Problem
* The OM retry policy retries a not-ready leader on the same OM, after a wait.
It does this only for a {{ServiceException}}.
{{GrpcOmTransport#unwrapException}} returns the OM exception itself. Thus, on
{{OMLeaderNotReadyException}}, the gRPC client fails over to the next OM with
no wait.
* {{GrpcOmTransport}} retries only the leader path. The follower read loop in
{{GrpcOMFollowerReadFailoverProxyProvider}} has no retry around it. Thus, on
{{OMLeaderNotReadyException}}, the loop falls back to the leader path and sends
a new request at once.
* The Hadoop RPC client does not have these problems. Its {{RetryProxy}} wraps
the follower read and the leader path, and it gives the retry policy a
{{ServiceException}}.
h3. Changes
* {{GrpcOmTransport}}: move the retry loop from {{submitRequestToLeader}} to
{{submitRequest}}, around the follower read and the leader path. Give the retry
policy a {{ServiceException}} with the {{RemoteException}} cause.
* {{GrpcOMFollowerReadFailoverProxyProvider}}: throw
{{OMLeaderNotReadyException}} to the retry loop, as
{{HadoopRpcOMFollowerReadFailoverProxyProvider}} does.
h3. Tests
* The gRPC client retries a not-ready leader on the same OM, after a wait.
* A follower read that gets {{OMLeaderNotReadyException}} goes through the same
retry.
Found in the review of HDDS-16737:
https://github.com/apache/ozone/pull/11418#discussion_r4369504955
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
