Mohammadhani Fouladgar created HDDS-16723:
---------------------------------------------
Summary: DN's should stop caching container reports when SCM says
to reregister.
Key: HDDS-16723
URL: https://issues.apache.org/jira/browse/HDDS-16723
Project: Apache Ozone
Issue Type: New Feature
Reporter: Mohammadhani Fouladgar
*Problem*
After an SCM has been down for some time, a datanode re-registering with an SCM
follower can fail to register. While the datanode is disconnected, Incremental
Container Reports (ICRs) accumulate in the datanode's per-endpoint report
queue. When the datanode re-registers, the next report sent to that SCM bundles
the Full Container Report (FCR) together with the entire backlog of accumulated
ICRs in a single message. This oversized message can exceed the maximum
RPC/protobuf message size, so the request is rejected.
Worse, the datanode is not aware that registration failed: it keeps sending
heartbeats as though registered, and the SCM keeps rejecting them. The datanode
never successfully (re-)registers with that SCM.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
