Janus Chow created HDDS-17787:
---------------------------------
Summary: SCM block deletion stops for good once
transactionToDNsCommitMap reaches its limit
Key: HDDS-17787
URL: https://issues.apache.org/jira/browse/HDDS-17787
Project: Apache Ozone
Issue Type: Improvement
Reporter: Janus Chow
{{SCMBlockDeletingService}} skips the whole round when
{{transactionToDNsCommitMap}} reaches
{{{}hdds.scm.block.deletion.txn.dn.commit.map.limit{}}}. Skipping the round
also skips {{cleanAllTimeoutSCMCommand}} (it runs inside
{{{}DeletedBlockLog#getTransactions{}}}) and the resending of transactions
whose commands timed out.
A delete command can be lost after SCM marked it sent, e.g. when the datanode
times out reading a large heartbeat response:
{{SCMNodeManager#processHeartbeat}} has already taken it off the command queue.
Its transactions stay in the map until they are resent and acknowledged. Once
enough of them pile up to reach the limit, nothing is resent any more, the map
never shrinks, and block deletion stays stopped until the SCM leader changes
(which clears the in-memory map).
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]