priyeshkaratha opened a new pull request, #11043:
URL: https://github.com/apache/ozone/pull/11043
## What changes were proposed in this pull request?
HDDS-15726 introduced rollback logic that reverts those counters on
IOException from addTransactionsToDB / removeTransactionsFromDB. That is
correct for definitive failures (not-leader, connection error,
state-machine IO error), but wrong for a Ratis timeout.
SCMRatisServerImpl.submitRequest calls
server.submitClientRequestAsync(request).get(requestTimeout, ms)
A TimeoutException from .get() only abandons the client-side wait. The
CompletableFuture is discarded, but the in-flight log entry continues through
the Ratis pipeline and can still be committed and applied after the caller has
already seen the exception. Rolling back the summary at that point produces:
- a double-decrement on the add path (summary goes negative), or
- an overcount on the remove path (summary stays inflated).
Both divergences persist until the next leader transfer, which reconciles
the counters by reloading from the durable state via
initDataDistributionData().
### Fix
Introduce isRatisTimeout(IOException) to distinguish ambiguous outcomes from
definitive failures. The check covers two cases:
- SCMException(ResultCodes.TIMEOUT): translateException maps a
java.util.concurrent.TimeoutException to this code.
- SCMException(ResultCodes.INTERNAL_ERROR) whose cause is
InterruptedException: .get() can also throw InterruptedException if the thread
is interrupted while waiting; the request is already in-flight, giving the same
ambiguous outcome as a timeout.
## What is the link to the Apache JIRA
HDDS-16142
## How was this patch tested?
Tested by added cases and existing tests.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
