krconv opened a new pull request, #7131:
URL: https://github.com/apache/hbase/pull/7131
This updates how the client reacts to region merges/splits while processing
`coprocessorService()`. Instead of failing the region, it resubmits the
affected range of row keys to try to save the call from needing to be aborted.
Added unit tests that test this scenario. I also compared this behavior to
the HTable from previous versions; HTable doesn't seem to have this same
assertion, nor proper handling for split regions, meaning it will potentially
return incorrect results.
The coprocessor service logic in the client goes like this when we need to
submit RPCs to multiple regions:
1. Start a chain of `AsyncTableRegionLocator` requests, starting with the
start key and continuing to locate the next region until we've located the
region containing the end key
2. Via a listener on the locator request for each region, we send the
coprocessor RPC to the region that was located as soon as the location is
resolved (likely from the meta cache)
3. The RPC connection is started to the region, at which point we may find
out that the region name has changed (and abort the request for that region)
4. For each region, we continue to send coprocessor RPCs until
`PartialResultCoprocessorCallback::getNextCallable` returns `null`
5. Once the last RPC has returned from the last region that the call had
identified, we call `CoprocessorCallback::onComplete` to indicate the call as
done.
This PR inserts logic at step 3, starting a new call to the affected range
covered by the (outdated) region returned by step 2.
Marking this as draft because I'll need some time to fix formatting/linting
issues, but I'd like to get tests running.
--
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]
