[
https://issues.apache.org/jira/browse/HBASE-14521?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14950283#comment-14950283
]
Nicolas Liochon commented on HBASE-14521:
-----------------------------------------
It's a good point: the existing implementation is confusing.
The patch looks good. It contains a lot of cleanup that will make the code
easier to read (thanks, Yu!)
I'm surprised by this:
{code}
@@ -137,7 +137,6 @@ public class TestAsyncProcess {
AsyncRequestFutureImpl<Res> r = super.createAsyncRequestFuture(
DUMMY_TABLE, actions, nonceGroup, pool, callback, results,
needResults);
allReqs.add(r);
- callsCt.incrementAndGet(); <=========== We should continue to count the
calls, no?
return r;
}
{code}
Note that setting retries to zero is most of the time an error as we can have a
retry in many cases, for example iif the client cache is not up to date
(contains the wrong region server for a region).
> Unify the semantic of hbase.client.retries.number
> -------------------------------------------------
>
> Key: HBASE-14521
> URL: https://issues.apache.org/jira/browse/HBASE-14521
> Project: HBase
> Issue Type: Bug
> Affects Versions: 0.98.14, 1.1.2
> Reporter: Yu Li
> Assignee: Yu Li
> Fix For: 2.0.0, 1.3.0
>
> Attachments: HBASE-14521.patch, HBASE-14521_v2.patch,
> HBASE-14521_v3.patch
>
>
> From name of the _hbase.client.retries.number_ property, it should be the
> number of maximum *retries*, or say if we set the property to 1, there should
> be 2 attempts in total. However, there're two different semantics when using
> it in current code base.
> For example, in ConnectionImplementation#locateRegionInMeta:
> {code}
> int localNumRetries = (retry ? numTries : 1);
> for (int tries = 0; true; tries++) {
> if (tries >= localNumRetries) {
> throw new NoServerForRegionException("Unable to find region for "
> + Bytes.toStringBinary(row) + " in " + tableName +
> " after " + numTries + " tries.");
> }
> {code}
> the retries number is regarded as max times for *tries*
> While in RpcRetryingCallerImpl#callWithRetries:
> {code}
> for (int tries = 0;; tries++) {
> long expectedSleep;
> try {
> callable.prepare(tries != 0); // if called with false, check table
> status on ZK
> interceptor.intercept(context.prepare(callable, tries));
> return callable.call(getRemainingTime(callTimeout));
> } catch (PreemptiveFastFailException e) {
> throw e;
> } catch (Throwable t) {
> ...
> if (tries >= retries - 1) {
> throw new RetriesExhaustedException(tries, exceptions);
> }
> {code}
> it's regarded as exactly for *REtry* (try a call first with no condition and
> then check whether to retry or exceeds maximum retry number)
> This inconsistency will cause misunderstanding in usage, such as one of our
> customer set the property to zero expecting one single call but finally
> received NoServerForRegionException.
> We should unify the semantic of the property, and I suggest to keep the
> original one for retry rather than total tries.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
