hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From James Moore <jcmo...@hubspot.com>
Subject Re: Pre-emptive Fast Fail not used for scanners?
Date Thu, 02 Mar 2017 17:57:29 GMT
I've created HBASE-17719
<https://issues.apache.org/jira/browse/HBASE-17719> with
an attached patch file that allows scanners to fast fail.

On Thu, Mar 2, 2017 at 10:56 AM, Ted Yu <yuzhihong@gmail.com> wrote:

> James:
> Can you log a JIRA (and attach patch since you have made some fix) ?
>
> Thanks
>
> On Tue, Feb 28, 2017 at 3:07 PM, James Moore <jcmoore@hubspot.com> wrote:
>
> > Hey Ted,
> >
> > Unfortunately, that change triggers NPEs in the client.  I tracked down
> the
> > source of the NPE as the ScannerCallableWithReplicas not extending
> > RegionServerCallable as ScannerCallable does.  I'm running some
> additional
> > tests now but should have an update shortly.
> >
> > On Tue, Feb 28, 2017 at 2:29 PM, Ted Yu <yuzhihong@gmail.com> wrote:
> >
> > > Seems to be a bug.
> > >
> > > Mind trying out this change ?
> > >
> > > http://pastebin.com/qim48Se9
> > >
> > > Thanks
> > >
> > > On Tue, Feb 28, 2017 at 11:06 AM, James Moore <jcmoore@hubspot.com>
> > wrote:
> > >
> > > > The value is true, and I've been able to verify FastFail behavior
> when
> > > > using Gets when I added a few logging statements into the
> > > RPCRetryingCaller
> > > > I noticed that on scans the interceptor for scans only was
> > > > NoOpRetryableCallerInterceptor
> > > > rather than a PreemptiveFastFailInterceptor.
> > > >
> > > > the basic testing code is
> > > >
> > > > Scan s =  new Scan();
> > > > > s.addColumn(Bytes.toBytes("0"),Bytes.toBytes("cq1"));
> > > > > //multi.add(s);
> > > > > ResultScanner scanner = table.getScanner(s);
> > > > > scanner.next();
> > > > > scanner.close();
> > > > >
> > > > >
> > > > On Tue, Feb 28, 2017 at 12:41 PM, Ted Yu <yuzhihong@gmail.com>
> wrote:
> > > >
> > > > > What's the value for "hbase.client.fast.fail.mode.enabled" ?
> > > > >
> > > > > The default value is false.
> > > > >
> > > > > Cheers
> > > > >
> > > > > On Tue, Feb 28, 2017 at 9:36 AM, James Moore <jcmoore@hubspot.com>
> > > > wrote:
> > > > >
> > > > > > We're considering rolling out Pre-Emptive Fast fail to replace
a
> > > > > homegrown
> > > > > > solution based on Hystrix for fast failing requests. However
in
> > > > testing,
> > > > > it
> > > > > > appears that scanners use the NoOpRetryableCallerInterceptor
> rather
> > > > than
> > > > > > the configured fast fail interceptor.  Is it intentional that
> > Scanner
> > > > > calls
> > > > > > don't use FastFail?
> > > > > >
> > > > > > thanks!
> > > > > >
> > > > > > --James
> > > > > >
> > > > >
> > > >
> > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message