asterixdb-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ian Maxon <ima...@uci.edu>
Subject Re: Issue 884 in asterixdb: Unexpected exceptions when the FrameSize and the PageSize is different
Date Mon, 18 May 2015 21:50:46 GMT
I run across that situation (#2) rather often as well in testing. Since
it's an OOM situation I think the only safe thing to do would be to somehow
ask a job to abort (assuming that doesn't require any more pages?...). I'm
not sure how the buffer cache would communicate that information up to the
correct layer though.

- Ian

On Mon, May 18, 2015 at 2:35 PM, Mike Carey <dtabass@gmail.com> wrote:

>  Regarding 2, this should not happen in production, obviously, but it
> would be nice to catch it when it happens (needing a victim, can't find
> one) and throw up our hands rather than hanging.
>
> On 5/18/15 10:02 AM, asterixdb@googlecode.com wrote:
>
>
> Comment #3 on issue 884 by kiss...@gmail.com: Unexpected exceptions when
> the FrameSize and the PageSize is different
> https://code.google.com/p/asterixdb/issues/detail?id=884
>
> I investigated this issue and found why this situation occurs.
> There are two problems behind this situation.
>
> 1. When an in-memory component is about to be scheduled to be flushed,
> the component's writer counter can be greater than 0 based on the current
> code base/design.
> So, this should be considered as a valid situation (rather than an
> exception) and
> dealt with appropriately, i.e., the component must not be scheduled to be
> flushed.
>
> 2. The number of buffer pages are not enough if the frame size is set to
> 4096 and the
> buffer cache page size is set to 327680 in default
> asterix-build-configuration.xml.
> If the configuration file is set in this way, there are at most 11 pages
> in buffer cache,
> so buffer cache can not find victim pages since all pages are used. Thus,
> the test
> is hung.
> So, the buffer cache size (not the buffer cache page size) should be
> increased
> appropriately to avoid this situation.
>
> I have a fix for problem 1 and I will send a code review after I add
> comments to the fixed
> code such as why the situation can occur with a detail example scenario.
>
> Lastly, this issue is different from the situation described in issue 878
> even though both issue came from LSMHarness layer.
>
>
>  --
> You received this message because you are subscribed to the Google Groups
> "asterixdb-dev" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to asterixdb-dev+unsubscribe@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.
>

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