manifoldcf-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Karl Wright <daddy...@gmail.com>
Subject Re: LockGate
Date Fri, 19 Dec 2014 07:34:05 GMT
I looked carefully at the code and can find nothing that would cause
functional problems.  I did, however, notice that LockGate itself is
crippled by a bug which essentially expires LockGate objects too
aggressively.  See CONNECTORS-1126.  I would be curious if this fix has any
effect on what you are seeing.

Karl


On Thu, Dec 18, 2014 at 7:27 PM, Karl Wright <daddywri@gmail.com> wrote:
>
> Hi Aeham,
>
> I honestly can't remember this issue even occurring in the first place,
> even though the code was written just a year ago.
>
> Normally for something like this I would have opened a ticket, but I can't
> find any.  That argues that there was a logic error that I found and fixed
> during the original development.  I certainly do not remember having anyone
> raise this issue at all in the past year.
>
> If you have a lock up, it's never wrong to get a thread dump and send it
> along to me.  The reason is that lockups often have causes that have
> nothing to do with lock management, but are the result of escalating
> carnage from one stuck thread, or from a deadlock.  It's also helpful to
> know when you first started to see the failure.  If this is right after the
> hashed lock names fix I just committed, then I would have concerns that
> that fix was in fact causative.
>
> Karl
>
>
> On Thu, Dec 18, 2014 at 7:00 PM, Aeham Abushwashi <
> aeham.abushwashi@exonar.com> wrote:
>>
>> Hi,
>>
>> While investigating a couple of issues where manifold seemingly locks up
>> for no obvious reason, I stumbled on the comments at the top of the
>> LockGate class which discuss some unexplained lockups.
>> Are the comments relevant to the latest versions of MCF? If so, is there a
>> JIRA ticket I could follow for more details?
>>
>> I was particularly intrigued because the pattern of the stack traces in
>> the
>> comment is similar to the one I'm investigating.
>>
>> Cheers,
>> Aeham
>>
>

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