Re: deadlock in pcas 3.14.12.6-rc1

[Bruce Hill <bhill@slac.stanford.edu>]

Hi Xiaoqiang,

I hesitate to say for sure as I'm not completely familiar w/ this code,
but as far as I can tell the casChannelI instance would have to be constructed
before creating a casMonitor instance and installing it for the eventSys to process.

It also appears the casMonitor would have to be uninstalled and deleted before
the casChannelI instance could be deleted.

As postEvent is processing events for casMonitors, that should prevent this deadlock
from occurring again.

Regards,
- Bruce

On 12/07/2016 03:08 AM, Wang Xiaoqiang wrote:
> Hi Bruce,
>
> Thanks. The crash cannot be reproduced with this fix.
>
> Since the casPVI lock is now acquire during channel creation,
> just a purely theoretical thinking, is there a chance that the deadlock would happen if a new CA client wants to 
> connect while PCAS application is calling postEvent?
>
> I am running the same test again, and in addition four loops to make continuous caget requests.
> $ while [ 1 ]; do caget MTEST:MinValue; done;
>
>
> Best
> Xiaoqiang
>
> On 12/07/2016 10:45 AM, Bruce Hill wrote:
> > Hi Xiaoqiang,
> > Thanks, I was able to reproduce the bug using your suggestions.
> > I've submitted a proposed fix on launchpad.   Please let me know if you
> > have a chance to try it.    My version ran all day today w/o crashing.
> > https://code.launchpad.net/~bhill/epics-base/pcas-deadlock-fix-3.14/+merge/312638
> > <https://code.launchpad.net/%7Ebhill/epics-base/pcas-deadlock-fix-3.14/+merge/312638>
> >
> >
> > Cheers,
> > - Bruce
> >
> >
> > On 12/6/2016 1:38 AM, Wang Xiaoqiang (PSI) wrote:
> > > Hi Bruce,
> > >
> > > I replied you here
> > > https://github.com/paulscherrerinstitute/pcaspy/issues/29#issuecomment-265103000
> > >
> > >
> > > I carried out the described test procedure on a SL6 64bit Linux PC.
> > > The problem appears definitely in a few seconds.
> > >
> > > Best
> > > Xiaoqiang
> > >
> > > -----Original Message-----
> > > From: tech-talk-bounces@aps.anl.gov
> > > [mailto:tech-talk-bounces@aps.anl.gov] On Behalf Of Bruce Hill
> > > Sent: Dienstag, 6. Dezember 2016 04:15
> > > To: tech-talk@aps.anl.gov
> > > Subject: Re: deadlock in pcas 3.14.12.6-rc1
> > >
> > > Hi Xiaoqiang,
> > > Do you have a pcaspy example that I can run to reproduce this?
> > >
> > > I'm running pcaspy 0.6.2 currently, but could also build from a github
> > > branch if available.
> > >
> > > Thanks,
> > > - Bruce
> > >
> > >
> > > On 11/10/2016 08:59 AM, Wang Xiaoqiang (PSI) wrote:
> > > > Hi,
> > > >
> > > > I am testing the support of dynamic array in the recent base
> > > > 3.14.12.6-rc1.
> > > > Using a normal pcaspy application, which updates ~15 Hz. It
> > > > repeatedly gets into a deadlock state.
> > > >
> > > > The analysis has been put up here
> > > > https://github.com/paulscherrerinstitute/pcaspy/issues/29
> > > >
> > > > The solution could be reversing the lock order during
> > > > casEventSys::process. But it does not seem to be easy, because the
> > > > chain of calls involved.
> > > >
> > > > Best
> > > > Xiaoqiang
> > > --
> > > Bruce Hill
> > > Member Technical Staff
> > > SLAC National Accelerator Lab
> > > 2575 Sand Hill Road M/S 10
> > > Menlo Park, CA  94025

--
Bruce Hill
Member Technical Staff
SLAC National Accelerator Lab
2575 Sand Hill Road M/S 10
Menlo Park, CA  94025