Re: IOC hangs (scanOnce crashes)

[lange@leo.acc.bessy.de (Ralph Lange)]

> Request are made to scanOnce by putting a request into a ring buffer.
> thus when scanOnce crashes the ring buffer ultimately fills up. This is
> what you are seeing.

Unwillingly.

> Note that NO epics tasks are automatically restarted when they crash.
> Sounds nice but can cause more problems than it solves.

The periodic scan tasks ARE automatically restarted. (See line 539 +-20 of
dbScan.c) Looks safe to me in that case. Generally I would agree with you.

> >         We inserted a patch in scanOnce to check the precord argument for
> [...]
> 
> So what did you find?

Due to oh-so-important measuring tasks we are not able to run further tests
on this IOC right now. During the measurements the IOC load is next to zero
- almost no devices are touched and the fault doesn't show up. Seems we'll
have to wait until the physicists are asleep.

> If you get a stack trace of crashed scanOnce task it should help.

We're trying to catch the task that is calling scanOnce() to see where the
crap comes from. I think the onceTask itself doesn't do too much
interesting things - it just reads from the ring buffer and calls
dbScanLock without checking the validity of the precord it got from the
buffer - then dies.

At least it's good to know that you're all out there ... ;-)
Ralph
-- 
      __  Ralph Lange                         Email:       lange@bii.bessy.de 
     /\ \                                     WWW: http://www.bessy.de/~lange
    /  \ \  BESSY II                          
   / /\ \ \  Berliner Elektronenspeicherring- Snail:                 BESSY II
  / / /\ \ \  Gesellschaft fuer Synchrotron-               Rudower Chaussee 5
 / / /__\_\ \  strahlung m.b.H.                       D-12489 Berlin, Germany
/ / /________\                                Phone:         +49 30 6392-4862
\/___________/ Control System Group           Fax:                ...   -4859