Re: odd ASYN behavior

[Mark Rivers via Tech-talk <tech-talk at aps.anl.gov>]

Please also consider setting

    callbackParallelThreads    

to enable multiple concurrent threads to be working on the callback queues. Making the callback queue larger is one way, having the IOC process the callbacks faster is another way to cope with high traffic situations.

That's a good point.  In this case the problem only occurs at iocInit because there are lots of records with SCAN=I/O Intr that get processed once at iocInit.  In this case I think increasing the queue size is fine, it just means that iocInit takes a bit longer to complete.  In the case when the problem happens due to a high callback rate during normal operation then callbackParallelThreads is a better solution.

Mark

---

From: Tech-talk <tech-talk-bounces at aps.anl.gov> on behalf of Ralph Lange via Tech-talk <tech-talk at aps.anl.gov>
Sent: Friday, March 18, 2022 4:54 PM
To: EPICS Tech Talk <tech-talk at aps.anl.gov>
Subject: Re: odd ASYN behavior

 

On Fri, 18 Mar 2022 at 22:11, Mark Rivers via Tech-talk <tech-talk at aps.anl.gov> wrote:

The problem was that the EPICS callback queue was overflowing.  John was getting a lot of these errors:

    callbackRequest: cbLow ring buffer full

By adding this command just before iocInit() the problem was fixed:

    callbackSetQueueSize(10000)

This is normally not needed if the IOC is controlling a single TetrAMM.   However, John is controlling several TetrAMM devices from the same IOC, and that led to more callbacks than the default queue size could handle.

Please also consider setting

    callbackParallelThreads    

to enable multiple concurrent threads to be working on the callback queues. Making the callback queue larger is one way, having the IOC process the callbacks faster is another way to cope with high traffic situations.

See the AppDevGuide for more details [1].

There are iocShell commands to show statistics on queue usage to allow for checking as well as proper and safe queue size configuration. The iocStats module [2] can be used for continuous monitoring of queue status.

Cheers,

~Ralph

[1] https://epics.anl.gov/base/R3-16/1-docs/AppDevGuide/EPICSGeneralPurposeTasks.html#x17-69800016.2.5

[2] https://github.com/epics-modules/iocStats