Experimental Physics and
Industrial Control System

"Jeff Hill" <[email protected]> · Mon, 27 Aug 2001 15:25:28 -0600

All,

This is a bug report for a problem discovered in EPICS R3.14,
EPICS R3.13, and possibly also EPICS R3.12.

Symptoms:

Clients disconnect unexpectedly when canceling event (monitor)
subscriptions using ca_clear_event(), but this may not be
noticed because the problem is probably rare under low or 
moderate load, and because CA clients automatically 
reconnect. The problem was only reproducible here under heavy 
load presented when the R3.14 client side regression tests were 
run on a fast Linux workstation. For example, when the tests 
were run on a 500 MHz NT host I was unable to reproduce the bug.

On vxWorks you may see threads with the name "CA client" and also 
"CA event" even if there are no clients connected (the casr 
diagnostic will also incorrectly report that clients are connected). 
As the IOC stays in long term operation without a reboot, and 
depending on the frequency of occurrence at your site (if any), you 
may also notice that memory and file descriptors are not reclaimed.

Fix:

I have committed a patch to CVS, and it will be incorporated
into the next release of R3.13 and also R3.14. 

Background:

The problem was discovered when regression testing EPICS R3.14.
The cause was a deadlock occurring when a CA event (monitor)
subscription was canceled when the CA circuit was simultaneously
in event (monitor) flow control mode and the event (monitor) 
subscription also had a corresponding update in the queue. 

Sorry about any incontinence this has caused.

Jeff
__________________________________________________________
Jeff Hill                     Internet     [email protected]
LANL MS H820                  Voice        505 665 1831
Los Alamos NM 87545 USA       FAX          505 665 5107

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System