Experimental Physics and
Industrial Control System

----- Original Message -----

From: Hill, Jeffrey O

To: Makarov, Oleg A. ; [email protected]

Sent: Tuesday, February 07, 2012 1:40 PM

Subject: RE: Virtual circuit disconnect - caRepeater bug?

Hello Oleg,

 

Does a simple client such as camonitor experience any disconnects at the same time that MEDM does? I am guessing probably not. Have we seen this symptom with any client other than MEDM?

 

The message that you are seeing indicates that the OS has informed the CA client library that the TCP circuit dropped. The channels can become disconnected in the MEDM application for two reasons; the client library might detect that the TCP circuit is unresponsive or in contrast it might be informed by the OS that the TCP circuit has disconnected. It appears that you are experiencing the latter situation.

 

Does the circuit called out in the diagnostic in question disconnect only briefly, or do any channels connected to PVs on that IOC remain in a disconnected state?

 

There is a “feature” in the CA client library where it will close down the TCP circuit if there are no-longer any channels which use it. One possibility might be that a new display replaces the old display and we are seeing a timing induced false positive disconnect message when all channels on the IOC mentioned in the diagnostic are deleted. Do you know how many channels remain connected to the IOC in question before and after the related display button is pressed?

 

Ø  MEDM build with Epics 3.14.12.2 does not experience such "Virtual circuit disconnect"

Ø  problem while running on a workstation with Epics 3.14.11 installed.

 

Presumably this is occurring on Linux, with an EPICS base build specifying shareable libraries (runtime linking) for EPICS base. I seem to recall that the currently the paths to the sharable libraries are specified when MEDM is linked so this is perhaps an unexpected behavior, but perhaps that part of the build system has changed.

 

Jeff

 

 

From: Oleg Makarov [mailto:[email protected]]
Sent: Monday, February 06, 2012 7:13 PM
To: [email protected]; Hill, Jeffrey O
Subject: Virtual circuit disconnect - caRepeater bug?

 

Is a caRepeater bug was introduced starting from Epics 3.14.12?

It is intermittent, and may occur when "Related Display" button on an medm screen is pressed.

We were able to reproduce "Virtual circuit disconnect" message (see below) with Epics 3.14.12, Epics 3.14.12.1, and Epics 3.14.12.2 but not with Epics 3.14.11.

Other observations:

1) Workstations with Epics 3.14.11 do not experience such "Virtual circuit disconnect" problem with the same IOC, so this bug is likely on a client side.

2) MEDM build with Epics 3.14.12.2 does not experience such "Virtual circuit disconnect" problem while running on a workstation with Epics 3.14.11 installed.

Is this bug in the caRepeater code?

Any suggestions?

 

Regards,

Oleg Makarov

 

 

 

Mon Feb 06 18:19:30 CST 2012

medmCAExceptionHandlerCb: Channel Access Exception:

Channel Name: Unavailable

Native Type: Unavailable

Native Count: 0

Access: Unavailable

IOC: Unavailable

Message: Virtual circuit disconnect

Context: bl3ioc1.gmca.aps.anl.gov:5064

Requested Type: TYPENOTCONN

Requested Count: 0

Source File: ../cac.cpp

Line number: 1214