Experimental Physics and
Industrial Control System

"Jeff Hill" <[email protected]> · Thu, 19 Apr 2007 13:18:29 -0600

Kay,

FWIW, I have been reviewing the differences in the source code between
R3.14.8.2 and R3.14.9. Before R3.14.9 there was a design flaw where the
access security callback was directly placing the response in the server's
output protocol stream. That design creates a vulnerability to any one
client that is not reading its input stream for any reason. If so, the
access security system could block in the rsrv callback until that client
does read its input stream, and in the interim any other client that tries
to register with the access security system will block for the access
security system's lock. This isn't a classic deadlock, but it would
definitely exhibit the symptoms you are reporting. A busy server is quite
likely to end up with a client that does not read its input at one time or
another (independent of course of what client library implementation is in
use).

So I believe that you are seeing another manifestation of a problem you have
previously identified, and we have already fixed in R3.14.9.

HTH,

Jeff

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System