Kbase P9866: WebSpeed agents become BUSY under heavy load on Solaris
| Autor |
Progress Software Corporation - Progress |
| Acesso |
Público |
| Publicação |
8/27/2003 |
|
Status: Verified
FACT(s) (Environment):
WebSpeed 3.1D
FACT(s) (Environment):
Sun Solaris SPARC
SYMPTOM(s):
WebSpeed agents become BUSY under heavy load
WebSpeed agents remain BUSY and are never released
Following calls are typically in pstack
free
msgSave
msgnCB
drSigMessage
drSigClient
Following calls are typically in pstack
free
malloc
msgSave
msgnCB
__sighndlr
sigacthandler
pstack shows msgnCB and msgSave after sigacthandler being called
** Pipe to subprocess has been broken (140)
Busy agents usually writes multiple 140 errors in the WebSpeed server log
CAUSE:
Bug# 20030203-021
CAUSE:
When web users click on stop button, the communication channel ("pipe" referred in error 140) from agent to messenger is interrupted, causing sigpipe to be sent. In the signal handler, msgSave is called which calls malloc and free, but on Solaris, malloc and free are not Async-Signal-Safe.
FIX:
This issue has been fixed in SP Progress 9.1D06