[cups.bugs] [MOD] STR #3257: cups-polld wedged at 100% processor usage

[Maciej Bliziński]

DO NOT REPLY TO THIS MESSAGE.  INSTEAD, POST ANY RESPONSES TO THE LINK BELOW.

[STR New]

The problem has been observed on Solaris. Not sure whether it's OS related
or not. 

There's a problem in which cups-polld (v1.3.9) daemons start consuming
100% of 1 CPU each. The issue occurs after certain time from the startup,
within the range from few hours to few days. Individual cups-polld
processes get into that state one by one, and then stay in that state
indefinitely. Their behavior resembles being stuck in busy-waiting mode.

- truss doesn't show anything (perhaps the process doesn't make any
syscalls)
- pstack shows that the process is spinning around httpUpdate() and
httpGets() inside cupsDoIORequest.

Example pstacks:

 ff365090 httpGets (ffbedf18, 8000, 48098, 48098, 1c00, 2400) + 4
 ff365ec4 httpUpdate (48098, 2e604, 5, ff394450, 8000, ffbedf18) + 88
 ff378b0c cupsDoIORequest (3, 458a8, 0, ffffffff, ffffffff, 48098) + 3e0
 00011714 poll_server (48098, 4, 277, 384, 458a8, 2203c) + ec
 00011550 main     (4, ffbffb84, 277, 384, 2203c, 22370) + 268
 00010eb8 _start   (0, 0, 0, 0, 0, 0) + 108

 ff23b750 _gnutls_recv_int (45e48, 17, ffffffff, 0, 800, 2400) + c10
 ff36514c httpGets (49cd8, 8000, ffbedf18, 48098, 1c00, 2400) + c0
 ff365ec4 httpUpdate (48098, 2e604, 5, ff394450, 8000, ffbedf18) + 88
 ff378b0c cupsDoIORequest (3, 458a8, 0, ffffffff, ffffffff, 48098) + 3e0
 00011714 poll_server (48098, 4, 277, 384, 458a8, 2203c) + ec
 00011550 main     (4, ffbffb84, 277, 384, 2203c, 22370) + 268
 00010eb8 _start   (0, 0, 0, 0, 0, 0) + 108

 ff365e88 httpUpdate (48098, 2e604, 5, ff394450, 8000, ffbedf18) + 4c
 ff378b0c cupsDoIORequest (3, 458a8, 0, ffffffff, ffffffff, 48098) + 3e0
 00011714 poll_server (48098, 4, 277, 384, 458a8, 2203c) + ec
 00011550 main     (4, ffbffb84, 277, 384, 2203c, 22370) + 268
 00010eb8 _start   (0, 0, 0, 0, 0, 0) + 108



The issue can be replicated consistently on many hosts. The operating
system is Solaris 10 (updates 5 and 6). The configuration is referring to
10 other cups servers to poll.

I'm still looking into ways of collecting more debugging data, but I've
decided to file the bug right now, perhaps someone would have ideas where
to look or what might be a potential cause of the problem.

Link: http://www.cups.org/str.php?L3257
Version: 1.3.10