murder idle hanging

[Andrew Morgan]

I'm running a cyrus murder with 2 frontends, 2 backends, and 1 mupdate 
master, all v2.2.13.  Occasionally, frontend processes seem to hang while 
handling IDLE commands from clients.  They all show the following in 
netstat's output:

tcp       54      0 128.193.4.143:54538     128.193.4.142:143       CLOSE_WAIT

(54 bytes sitting in the Recv-Q from the backend).


gdb shows the following stack:

(gdb) where
#0  0xb7c3d9f8 in select () from /lib/tls/libc.so.6
#1  0x08076c7c in prot_select (readstreams=0x8126668, extra_read_fd=-1, 
out=0xbfa11f84,
     extra_read_flag=0x0, timeout=0xbfa11f88) at prot.c:1093
#2  0x080537d8 in cmd_idle (tag=0x8145398 "nhqt") at proxyd.c:2678
#3  0x08050dbe in cmdloop () at proxyd.c:1701
#4  0x0804f78c in service_main (argc=2, argv=0x811d008, envp=0xbfa153f0) 
at proxyd.c:1306
#5  0x0804c380 in main (argc=2, argv=0x3a805a, envp=0xbfa153f0) at 
service.c:533


and strace of this process repeats the following:

open("/var/spool/cyrus/config/msg/shutdown", O_RDONLY) = -1 ENOENT (No such file or directory)
time(NULL)                              = 1161625450
select(1, [0], NULL, NULL, {60, 0})     = 0 (Timeout)
time(NULL)                              = 1161625510


There seems to be a bug in the handling of IMAP IDLE commands in proxyd. 
I found bug #2762 in the Cyrus bugzilla which fixed a race condition with 
IDLE for imapd, but it did not include changes for proxyd.  I don't know 
if this bug is relevant here in any case.

Has anyone seen this before?

 	Andy