Bug 1467395

Reporting Customer: Motorola
TAM Account: YES

Description of problem:
winbind seems to be leaving sockets open and eventually hitting clients limit, ending up in dumping a lot of text in the log files, affecting machine performance to the point that it might not responds to the console anymore.

Here's the problem as described by the customer:

====

I understand that actual root cause is that (for unknown reason) winbind __does not close sockets__
Here what scenario looks like (on actual production system, as mentioned before we are not able to reproduce it):

At some point in time following messages appears (this looks like random, no apparent issues are reported by OS before):
May  1 11:03:27 zc01.zone1 winbindd[5556]: [2017/05/01 11:03:27.380063,  0] winbindd/winbindd.c:975(winbindd_listen_fde_handler)
May  1 11:03:27 zc01.zone1 winbindd[5556]:   winbindd: Exceeding 200 client connections, no idle connection found
May  1 11:03:27 zc01.zone1 winbindd[5556]: [2017/05/01 11:03:27.388117,  0] winbindd/winbindd.c:975(winbindd_listen_fde_handler)
May  1 11:03:27 zc01.zone1 winbindd[5556]:   winbindd: Exceeding 200 client connections, no idle connection found
May  1 11:03:32 zc01.zone1 winbindd[5556]: [2017/05/01 11:03:32.119143,  0] winbindd/winbindd.c:975(winbindd_listen_fde_handler)
May  1 11:03:32 zc01.zone1 winbindd[5556]:   winbindd: Exceeding 200 client connections, no idle connection found
May  1 11:03:32 zc01.zone1 winbindd[5556]: [2017/05/01 11:03:32.119528,  0] winbindd/winbindd.c:975(winbindd_listen_fde_handler)
May  1 11:03:32 zc01.zone1 winbindd[5556]:   winbindd: Exceeding 200 client connections, no idle connection found
(...)
Winbind will stay at this state for some time, but after an hour or so process runs out of descriptors and message flood begins (this means like 10000+ entries/s), with following logs:

May  1 12:19:31 zc01.zone1 winbindd[5556]:   winbindd: Exceeding 200 client connections, no idle connection found
May  1 12:19:32 zc01.zone1 winbindd[5556]: [2017/05/01 12:19:32.468367,  0] winbindd/winbindd.c:975(winbindd_listen_fde_handler)
May  1 12:19:32 zc01.zone1 winbindd[5556]:   winbindd: Exceeding 200 client connections, no idle connection found
May  1 12:19:32 zc01.zone1 winbindd[5556]: [2017/05/01 12:19:32.468437,  0] winbindd/winbindd.c:817(new_connection)
May  1 12:19:32 zc01.zone1 winbindd[5556]:   Faild to accept socket - Too many open files
May  1 12:19:32 zc01.zone1 winbindd[5556]: [2017/05/01 12:19:32.468608,  0] winbindd/winbindd.c:975(winbindd_listen_fde_handler)
May  1 12:19:32 zc01.zone1 winbindd[5556]:   winbindd: Exceeding 200 client connections, no idle connection found
May  1 12:19:32 zc01.zone1 winbindd[5556]: [2017/05/01 12:19:32.468630,  0] winbindd/winbindd.c:817(new_connection)
May  1 12:19:32 zc01.zone1 winbindd[5556]:   Faild to accept socket - Too many open files
May  1 12:19:32 zc01.zone1 winbindd[5556]: [2017/05/01 12:19:32.468685,  0] winbindd/winbindd.c:975(winbindd_listen_fde_handler)
May  1 12:19:32 zc01.zone1 winbindd[5556]:   winbindd: Exceeding 200 client connections, no idle connection found
May  1 12:19:32 zc01.zone1 winbindd[5556]: [2017/05/01 12:19:32.468704,  0] winbindd/winbindd.c:817(new_connection)
May  1 12:19:32 zc01.zone1 winbindd[5556]:   Faild to accept socket - Too many open files

This of course provides huge impact on I/O, /var/log gets full in the matter of minutes and process itself eating up CPU making server so busy it's impossible to login on console or ssh. To recover the server you need to power-cycle & do some clean up on /var/log.

We would expect that "winbind request timeout" is backported to samba-winbind-3.6 as per https://attachments.samba.org/attachment.cgi?id=10128
 and https://bugzilla.samba.org/show_bug.cgi?id=3204

===


While the customer understands that RHEL6 is late into its production cycle and a backport might not be possible, they expect us to at least provide them with a supported workaround that inhibits the spam in the logs and therefore protects the machine from being impacted.