maximum number of FD events (64) received

[JINMEI Tatuya / 神明達哉]

At Mon, 27 Sep 2010 13:27:01 +0400,
Samer Khattab <skhattab at gmail.com> wrote:

> I'm using Bind as a caching name server and serving around 2000 req per
> second, and recently have the following messages showing up from time to
> time in the general.log.
> 
> 27-Sep-2010 10:45:47.639 sockmgr 0x2ad7af2f5010: maximum number of FD events
> (64) received
> 27-Sep-2010 10:45:47.872 sockmgr 0x2ad7af2f5010: maximum number of FD events
> (64) received
> 
> BIND BIND 9.7.1-P2
> RHEL 5.5       kernel 2.6.18-194.11.3.el5
> 
> What is the meaning of these messages ? Are they related to the system file
> descriptors ?

These logs are not (directly) related to file descriptors.  They mean
epoll returned more socket events than the implementation normally
expects (which is 64).  This is not necessarily an error because the
remaining events will be returned with the next call to epoll_wait().
However, the event loop should generally runs pretty quickly, so it's
still an unexpected situation.

You may want to check overall stability of the server, e.g., in terms
of the ratio of server failures (SERVFAIL) that your server returns to
the clients, cache memory footprint, cache hit ratio, number of query
drops (if any), etc.   If these are okay and you only see the log
messages occasionally, you can probably ignore them.

Otherwise, if you use multiple threads on a multi-core machine and you
set max-cache-size to some finite value, you may be hit by a recently
found bug in the cache memory management, which can make a caching
server very busy.  (but it's a wild guess: I've personally never seen
this bug trigger the log message in question).  This bug will be fixed
in 9.7.2.

---
JINMEI, Tatuya
Internet Systems Consortium, Inc.