Bug #1018
1.50 r1605 out fds, very high max-fds, lock up
| Status: | Fixed | Start: | ||
| Priority: | Normal | Due date: | ||
| Assigned to: | - | % Done: | 0% |
|
| Category: | core | |||
| Target version: | - | |||
| Pending: | Resolution: | fixed |
||
Description
Hi,
I am running 1.50r1605 on Debian 3.1 (kernel 2.6.8 k7) and experiencing
out of fds errors and lockups by lighty. The instance serves large file
downloads, and handles around 150 conncurrent connections (according to
server-status) pushing around 150Mbps. The lockups occur after several
hours with the very high max-fd's limit. When the limit was lower (1024)
it would lockup very quickly.
server.c.803: (trace) note servers sockets got disabled for a while as
we are out of fds. We use 14737 of 16374 right now. (you may want to
raise server.max-fds)
Here is the pertinent config settings:
server.modules = (
"mod_access",
"mod_alias",
"mod_accesslog",
"mod_auth",
"mod_status"
)
connection.kbytes-per-second = 640
server.network-backend = "linux-aio-sendfile"
server.use-noatime = "enable"
server.max-stat-threads = 1
server.max-read-threads = 64
server.max-connections = 5461
server.max-fds = 16384
server.event-handler = "linux-sysepoll"
I am not sure if this is a bug with lighttpd or some config setting on
my server that needs to be tweaked. If you need more information from me
to track it down, let me know what I need to get for you.
I changed the max-fds and max-connections to their defaults and restarted lighttpd attaching the strace to record what happens. The server reported the out of fds error after only 5 minutes under the same load. I will attach the output of the strace to this ticket.
History
02/09/2007 05:32 PM - complexero
The strace file was too large to attach, I have made it available for download here:
http://complexero.com/lighttpd/lighttpd.trace.gz
Let me know if you need any other information.
Thanks
Mike