It seems we finally have the process limit and fd limit bumped up enough to
handle the load, but we have been looking for a linux utility similar to sun's
"pstat" that will count the current running processes, open descriptors, and
memory usage, and print it out. We are currently using "ps ax | wc -l" to get
the process total, but have not yet found a way to count total open files.
Sun's "pstat" looks like this:
~ (/dev/ttyud) dwild@eskimo> pstat -T
1013/4500 files
1951/4000 inodes
445/1500 processes
299860/1165040 swap
~ (/dev/ttyud) dwild@eskimo>
It would be much appreciated if you could point me to something like this, even
if it is alpha code. This way if it's already started we won't need to spend
time re-writing it.
Another problem we are having is with NIS lookups through libc. It appears,
when using the "getpwnam" and "getpwuid" calls, the system is returning
incorrect data on the first hit, but correct data when it reads from the cache
(for NIS records only). This was causing sendmail to bounce messages to users
who exist on the system; when the load on the system was high, mail was being
delivered to the wrong users. There appears to be no pattern to the
misdeliveries, I havn't really looked at the libc code yet but it appears there
is some garbage somewhere that is being cast or wrapped to a random UID. We
have worked around this problem temporarily by keeping a local copy of the
password file on the system.
If any of you could help us out with this, or possibly point me to a linux
developers' list, it would be much appreciated.
Thanks!
-- Joel Boring (aka Derek Wildstar) Eskimo North dwild@eskimo.com, DWildstar@IRC Seattle, WA IRC Admin / Assistant Sysadmin / Tech Support (206) 361-1161