My question to alpha-osf-managers is at the end. Sorry about the long
intro, but the background has important information.
I posted this recently to amanda-users:
I'm struggling along to get amanda working on my net. I keep getting the
following errors in sendsize.debug:
setpgrp(0,20574) failed: Invalid argument
sendsize-dump: running: "dump 0f - /dev/rz2a"
dump: Dumping from host mays.mrj.com
dump: Date of this level 0 dump: Tue Nov 21 16:33:17 1995 EST
dump: Date of last level dump: Dumping /dev/rrz2a (/) to standard output
dump: Mapping (Pass I) [regular files]
dump: Mapping (Pass II) [directories]
dump: Cannot get semaphores
get_semaphore(): semget() for fifo_free_semid: No space left on device
dump: SIGTERM received -- Try rewriting
dump: Unexpected signal -- cannot recover
dump: Cannot remove semaphores
remove_semaphore(): semctl() for fifo_free_semid: Invalid argument
dump: Cannot remove shared memory
remove_shared_memory(): shmctl(): Invalid argument
dump: SIGTERM received -- Try rewriting
dump: Unexpected signal -- cannot recover
When I run ipcs, I see the following:
mays.mrj.com{kmayer}444: ipcs
Message Queues:
T ID KEY MODE OWNER GROUP
q 0 1090529888 --rw------- root wheel
Shared Memory:
T ID KEY MODE OWNER GROUP
m 0 0 --rw------- root daemon
m 5377 0 --rw------- bin daemon
m 258 0 --rw------- bin daemon
m 259 0 --rw------- bin daemon
m 260 0 --rw------- bin daemon
m 261 0 --rw------- bin daemon
Semaphores:
T ID KEY MODE OWNER GROUP
s 0 1090529888 --ra------- root wheel
s 225 0 --ra------- bin daemon
s 226 0 --ra------- bin daemon
s 227 0 --ra------- bin daemon
s 36 0 --ra------- bin daemon
s 37 0 --ra------- bin daemon
s 38 0 --ra------- bin daemon
s 39 0 --ra------- bin daemon
s 40 0 --ra------- bin daemon
s 41 0 --ra------- bin daemon
s 42 0 --ra------- bin daemon
s 43 0 --ra------- bin daemon
s 44 0 --ra------- bin daemon
s 45 0 --ra------- bin daemon
s 46 0 --ra------- bin daemon
s 47 0 --ra------- bin daemon
The shared memory and the semaphores DO NOT go away unless I manually
delete them with ipcrm.
It seems that my OS configuration only allows 16 semaphores. Once the
slots are filled, dump dies (with our without Amanda!) and until I
empty out the cruft, won't run. Has anyone else run into this? Know
what's going on? I'm running OSF/1 3.2, and Amanda 2.2.6.6.
This morning I got a message from Phil Newman:
From newman_at_pandora.gsfc.nasa.govWed Nov 22 11:05:21 1995
Date: Wed, 22 Nov 1995 10:48:46 -0500 (EST)
From: Phil Newman <newman_at_pandora.gsfc.nasa.gov>
To: Ken Mayer <kmayer_at_mrj.com>
Subject: Re: HELP! Getting started with an OSF/1 client
Ken,
To increase the number of semaphores on a OSF/1 machine, edit
the kernel configuration file, /sys/conf/<NODE> where node is
the name of the machine in caps. Scroll down to special options
and put an entry in for semmsl after maxssiz:
# Special options (see configuring the kernel chapter
# in the Guide to System Administration)
#
timezone 0 dst 0
dfldsiz 134217728
maxdsiz 1073741824
dflssiz 2097152
maxssiz 33554432
semmsl 2500 # <<< ADD THIS <<<
cpu "DEC3000_500"
maxusers 32
machine alpha
The choice of 2500 is rather arbitrary. Most machine it
is 200 or 300. It seems that the larger you make this
number, the longer it takes before you run into problems.
OSF/1 is apparently not giving the semmaphores back to
the OS after the process has finished.
Close the file and rebuild the kernel...
... stuff about installing the new kernel ...
Hope this helps!
-Phil
>>> OKAY, HERE'S MY QUESTION <<<
Is this 'not giving semaphores back to the OS after the process has
finished' an OSF/1 bug or an Amanda bug? I'm runing 3.2, is there a patch
fix?
TIA,
Ken
Received on Wed Nov 22 1995 - 18:03:21 NZDT