Package: slapd
Version: 2.2.23-1
Hello,
slapd/slapadd use sched_yield on places where it really
should not be idling. Due to this upgrade from 2.1.30 to
2.2.23 takes more than 1 day on our 100000 objects tree
we have, with about 40 indexed attributes (after one day
I gave up, attached debugger to slapadd and overwrite
sched_yield procedure in slapadd with 'ret'; import
finished ~1 hour after I did that).
So I created shared library with function below, and
LD_PRELOADed it before upgrading replicas - replicas
were upgraded in about 1h15m, so I assume that slapadd
without killing sched_yield would take at least 4 days
with our configuration, which seems a bit too long to me.
It would be nice if upstream could eliminate at least
some useless sched_yield from code (if not all these
calls, I do not believe they are doing anything good
on 2.6.x kernels). Or you should warn users that
slapadd can take days, maybe weeks.
/*
%.so: %.c
gcc -W -Wall -O2 -shared -o $@ $< -ldl
*/
#define _GNU_SOURCE
#include <dlfcn.h>
int sched_yield(void) {
return 0;
}
Thanks,
Petr Vandrovec
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
