[Date Prev][Date Next]
[Chronological]
[Thread]
[Top]
Re: slapd goes <DEFUNCT> after random time (ITS#248)
"Kurt D. Zeilenga" wrote:
> Under UNIX semantics, a process is considered defunct if it has exited
> and has a parent which has not waited for it. This is often referred
> >For that matter, I can't even figure out what "DEFUNCT" means!
> man ps
Grrrr... I thought it reported them as "Zombie". Silly me. I assure
you I am not that much of an idiot. :) I /know/ some tool calls them
zombies...
> You should look at the server log files for hints to why it shutdown.
> Usually the entries associated with the last connection are enough.
slapd[30042]: conn=80 fd=7 connection from huey.hidden.com
(10.0.0.100) accepted.
slapd[1984]: conn=80 op=0 BIND dn="" method=128
slapd[1984]: conn=80 op=0 RESULT err=0 tag=97 nentries=0
slapd[1985]: conn=80 op=1 SRCH base="O=HIDDEN" scope=2
filter="(|(cn=KRE*)(mail=KRE*)(sn=KRE*))"
slapd[30042]: conn=81 fd=12 connection from cogsworth.hidden.com
(10.0.0.107) accepted.
slapd[1986]: conn=81 op=0 BIND dn="" method=128
slapd[1986]: conn=81 op=0 RESULT err=0 tag=97 nentries=0
slapd[1987]: conn=81 op=1 SRCH base="O=HIDDEN" scope=2
filter="(|(cn=CHOE*)(mail=CHOE*)(sn=CHOE*))"
<hang> (I wasn't here during this hang, so I can't tell you /which/
process went defunct. Sorry. I'll have that for you next time this
happens. This was an access via Netscape Messenger.) I fear not much
help, here. I can turn up the logging and try again.
> If it's aborting due to a memory fault or such, I recommend configurating
> your environment so that slapd can dump core. If you do have a core
> dump, a traceback would be useful (in conjunction with the log)
> in diagnosing the problem.
It isn't aborting. It is allowed to dump core, but doesn't appear to
be dumping one. It just silently dies...
> >My current working theory is that my problems are somehow connected to SYN
> >COOKIES.
> I think you are on a wild goose chase. Your problem is much
> more likely something quite simple.
I like simple. However, this one appears to be /too/ simple. It is
NOT leaving core files, it is NOT leaving death messages in the logs,
and it does NOT recover.
I will turn up logging...