Date: Thu, 18 May 1995 12:31:25 -0400 (EDT) From: Mark Hittinger <bugs@ns1.win.net> To: current@FreeBSD.org, hackers@FreeBSD.org Subject: re: Help! (in, freezups) Message-ID: <199505181631.MAA19578@ns1.win.net>
next in thread | raw e-mail | index | archive | help
> From: "Alok K. Dhir" <adhir@iagi.net>
> I've got a news server (INN1.4sec - latest port on freebsd.cdrom.com) 
> running -current (as of yesterday).  For the last 2.5 weeks or so, I've 
> been SUPing just about every other day...  Why?  Because I am waiting for 
> a kernel that doesn't crash every day.  Before this, I was running 
> -current as of mid April (roughly), and did not have this crashing 
> problem.  I can't seem to pinpoint exactly what went wrong.
I'm not seeing terrible problems but I'm not using MMAP yet :-)
Make sure you have this patch installed in your INN:
I forget where it came from - may have been direct from rsalz) and
patched it on to INN 1.4 here.  The relevant area, as kluged, now reads
(around line 82 in site.c):
    if (AmRoot)
        xchown(name);
    if (cp) {
if (cp->fd >= 0)
syslog(L_ERROR, "DEBUG ERROR SITEspool trashed:%d %s:%d", cp->fd, sp->Name, i);
        WCHANremove(cp);
        RCHANremove(cp);
        SCHANremove(cp);
        close(cp->fd);
        cp->fd = i;
        return TRUE;
    }
    sp->Channel = CHANcreate(i, CTfile, CSwriting, SITEreader, SITEwritedone);
    if (sp->Channel == NULL) {
-----------------------
That fixes a file descriptor leak in INN.
I think you should probably re-compile with MMAP off if it is on, and put
this patch in.  Then re-install the re-linked images.
Just out of paranoia I'd run fsck on your news partition, rebuild your
history files, and do an expireover -s if you are using the NOV database.
Depending on your load you might also want to consider these options
in your kernel config file:
MAXUSERS 256
options "CHILD_MAX=128"
options "OPEN_MAX=128"
options "NMBCLUSTERS=512"
But you should pick your own numbers for these.
My average uptime is out to 4 days now on my heaviest hit box, the other
boxes that have lighter load have very good uptimes.  I have also noticed
very nice performance improvements over the earlier kernels.
The failures that I am seeing are scsi buffer already done freezups.  About
once a week on my 4 gig news harddrive.
Regards,
Mark Hittinger
bugs@win.net
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199505181631.MAA19578>
