[Raw Msg Headers][Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Router dumping core



...
> > > >        I'm just wondering if anyone has been experiencing their router
> > > >process dumping core.  I am running version 2.90-940626 on a Sparc LX,
> > > >sunos 4.1.3_U1 ...  Zmailer was compiled with gcc 2.5.8  ..
...
> > 	I am seeing it too, but beats me what is up..
> > 	(careless string-pointers somewhere ?)  Maybe this night..
> > 	Maybe a problem with  rfc822.ssl -- use the old version ?
> 
> Which version, the original 2.2 ? or one of your intermediates.  BTW,
> just happened again ...  I get several core file in the top level
> /var/spool/postoffice  it looks like only three of the router deamons
> dumped core ?

	My lattest dump, of course (when I talk about cores, when others
	say, I don't know..)

	Figuring which router dumps core is "a bit" difficult, thus
	the configuration script does its own tricks on getting unique
	id for the file.  Maybe not the best way to do it, but livable..

	Anyway, I got HUGE logs into  $LOGDIR/router  when the script
	left cores into spool...  Also the idle loop spinned considerable
	times.  (Ok, a bit of creativity within  rd_stability() -routine
	could do the same as the scrips do now when they move the cores
	away, but it would be less flexible..)

> Question, it seems that when running 2.2, the router processes each
> consumed about equal CPU, even over the long haul, now there are bit
> differences between them ....

	Yes, Though most likely that is due to differing kind of jobs
	those routers have gotten over their life-time.  They do race
	asynchronously, after all..

	Here at UTU.FI the typical job-mix resolves easily, thus their
	(4 routers) memory images are about identical in size, and
	CPU-times are the same.

	At NIC.FUNET.FI  two routers have handled postings to long lists
	and those have huge core-sizes (30MB), one has medium size core,
	and the fourth one is 1M..  CPU times vary from 45 sec to 100 sec
	and don't depend upon core-size. When one routes long list, others
	can spend time spinning small quick jobs..

> Looks like I have to keep a close eye on things ...
> 
> Cheers,
> 
> Marco


	I have some ideas about handling  tmalloc()ed memory, lets see
	if I can get some order out of it.  Basically the intention is
	to let all non-local objects to ride on "ring 1", and rest to
	alloc (and release) rings with higher orders when a recursion
	is used.  Bringing return data from recursion level to the
	caller level does need some creative copying, though.

	I think there can be practically infinite ( > 100 ) number  of
	levels of tmalloc() rings.  That should handle all our needed
	recursions.

		/Matti Aarnio

	PS: My CNAME recursion blocker was coded in a year or so ago..
	    Could you see, why sometimes  router  can route, but SMTP
	    doesn't find MXes, nor As ?  Wierd, isn't it ?