[OpenIndiana-discuss] oi_151a3 svc.startd chewing up memory

[Milan Jurik]

Hi,

Jim Klimov píše v út 29. 05. 2012 v 04:16 +0400:
> 2012-05-29 3:39, Jim Klimov написал:
> > Hello all,
> >
> > On a test box running OI oi_151a3 I found that it can not restart
> > services due to svc.startd being 2500M in VM size. As a consequence,
> > fork() fails due to insufficient free VM (swap space) and services
> > can not start.
> >
> > I guess there is a leak somewhere, was anything like that fixed in
> > the past month or so?
> 
> A bit more research points to an old problem, marked as fixed:
> http://defect.opensolaris.org/bz/show_bug.cgi?id=15761
> 
> This box did have a pkg/server (startd/duration=child) instance
> which did not start well and was left astray. From what I see
> now, a daemon is started and in particular grabs the TCP port,
> but SMF thinks the service has failed and restarts it. Further
> invokations fail due to busy port, but the service does not go
> into maintenance. I see svc.startd occupying more RAM at a rate
> of about 1Mb/1-2mins (glancing at top).
> 
> The symptomatic part can be fixed by simply changing the base
> pkg/server startd/duration attribute from "child" to "contract",
> but the core problem - svc.startd leaking memory in case of such
> unlimited restarts - is still in place. Also, for the past ten
> minutes or so since I fixed the pkg/server, the restarter hasn't
> released a byte ;)
> 

I would start with simple gcore file analysis - svc.startd is linked
with libumem, so you can take gcore and look at leaks and memory
consumers with mdb.

> HTH,
> //Jim Klimov
> 
> 

Best regards,

Milan