[9fans] closeproc slow

[9fans] closeproc slow

[erik quanstrom]

i killed off some webfs processes that were lingering.  it took
the kernel about 5 second per process to reap them on an otherwise-
unloaded system:

; while(sleep 5)
	ps|grep closeproc|wc
     36     252    2304
     35     245    2240
     34     238    2176
     33     231    2112
[...]

either all of these processes are in state Queueing or all but one
are in state Queueing and the last is in state Wakeme.

so the (probablly useless) question is why does this take so long?

- erik

Re: [9fans] closeproc slow

[erik quanstrom]

that makes sense.

do things happen differently if the process is not slain
but dies a natural death?

- erik

On Sun Jul 23 10:32:39 CDT 2006, rsc@swtch.com wrote:
> Your processes died a quick and painless death.
> The closeproc helpers, called upon to clean up after
> the battle, leave at a rate of one per five seconds.
>
> I was trying to avoid repeatedly creating and deleting
> processes if someone is killing procs frequently.
> In that case, the right number of closeprocs will stick
> around and get reused.
>
> Russ

Re: [9fans] closeproc slow

[erik quanstrom]

thanks.

- erik

Re: [9fans] closeproc slow

[Russ Cox]

Your processes died a quick and painless death.
The closeproc helpers, called upon to clean up after
the battle, leave at a rate of one per five seconds.

I was trying to avoid repeatedly creating and deleting
processes if someone is killing procs frequently.
In that case, the right number of closeprocs will stick
around and get reused.

Russ

Re: [9fans] closeproc slow

[Russ Cox]

> do things happen differently if the process is not slain
> but dies a natural death?

yes.  the closeprocs only come out when you run slay.
their job is to close the file descriptors that the slain
processes left behind.  you have to do it in a different proc
to avoid deadlocks in the general case.

see http://9fans.net/archive/2006/01/48 and the ensuing
discussion.

russ