httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
Subject Re: Whacking shared memory on graceful restarts is bad
Date Sun, 30 Jul 2000 15:46:18 GMT

The problem is easy to solve.  The APR shared memory stuff shouldn't be
registering a cleanup with the pool.  I always had a problem with making
the shared memory stuff register a cleanup, because the lifetime of the
pool wasn't the same as the lifetime of the shared mem.  If we make the
shared memory stuff use malloc and free, then we are all set.

I'll make this change later today.


On Sun, 30 Jul 2000, Manoj Kasichainula wrote:

> Dexter, mpmt_pthread, and probably the other Unix MPMs react in a
> SIGSEGV kind of way when sent a graceful restart signal.
> In these MPMs, we're actually now freeing shared memory upon cleaning
> the pconf pool. This is not a good thing, because during graceful
> restarts, threads that are continuing to process requests must still
> be able to write to their old shared memory region. This change
> appears to have been made when switching to APR for shared memory.
> I'm not too familiar with the pre-APR shared memory code; it appears
> that in 1.3 we just didn't free the shared memory on restart, except
> for USE_POSIX_SCOREBOARD, where we did shm_unlink, whatever that is.
> If it wasn't for the POSIX shmem case, my suggested fix would be to
> just eliminate that cleanup.

Ryan Bloom               
406 29th St.
San Francisco, CA 94131

View raw message