[Buildbot-devel] `make clean' is too slow on buildbot

[Weongyo Jeong]

On Fri, Apr 23, 2010 at 12:23:12PM -0700, Dustin J. Mitchell wrote:
> On Fri, Apr 23, 2010 at 12:11 PM, Weongyo Jeong <weongyo.jeong at gmail.com> wrote:
> > It's happening for all commands that `make clean' was a just example.
> > The following is a result when `make clean' is running.
> 
> Try running strace against it to see what it's up to?  You may want to
> pick something simpler than 'make', if everything is this slow -
> /bin/true should be pretty simple.

I tried to see what's going on at `make' command using strace(1) and the
result is as follows:

Process 488 attached - interrupt to quit
brk(0x11b3b000)                         = 0x11b3b000
brk(0x11b5c000)                         = 0x11b5c000
brk(0x11b7d000)                         = 0x11b7d000
brk(0x11b9e000)                         = 0x11b9e000
brk(0x11bbf000)                         = 0x11bbf000
brk(0x11be0000)                         = 0x11be0000
brk(0x11c01000)                         = 0x11c01000
brk(0x11c22000)                         = 0x11c22000
brk(0x11c43000)                         = 0x11c43000
brk(0x11c64000)                         = 0x11c64000
brk(0x11c85000)                         = 0x11c85000
brk(0x11ca6000)                         = 0x11ca6000
brk(0x11cc7000)                         = 0x11cc7000
brk(0x11ce8000)                         = 0x11ce8000
brk(0x11d09000)                         = 0x11d09000
brk(0x11d2a000)                         = 0x11d2a000
brk(0x11d4b000)                         = 0x11d4b000
brk(0x11d6c000)                         = 0x11d6c000
brk(0x11d8d000)                         = 0x11d8d000
brk(0x11dae000)                         = 0x11dae000
brk(0x11dcf000)                         = 0x11dcf000
...
...
and so many brk() system calls.

It looks make(1) command consumes 100% CPU with calling brk(2) system
calls and it takes about 250MB ~ 300MB memory!  Normally `gmake' worked
without consuming so many memory but on buildbot it consumes so many
memory and spends most of time to call brk(2).  After calling brk(2) it
seems the process becomes a normal as it did.

It looks there are some issues combining GNU make and buildbot.  Any
ideas why it comes out on buildbot specially?

regards,
Weongyo Jeong