Re: testsuite failure - 193 parallel execution

[Eric Blake]

On 07/20/2010 03:17 PM, Paul Eggert wrote:
> On 07/20/10 14:11, Eric Blake wrote:
> 
>> Hmm, maybe we should do more validation in the parent - not just read
>> from the fifo, but validate that we actually read a token.  That way, we
>> can ignore EOF (on the grounds that it may be a temporary condition, and
>> that more children remain to be run) and only advance when the token we
>> read back is non-empty.
> 
> That would be better than now, but taking a step back, why are we using
> a named fifo at all?  Why not simply use a pipe?  Then we wouldn't have
> to worry about all this after-you-Alphonse business.  Children could
> write to the pipe when they're done.  The parent could read from the pipe
> to find out when a child died.

How do you propose to set up the pipe in the parent in a way that you
can control which fd it is tied to in the child?  The only way I know to
do that is via a named pipe (aka fifo).

Unless maybe something like this would do (untested)...

exec 7>&1 # save orig stdout
{ exec 6>&1 1>&7 # save pipe's stdout, restore stdout
  for each child do
    wait for permission from parent to start test
    (
      test
      echo >&6
    )&
  done
} | {
  wake up n children
  for loop that reads stdin, and wakes up children as data comes in
  done
  prune remaining n reads off stdin
}
exec >&7 7>&-

But how does a child block until the parent is ready for the next
parallel group, and how does the parent signal that a child can run?
Here, maybe one named fifo per child will work, where the child blocks
by doing:

exec 8>test-specific-fifo

the parent wakes it by doing:

exec 8<test-specific-fifo

then both processes can close the fifo, without doing any I/O over the fifo.

And how does signal handling fit in with interrupting a child test?

-- 
Eric Blake   address@hidden    +1-801-349-2682
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature