> On Sun, Jul 25, 2004 at 05:12:52PM +0200, Jean Delvare wrote: > > 3* Any idea what the problem could be? How would I investigate? > > My guess would be that interrupting the compile caused distcc to mark > one or more hosts as unresponsive, which causes distcc to temporarily > ignore the host. Do you see any log messages about a host being in a > "backoff period"? I think a file will end with the string "backoff" > in the "lock" subdir of the .distcc tmpdir when this happens (though > the comment at the start of timefile.c says that it uses the state > dir, it appears to me that the code disagrees with this).
Interesting guess but it doesn't quite match what I observe. Just rebooted both systems, deleted any temporary file in ~/.distcc/*, checked that nothing was left in /tmp, ran "make clean" then "make CC=distcc -j4" in the linux-2.6.8-rc2 directory. I got only 3 jobs. So it can happen without a previous compilation having been interrupted. I then interrupted that build, and restarted it... 4 jobs this time. Interrupted, restarted, 1 job only. No stale file anywhere as far as I can see. Also note that the missing jobs are not per-host. When I get 3 jobs, one of them is running on the remote server, which is supposed to get the missing 4th job. And for confirmation I don't see any "backoff" file. If it was a distcc problem I would expect the misisng job(s) to come back after some time, but they never do. Instead, they come back when a new build starts. This is what makes me think that this is really "make" incorrectly defining the number of jobs as it starts. But then I don't get why using distcc seems to trigger the bug. Thanks for the pointer anyway. What I find really strange is that this happens only when I run Linux 2.6 and only when I compile Linux 2.6. I cannot see how this is supposed to matter, but still it seems to do. I believe that my best chance now would be if someone else would succeed in triggering the same problem. If nothing else it would demonstrate that I am not crazy ;) Thanks again, -- Jean "Khali" Delvare http://khali.linux-fr.org/ __ distcc mailing list http://distcc.samba.org/ To unsubscribe or change options: http://lists.samba.org/mailman/listinfo/distcc
