Hi!
I have finally reproduced the "long lost child" messages on my
development machine and made a fix for it. First of all, these messages
were silently ignored by the server if no graceful restart had been made
on the server. The patch below also removes this restriction, so it
would report these children in any state.
These messages were logged if a kill signal was sent to a processor, but
before the child could exit, the multiplexer would try to revive the
child (multiplexer always keeps the processor child alive, if the vhost
is active), which would create a new child in the slot and thus making
the previous processor gone from the scoreboard. No other harm was done
except the log messages because the previous processor was shutting down
anyway. Potential improvement here may be to not kill the processor if
the vhost has active workers or make the multiplexer activate the
processor only if the server environment has no workers alive.
The patch is located here: http://source.kood.ee/peruser-lostchildfix.patch
(should apply to both v13 and v15)
--
Taavi Sannik
DataCode OY
_______________________________________________
Peruser mailing list
[email protected]
http://www.telana.com/mailman/listinfo/peruser
