Hi Willem,
Indeed, NSD 4.8.0 did not log this condition as an error message and
just proceeded if the old-main would quit.
With 4.9.0 reloading was refactored to reap exited old serve childs in
order to reduce the number of "defunct" or "zombie" processes that can
emerge (for example when one old-serve child is still busy, for example
serving an AXFR or so).
When old-main is done with is job during reload (killing the old serve
children), it informs the reload process and then immediately exists.
The detection of the closed pipe (because of exited old-main) could very
well become before the information that old-main is done on that pipe on
some platforms.
Thank you for the fantastic explanation! The graphic is also very
helpful for understanding the logic flow.
So I consider this "reporting of an exited old-main" at this point in
the code a bug, and changed it into a debugging warning level message
here: https://github.com/NLnetLabs/nsd/pull/421
PS. For completeness a strip of a successful NSD reload below. Your
issue would occur if old-main(5) would exit before the load(2) process
received the NSD_RELOAD in picture 6.NSD successful reload
That makes sense given the previous information you provided. I look
forward to running a future NSD version with that patch :-).
Cheers,
Otto
_______________________________________________
nsd-users mailing list
nsd-users@lists.nlnetlabs.nl
https://lists.nlnetlabs.nl/mailman/listinfo/nsd-users
