Re: make the stats collector shutdown without writing the statsfiles if the immediate shutdown is requested.

On 2021/03/18 11:59, kuroda(dot)hayato(at)fujitsu(dot)com wrote:
> Dear Ikeda-san,
>
> I confirmed new patch and no problem was found. Thanks.
> (I'm not a native English speaker, so I cannot check your comments correctly, sorry)

One user-visible side-effect by this change is; with the patch, the stats is
cleared when only the stats collector is killed (with SIGQUIT) accidentally
and restarted by postmaster later. On the other than, currently the stats is
written in that case and subsequently-restarted stats collector can use
that stats file. I'm not sure if we need to keep supporting this behavior, though.

When only the stats collector exits by SIGQUIT, with the patch
FreeWaitEventSet() is also skipped. Is this ok?

- * Loop to process messages until we get SIGQUIT or detect ungraceful
- * death of our parent postmaster.
+ * Loop to process messages until we get SIGTERM or SIGQUIT of our parent
+ * postmaster.

"SIGTERM or SIGQUIT of our parent postmaster" should be
"SIGTERM, SIGQUIT, or detect ungraceful death of our parent postmaster"?

+SignalHandlerForUnsafeExit(SIGNAL_ARGS)

I don't think SignalHandlerForUnsafeExit is good name. Because that's not
"unsafe" exit. No? Even after this signal handler is triggered, the server is
still running normally and a process that exits will be restarted later. What
about SignalHandlerForNonCrashExit or SignalHandlerForNonFatalExit?

Regards,

--
Fujii Masao
Advanced Computing Technology Center
Research and Development Headquarters
NTT DATA CORPORATION