samba - Jun 2021 - OOM killed samba

One of the Samba DC's was killed by the the out of memory manager.
This gave us some problems which we have resolved now.

Though I am wondered, is it possible to configure OOM as follows:


If the memory usage is too high, kill the process,

?stop samba, wait 5 minutes, start samba,

?if starting fails 2 times (thus thats 10 minutes later) restart the whole
server.

?(Sending a message optionally would be great too).





In windows services are often configured with failover options, 1st time restart
2nd time restart, 3th time reboot



---

tldr the server in question had some memory issues which operantly it solved by
killing randomly samba processes
it never draw much attention as it never resulted in any problem
As it was only a process and not the whole server it didnt get notified, Samba
always seamed so stable.
Till that day the other DC wanted to talk with it just halfway the moment it got
killed.
The good working DC came into error state and stopped working as its partner was
down.
Still investigating to the root cause of memory problems we added extra memory.

The kernel OOM killer is rather inflexible and there's not much you can 
do about it.

What you can do is

a) Restart samba regardless of why it has been killed ? most init 
systems more sophisticated than sysvinit can do it out of the box, and 
there's workarounds for sysvinit too (inittab, pipexec)

b) Preempt the kernel OOM killer with an userspace one. systemd-oomd is 
the only I'm aware of, it relies on systemd service definitions to set 
up automated restarts.

Generally, if you want Linux services to act like Windows services, you 
want systemd, that's what it's made for. (For better or worse.)

-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature
Type: application/pgp-signature
Size: 665 bytes
Desc: OpenPGP digital signature
URL:
<http://lists.samba.org/pipermail/samba/attachments/20210607/93f31027/OpenPGP_signature.sig>