[QA] Upgrade procedure fails to shut down ocis when using systemd

[jnweiger]

The upgrade guide correctly suggests to stop ocis before updating. The binary / bare metal install case is discussed in https://github.com/owncloud/docs-ocis/blob/2360692c5141e4fb8145a80fe46f52b6a7a87c4a/modules/ROOT/pages/migration/upgrading_5.0.x_7.0.0.adoc#2-shut-down-the-infinite-scale-instance

which correctly refers the user to: https://github.com/owncloud/docs-ocis/blob/master/modules/ROOT/pages/depl-examples/minimal-bare-metal.adoc#stopping-infinite-scale

There we learn, that for a graceful and complete shutdown, a rather complex shell pipeline should be used:

ps -ax|grep "ocis server" | grep -v grep | awk '{ print $1 }' |   xargs sudo kill -15

But: When the bare metal install was done according to https://doc.owncloud.com/ocis/next/depl-examples/bare-metal.html#setup-the-systemd-service this will not have the desired effect.

Systemd treats the kill -15 command as an incorrect shutdown and restarts ocis immediately. This behavior is correctly configured in the service file via Restart=always - This is a valuable safeguard against spurious crashes.

Thus following this procedure, we'll have ocis unintentionally running while doing the remaining upgrade steps. This most likely leads to errors, frustrated users or even data-loss.

I suggest, that for each setup, the respective tooling is used and we properly distinguish the systemd case from other binary deployments. For a systemd setup, the correct command for a graceful and complete shutdown is: service ocis stop

[jnweiger]

For the non-systemd case, I'd suggest to consider killall -v -s 15 ocis (or a regexp variation of this)

[mmattel]

ok. this sounds valid. I will create a new PR independent to the migration so we can backport to cover:

• killall
• systemd stop

Thanks for the investigation in filing the issue 👍