Make sure the right beat service gets restarted

[nyetwurk]

When running more than one role at once, make sure all of the right notifies
get triggered, not just the one for the last beat role run.

[elasticmachine]

Since this is a community submitted pull request, a Jenkins build has not been kicked off automatically. Can an Elastic organization member please verify the contents of this patch and then kick off a build manually?

[jmlrt]

jenkins test this please

[jmlrt]

Hi @nyetwurk,
Thanks for this PR.

Every "debian-based" tests are failing => elastic+ansible-beats+pull-request/10.
It seems that you forgot to rename notify statement in beats-debian.yml.
Can you fix that?

[nyetwurk]

Hi @nyetwurk,
Thanks for this PR.

Every "debian-based" tests are failing => elastic+ansible-beats+pull-request/10.
It seems that you forgot to rename notify statement in beats-debian.yml.
Can you fix that?

Hopefully fixed. Sorry about the oversight :)

[jmlrt]

jenkins test this please

[nyetwurk]

Can someone please approve so I can keep my downstream clean :) thx

[jmlrt]

Hi @nyetwurk,
Code review is good, Automatic test are almost good.

For any reason, I don't understand {{ beat }} isn't expanded on restart handler name on centos (see elastic+ansible-beats+pull-request/11/OS=centos-7,TEST_TYPE=multi at 17:52:40) also it's OK on debian based OS.

I'll relaunch tests as I don't what's the problem in the code.

In addition, I think we could use {{ beat }} in Start beats service task name also.

[jmlrt]

Jenkins test this please

[nyetwurk]

@jmlrt agreed on the Starts beats service name as well. I will add it time permitting. Unfortunately I don't have CentosOS to test on, but do have Ubuntu/Debian/MacOS and all three of those seem fine.

[nyetwurk]

@jmlrt I think there is something wrong with the tests. It doesn't seem as if the handler isn't working, it seems as though none of the tasks that notify are being set as changed: true

[jmlrt]

FYI you can run automated test locally on all supported OS with our KitchenCI setup assuming you have ruby and docker installed.

For this PR, you would need to run make converge PATTERN=multi-centos-7 to test a deployment with filebeat and metricbeat on centos (playbook test/integration/multi.yml).

More details on the testing setup here.

[jmlrt]

@jmlrt I think there is something wrong with the tests. It doesn't seem as if the handler isn't working, it seems as though none of the tasks that notify are being set as changed: true

The restart {{ beat }} handler is executed only when beat wasn't started by Start beats service as you can see in the not beats_started.changed condition.

This way beat doesn't need to be restarted by handler during first ansible role deployment because it is already started by Start {{ beat }} service with up to date configuration.

But if you change beats config and run the ansible role another time with beats already started, Start {{ beat }} service task will be in ok status (beats_started.changed == false) so the restart handler will restart beats to ensure new config is loaded.

[jmlrt]

LGTM 👍

[jmlrt]

Jenkins test this please

[nyetwurk]

Does that mean we need separate beats_not_changed for each kind of beat like we did for the handler (e.g. {{ beat }}_started.changed or {{beat}}_started.changed) since if you run two instances of the role from the same parent, the last one wins?

[jmlrt]

Yes we may need it, also I'm not sure.
I'm in holiday next week so I won't be able to test it.
Feel free to do some test if you have time for it or I'll take some time for that the week after.

[jmlrt]

@nyetwurk
I finally found the time to do a few tests.

I can confirm you that we don't need separated not beats_started.changed conditions.

When testing it with make converge multi-centos-7:

• If I change only the configuration of filebeat (first instance in the role) and replay the role => only filebeat is restarted
• If I change only the configuration of metricbeat (second instance in the role) and replay the role => only metricbeat is restarted
• If I change only the configuration of both filebeat and metricbeat and replay the role => the two beats are restarted.