generate unit file for st2api, st2auth and st2stream

[nzlosh]

Move api, auth and stream service unit files to generator so st2.conf variables are used consistently for .socket and .service files.

[cognifloyd]

Looks like a bit of copy pasta. I left suggestions to fix the ports.

Also, I wonder if we could combine these so one generator handles all 3 services. But it's probably not worth the effort.

[cognifloyd]

I just noticed that the generator does not include EnvironmentFile any more.

One good thing, is that makes the generator truly agnostic between debian vs EL.

One potential gotcha is that if anyone is relying on this, then that file will no longer be sourced. They can, however, add EnvironmentFile to a systemd service override file /etc/systemd/system/st2api.conf.d/some-file-name-here.conf. That can also be used to override anything else if desired. (The same applies to st2auth and st2stream)

[nzlosh]

As far as I know, the /etc/default/st2api configuration file is how Debian based distributions allowed overriding sysv and upstart service scripts. As you've rightly pointed out, people should use the built-in systemd override mechanism and all currently supported OS' are systemd based. We can highlight this in the v3.8 upgrade notes.

[cognifloyd]

I'm pretty sure tests are failing because this line is having an error that has been silenced:

st2-packages/packages/st2/rpm/st2.spec

Line 89 in 75a0e5e

%service_post st2actionrunner st2api st2stream st2auth st2notifier st2workflowengine

That line lists the 6 services that aren't getting enabled correctly.

[cognifloyd]

Looking at the test error for EL distros:

I think this is already handled by the %post macro which uses the %service_post helper. That runs:

st2-packages/rpmspec/helpers.spec

Line 86 in 23f100c

systemctl --no-reload enable %{?*} >/dev/null 2>&1 || : \

What if you remove the output redirection in the %service_post macro helper? Then maybe we'll see if there's an error. And maybe remove it here too.

[nzlosh]

There appears to be a worker timeout for gunicorn on EL7, but I don't think that's the cause of the services failing to start for the RPM based distros. I still don't see any errors in the logs that indicate why the service is not enabled and started.

[cognifloyd]

aargh. There's still no helpful output during yum install. I don't know which verbosity flag(s) to use (--verbose or --rpmverbosity or both?), but what if we add some verbose settings on this line:

st2-packages/scripts/install_os_packages.sh

Line 15 in 6fba855

sudo yum -y install $(lookup_fullnames $@);

scripts/install_os_packages.sh gets called in the package tests here:

st2-packages/rake/build/setup.rake

Line 13 in 739d24e

execute :bash, "$BASEDIR/scripts/install_os_packages.sh #{opts[:package_list]}"

We need to figure out why the %post bits aren't running like we expect. ☹️

[nzlosh]

I'm going to debug this on a VM, pushing and waiting for the CI to fail is a hindrance and unreasonably slow.