SP metadata cache keeps growing

[Peter Schober]

* Cantor, Scott <cantor.2 at osu.edu> [2020-02-26 22:37]:
> It's not the metadata, those are the JSON feeds. The cleanup bug
> reappeared a while ago and was fixed again in 3.0.
> 
> https://issues.shibboleth.net/jira/browse/SSPCPP-731

Lol, I even quoted
https://issues.shibboleth.net/jira/browse/SSPCPP-612
https://issues.shibboleth.net/jira/browse/SSPCPP-731
to the reporter but said those won't apply directly since those were
about the JSON feeds, not the XML.

> Restarting is stopping and starting, and even parsing the metadata
> alone, without the signature check, takes a long time. The systemd
> behavior is similar to before, you still have to customize the delay
> using a custom unit file.
[...]
> My guess is systemd is actively killing the process perhaps? The old
> init.d wait didn't really have that effect. Anyway, the delay is
> adjustable with the unit file.

I'll try to get a CentOS 8 up and running (have been meaning to do so
for a while now, my test SP still is on CentOS 6, i.e., a SysV init)
and recommend an override, thanks.

> They aren't metadata copies, they're the JSON feeds (just peek at
> them).

I was going by the directory listing I posted earlier and my test
system where /var/cache/shibboleth/*.xml are XML files and the others
are named *.json, so that didn't match the above issues.

Cheers,
-peter