fix(cypress): improve cypress performances

[skjnldsv]

• Prevent fetching all deps again with npm ci for each runner (node_modules is already saved in the init step)
• Enable redis to speed things up
• Allow restoring DB snapshots and implement it o the profile spec (personal-info.cy.ts)
• Make Nextcloud setup into RAM

Before	After

[skjnldsv]

Also tested a new approach, restroring DB snapshots.
A good example is the profile test going from 3 minutes, 22 seconds to 2 minutes, 36 seconds

[artonge]

Seem arbitrary. Can't we wait for something more specific?

[skjnldsv]

Yeah, we'll standardise this a bit more in a followup.

[susnux]

Why is this needed?
As soon as this part is reached the create user request has terminated so all database actions are already finished as the PHP process is done.
So normally this should not be needed I guess?

[skjnldsv]

Follow-up, use functions from @nextcloud/cypress

[susnux]

This is much better, thank you!
But there is still a performance issue, see runner 3:

Overall Run: 13 minutes 14 seconds
Cypress tests : 6 minutes 22 seconds
Pulling image: 1 minute
Restore context: 3 minutes

more than 50% of the CI time is still not related to tests.

Pulling the image takes locally for me much less time, I assume our runners do not have it cached and thus need to externally fetch it from GH?

But the thing that is weird: Creating the context takes ~2:30 while restoring takes ~3 minutes.
This are 27 minutes wasted by restoring the context, maybe this is something we can improve in a follow up?

[susnux]

🚀

[skjnldsv]

But the thing that is weird: Creating the context takes ~2:30 while restoring takes ~3 minutes.
This are 27 minutes wasted by restoring the context, maybe this is something we can improve in a follow up?

It's decompressing the archive I think, it takes many IO.

more than 50% of the CI time is still not related to tests.

That would be my guess as well, the more we have running, the more load our CI servers have. Meaning it will obviously have an impact on other runs.

[skjnldsv]

Pulling the image takes locally for me much less time, I assume our runners do not have it cached and thus need to externally fetch it from GH?

It does, we have a docker layer cache. We store images.
But this one is a new image, and we have 4 CI servers. Maybe that was a cache miss occurence, should be better now 👍