How We Made Our CI Pipeline 5x Faster | ...

[alpine igloo]

Social links if you want to boost:

https://x.com/sagikazarmark/status/1829090627151376705
https://www.linkedin.com/feed/update/urn:li:share:7234856320661635074/

[warm bridge]

Thanks @alpine igloo ! This is awesome.

[barren trench]

@alpine igloo How long does a GHA workflow run wait on the runner to start? We use ephemeral EC2s right now that take 20-40s before a GHA workflow run actually starts on a runner. In any case, great work! I'm seriously considering Depot.

[alpine igloo]

I haven’t measured it, but it’s definitely under 40s

[barren trench]

I'm seeing it take 30s with depot-ubuntu-22.04-arm-32-io and over 60s with depot-ubuntu-22.04-arm-32. Bummer, I'd like to get that much lower. I'm also seeing Depot runners take about 2x on certain tasks like 'setup job' and 'checkout repo' compared to our runs-on.com runners. I'll keep fiddling with it. Did you integrate dagger with Depot at all? I'm also interested in reducing the duration of the dagger engine cache pull that occurs on each runner startup.

[alpine igloo]

We don't use ARM runners...those may take a bit longer, I'm not sure.

Did you integrate dagger with Depot at all?

Not sure what you mean by that, but there is a depot module on daggerverse.

I'm also interested in reducing the duration of the dagger engine cache pull that occurs on each runner startup.

Yeah, that's still a headache for us too. Since the engine cannot tell in advance which cache volumes are going to be necessary, it simply downloads all.

@lyric violet mentioned some improvements are coming on this front.

[barren trench]

By integration with depot I meant in regards to improving the dagger engine cache pull, even broader use of depot persistence.

Thanks for the info!

[alpine igloo]

I guess that's up to Dagger and Depot to figure out.

[warm bridge]

@primal verge @hasty swan @upper karma 👀

[hasty swan]

We're working on this on two fronts:

1/ making it possible to store dagger volumes on local dirs, this will allow to store the volumes cache on depot persistent storage (same mechanism than the normal Github runner with the cache action).

2/ we're working with the Depot team on figuring out the best integration for dagger for faster pipeline, your feedback there will help, starting with your comments above.

[lyric violet]

@alpine igloo @barren trench good news, we have PR ready that will make cache pulls on-demand which significantly improves the experience of the distributed cache usage. We're expecting to release this some time by the end of this week / beginning of the next one 🙏