RFR: 8284389: Improve stability of GHA Pre-submit testing by caching cygwin installer

Christoph Langer clanger at openjdk.java.net
Wed Apr 6 10:35:40 UTC 2022


On Wed, 6 Apr 2022 07:40:41 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:

> Let's not switch `v2` -> `v3` in this PR.

You're right. I'll do another PR to update the actions versions. There are also updates possible for upload and donwload actions.

> So `windows_aarch64_build`, `windows_x64_build` checks the cache and adds recovery step. But `windows_x64_test` does not do it? Why? If the cygwin installer was cached, it would work fine, but if it does not -- for example if `windows_x64_test` job is restarted in isolation after cache flush -- it would break? I think we should be consistent in this: if installer is not in cache, we always do the recovery.

Yeah, I wasn't happy with this, too. I, however, thought that we could maybe assume that the build step was run successfully before the test step and therefore the cygwin installer should be in the cache and we would never fail. But, I guess you're right, the cache could be evicted in between due to whatever reason and we're unnecessarily abstaining from a safety net.

What I want to accomplish is that if it fails due to an issue with downloading the cygwin installer, it should fail in the build job, not later. But this should be the case in 99.99% now anyway, due to the added caching.

I will implement your suggestions.

-------------

PR: https://git.openjdk.java.net/jdk/pull/8113



More information about the build-dev mailing list