Hi @HaoK. If this is not a tell-mode PR, please make sure to follow the instructions laid out in the servicing process document.
Otherwise, please add tell-mode label.

@dougbu this is probably worth a shot for servicing helix runs, always retry 3 times for failures universally

Sorry, I'm not going to approve this without buy-in on ignoring flakiness in our servicing tests from @mkArtakMSFT @adityamandaleeka @rafikiassumani-msft and hopefully @Pilchie. This is a heavy hammer and, as you said elsewhere, it's not supported in the release/5.0 branch (or did I misread something❔).

Another option would be to make it our policy to backport all test-only changes as far as they're relevant.

Leads, see also discussion in #40467

I'm sure it is supported in release/5.0 as its a helix feature, but I don't think the retry stuff was ever ported to 5.0, so it'd require that. Just to be clear, I'm not pushing for this, this was just in response to your comment

we need to keep on top of flaky test fixes in our servicing branches because the window the branches are open doesn't leave us much time for retrying PRs and the impact of flakiness here can be delayed releases.

This would get you what you want (more green servicing builds), without whack-a-mole in servicing for tests

I'm sure it is supported in release/5.0 as its a helix feature, but I don't think the retry stuff was ever ported to 5.0, so it'd require that.

Do you mean our retry stuff wasn't ported to release/5.0 in this repo❔ We definitely don't have an eng/test-configuration.json file.

In any case, Helix SDK features vary between the Arcade branches. My question was whether their down-level branch contained the support. Reading https://github.com/dotnet/arcade/blob/release/5.0/src/Microsoft.DotNet.Helix/Sdk/tools/Microsoft.DotNet.Helix.Sdk.MonoQueue.targets (which is where test-configuration.json is found in release/6.0), the answer seems to be "no". We'd either need to ask for the feature to be backported (probably easily justified) or choose another option.

Just to be clear, I'm not pushing for this, this was just in response to your comment

No worries. I'll admit I wasn't positive how strongly you felt about this option and appreciate the clarification❕

Personally, I'm in favor of both approaches:

1. When making flaky test fixes in main, spend the time to see if they also apply in servicing branches and can be easily ported there. If so, then port them.
2. Increase retries for servicing branches to help get PR builds through in the limited windows we have.

Both together would reduce risk of retries masking real issues, while also helping get PRs through servicing branches.

Increase retries for servicing branches to help get PR builds through in the limited windows we have.

To clarify: Increase retries for all Helix work items or use retries similarly to what we have in 'main' (possibly w/ additions for servicing-specific test flakiness)❔