+Ryan/Erik for eventual migration to milo, which could (?) help.

FYI, you need to log in to see data on the dashboard.

The problem is that the step is named "page_cycler_v2.intl_es_fr_pt-BR on ATI GPU on Mac on Mac-10.11", but the link assumes the name is "page_cycler_v2.intl_es_fr_pt-BR"

This is caused by the swarming code, which adds something about ATI GPU etc.. to the step. I've seen this code before, let me find it.

I don't like it, because it confuses sheriff-o-matic. I'm fine with removing it, but we should see why it was added in the first place.

https://codereview.chromium.org/1720873002 was the change which added this. 

Assigning to pawel; do we still need the os dimensions in the step name?

The reason for adding the OS is that the same test can be triggered multiple time on multiple OSes. The OS name is only present when it's not using the "default" value as per the recipe.

+1 to removing this at least for the perf use case

Besides the issue with the perf dashboard links (which I am wondering if we can somehow fix in the upload script and strip the step name) this makes it very hard to quickly see what is failing on a given host bot on the perf waterfall.

For the perf use case, each host bot will be triggering jobs on swarming bots that all have the same dimensions (os, gpu etc).  Therefore, if you want the name of the os  you can look at the host bot name.  When looking at the waterfall overview page it is hard to determine at a glance what is failing when every failing benchmark as the boilerplate os and version as it is part of the failing step name.

My suggestion would be to make dimensions part of step text by default (not step name), and only add them to step names if we need to disambiguate (when same test is triggered with different dimensions on the same bot).

Important thing to note is default swarming dimensions may be different from OS running the buildbot slave, so I found it helpful to make effective dimensions clearly visible somewhere in the UI to notice mistakes/mismatches easier.

Assigning to Andy for crossover team. Feel free to assign back to me.

This bug is not untriaged. 

ping

Who is owning the implementation of this so we can prioritize it and move it forward?  I would be happy to help out with implementation with a little guidance.

Go for it, Emily. Feel free to reach out on this bug as your progress through the problem. The right group of people is CC’d here.

Personally, I think just adding the OS information to the step text would be good enough. Unfortunately, a lot of stuff depends on step names, so I think we should just keep the step name clean and equal to the test name.

I'm going to take this.

I think the consensus here was that step names aren't great, but there's not an easy way to fix them? I'm not sure.

Ned, should this still be worked on?

I am not sure this feature is still useful in the world of single perf setp, where a log may be quite big.

I put this blocked on CTU. We can reevaluate this once that project is done.

I am closing this given we now have a direct link to "Benchmark_logs" from the build page as well as included links in SOM.