M(ss) job are getting coalesced on m-c and this makes comparisons between pushes more complicated and breaks automated change alerts since they are based on comparing two pushes not each platform separately within the pushes. Is there an easy way to disable coalescing for both buildbot and TC for this job? Taskcluster config: https://dxr.mozilla.org/mozilla-central/rev/a08ec245fa24d573fc99e81210ecc09de734cdd3/taskcluster/ci/test/tests.yml#501 Buildbot config: https://hg.mozilla.org/build/buildbot-configs/annotate/d816f8ed797e49d73755ddcf16f616b153f30ffe/mozilla-tests/config.py#l407

on m-c, searching for screenshot, I see just linux64 M-e10s(ss): https://treeherder.mozilla.org/#/jobs?repo=mozilla-central&filter-searchStr=screenshot&fromchange=106a96755d3bcebe64bbbc3b521d65d262ba9c02 I see this per commit (unless we have a DONTBUILD), a bit more information can help me find the root cause and get a fix in place!

oh wait, tier-3 for osx and windows- let me look a bit more, I see some holes.

Yeah, they are tier 3 for the buildbot ones (The linux64 recently defaulted to a different tier I think). Also note that I used trigger.py to fill in most of the gaps since the beginning on January so you may not see the gaps as easily unless you look at the timestamps to notice ones where I backfilled.

I'm not sure if disabling coalescing is the right solution to your problem. Would backfilling be a better answer?

(In reply to Chris AtLee [:catlee] from comment #4) > I'm not sure if disabling coalescing is the right solution to your problem. > Would backfilling be a better answer? If you mean that a user has to manually backfill, then no, this is for a fully automated job where humans only get told about the changes between revisions and won't know to backfill every missing job. The job isn't a pass/fail job, it's basically always green and produces output which is analyzed outside of TreeHerder.