I looked at one of the logs from bug 1425365 but unfortunately the sccache log didn't contain useful info. As it turns out, that's because we shut down sccache after the build: https://dxr.mozilla.org/mozilla-central/rev/56b988a937689d5599400afa59b72c390b40abf2/client.mk#160 However, in rusttest builds when we run `make check` we wind up building more Rust code which will start sccache again, but without logging enabled, which is unfortunate. I guess the best fix here would be to move the sccache shutdown into mozharness so it runs after the build and check steps.

In builds that use sccache, the sccache server would be shut down at the end of the build step in client.mk. Unfortunately in rusttest builds the check step winds up compiling more Rust code which restarts the sccache server but without the proper configuration. This patch moves sccache shutdown to a post-script step in mozharness so that the server will stay running through the check step.

This is looking pretty good on try so far. I'm retriggering lots of win64-rusttest jobs to see if any of the bugs in the dependency graph crop up.

Unfortunately the answer is yes. bug 1470384: https://treeherder.mozilla.org/logviewer.html#?job_id=203191245&repo=try&lineNumber=48225 https://treeherder.mozilla.org/logviewer.html#?job_id=203191248&repo=try&lineNumber=48224 bug 1396845: https://treeherder.mozilla.org/logviewer.html#?job_id=203191252&repo=try&lineNumber=48397 It doesn't seem to have made anything worse, however.

Pushed by tmielczarek@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/47bcc150ba0b shut down sccache server in mozharness after all build steps. r=firefox-build-system-reviewers,gps