I had an extensive discussions with Takuto on cq false rejections, and while there are efforts going on reduce false rejections, there haven't been much attempts on minimizing the impact of cq false rejections.

One idea we generated is:

Assuming a CL is sent to CQ, and the first attempt failed due to a test failure on mac_chromium_rel_ng builder, browser_tests, ExtensionBrowserTest.OverrideStartupPagesSettings, then another builder-level retry is immediately initiated on mac_chromium_rel_ng, and *all* test targets will be re-compiled and *all* tests will be re-run.

However, it is sub-optimal and unnecessary. A better approach is to *only* re-compile the failed test targets and *only* re-run the failed tests in the second attempt on the mac_chromium_rel_ng builder.

It is of course lower priority than reducing CQ false rejection rate itself, but worths thing more along the line.