Experiments build the debug version of the images.
Both debug and non-debug builders use the same export tag
If the debug builder ends after the non-debug builder, it will overwrite the tag.
That causes a whole chain of issues:
1) Runners don't find target binaries.
2) Runners generate an empty corpus
3) Measurers pick up an empty corpus
4) Measurers try to run coverage with empty corpus.
5) "Coverage run failed" errors
Depending on the fuzzer+benchmark being built the issue may be more or less frequent.
That causes a whole chain of issues: 1) Runners don't find target binaries. 2) Runners generate an empty corpus 3) Measurers pick up an empty corpus 4) Measurers try to run coverage with empty corpus. 5) "Coverage run failed" errors
Depending on the fuzzer+benchmark being built the issue may be more or less frequent.