Sometimes the benchmarks fails out of bad luck (e.g. network failures during the docker build). One can hit the "Rebuild" button on the build logs page which will fix the misstep. However, the current-bench database still remembers the initial failure with the failed=true and reason column -- and the frontend keeps displaying the commit as failing.
(It's probably not ideal that the database has both success, failed and cancelled columns, I'm not sure how to interpret them when more than one is true!)
Sometimes the benchmarks fails out of bad luck (e.g. network failures during the docker build). One can hit the "Rebuild" button on the build logs page which will fix the misstep. However, the current-bench database still remembers the initial failure with the
failed=trueandreasoncolumn -- and the frontend keeps displaying the commit as failing.(It's probably not ideal that the database has both
success,failedandcancelledcolumns, I'm not sure how to interpret them when more than one istrue!)