In rare cases, the filesystem appears to send two events for file creation.
The file is scheduled twice, evident in the logs by INFO [Scheduler] ▶ Sending <{file}> to the classifier event timestamps are identical to the millisecond
Most often this results in hornet completing the process for the first time before the second time is classified. The file is migrated to DIRAC intact, but a classifier error is then logged to slack:
CRIT [Classifier] ▶ [classifier] file <{file_path}/{file}> does not exist
In rarer instances, both versions are classified, and two moves are attempted; the first move will succeed, the second move will fail, but will remove the completed file from warm and the data will be lost:
ERRO [Copy] ▶ File copy failed! ({file_path}/{file} -> {warm_path}/{file}) [open {file_path}/{file}: no such file or directory]
Lesson for future upgrade (in dragonfly migration):
gracefully handle double-scheduling (probably still slack warning so we know it's happening)
In rare cases, the filesystem appears to send two events for file creation.
INFO [Scheduler] ▶ Sending <{file}> to the classifierevent timestamps are identical to the millisecondCRIT [Classifier] ▶ [classifier] file <{file_path}/{file}> does not existERRO [Copy] ▶ File copy failed! ({file_path}/{file} -> {warm_path}/{file}) [open {file_path}/{file}: no such file or directory]Lesson for future upgrade (in dragonfly migration):