How should decoding keys be handled?
Currently all Run 3 grid jobs are cloning file-content-metadata
in every job. Fortunately the clone is from CVMFS but as the repository grows this is going to become increasingly unsustainable.
The write_decoding_keys_to_git
option avoids this however it always prints this warning and then continues to run the job regardless:
# WARNING: you have specified an output file, but also blocked any writing of decoding keys -- as a result, you will likely not be able to decode the data written
I see there was discussion around this in !3842 (merged) but it's still not clear to me how this is supposed to be handled correctly offline.
If I set write_decoding_keys_to_git: false
in all grid jobs then we avoid the clone but have a misleading warning and potentially unusable output data. The only option for enforcing that the keys are already in CVMFS can only be set from user code and isn't configurable by LHCbDIRAC.
Related, is there any documentation to point people to for how they should add extra keys to file-content-metadata
?