Present: Brian Paul Bockelman, Dave Dykstra, David Crooks, Miguel Martinez Pedreira, Mischa Salle, Vincent Brillault
● Welcome and minutes from last meeting
- No comment on the minutes of the previous meeting
- Vincent reported that at the last GDB, a new WG was created, the "container" working group, which will take care of the uniformed deployment an isolation solution (e.g. singularity). As a result, this working group will now only focus on traceability
● Compute traceability: artefacts & challenges
Two axis to work on (potentially in parallel):
- Runtime traceability: finding more details (e.g user identifier) about a running job from artefact left by the pilot/VO framework (file name, log files, etc)
- Needed for a better challenge capable of validating the VO findings
- Actions on VOs: identify what is already available and can be collected
- Offer a possibility for sites to collect these logs (like FNAL for CMS):
- HTCondorCE has a new feature, which can produce audit logs if the jobs push backs the right information.
- Dave will share a command line that can be used to produce that data
- This does not cover all CEs and might only be a gap solution
- What should be required from VO?
- First estimation: Start/stop action, user unique identifier (opaque string)
- More debug information from VOs? (e.g. pilot id, job id for the pilot framework, etc)
- Any format needed?
- How to collect it from other CEs, for other VOs?
- Discussions to be started by email
● Storage traceability
- Brian presented a pilot using tokens to authenticate and authorize for storage (XROOTD) actions
- This is based on standard technology and existing work (e.g. auth0)
- As discussed at previous meetings, this approach is compatible with Alice current system, so a convergence could be possible on the long term
- As nobody from ATLAS or LHCB was present, there was no feedback from them.
- This will be presented at the next GDB, at which time it will be possible to collect interest from all participants
- Brian suggested to run a storage challenge
- This would be technically more difficult than the simple compute challenge currently discussed
- To be discussed over mail as what can be tested and how
● Actions & next meeting
- AOB:
- Dave noted that the name of the WG was now not in line with the mailing list and the website: Vincent will see what can be done.
- Actions:
- VOs: identify and report artifacts that can be used for identifying jobs/users
- Dave: share commands that can be run to populate the HTCondorCE logs from a running job
- Vincent start mail discussion on:
- How pilots can push user/job information to sites? What should be pushed?
- Ideas on how to do a Storage Challenge
- Given the lack of progress and activity, no meeting is schedule yet. When the discussion will have progressed enough by email, another meeting should be schedule via foodle