Infrastructure
------------------------------------------------------------------------------------------------------------
Complete migration of HC hosts from quattor to AI (VMs and puppetising)
Get rid of NFS both for logs and SW
Agree on decoupling of test and production infrastructure under condition there is no big latency between testing and deployment.
CMS is less critical regarding latency, since there is no site blacklisting based on HC
There was a discussion that 3 layer infrastructure is needed, development, testing and production. It was decided to start with two layers and to see later whether we have to go for 3 layers.
What concerns policy for emergency intervention, experiment experts should have access for these machines. However, SDC should be informed about manual interventions to the production hosts.
It was agreed that SDC would take care about infrastructure. Puppetising for submission hosts will be done with joined effort between SDC and experiment teams
Support workflow
--------------------------------------------------------------------------------------------------------------
Experiments would like to keep the problem reporting in the way it is currently organized (up to them how it is done). In case experiment experts understand that the problem should be escalated to SDC, the corresponding GGUS HC support unit will be used.
Repository
------------------------------------------------------------------------------------------------------------
Split the repository in the following way : core, ATLAS, CMS, LHCb...
ATLAS team agreed that it makes sense. Would be even better if we can split ATLAS part according to payloads, should be investigated. Files for building RPMs should be also included in the repository.
Question:
Are there cases when several different versions are installed ?
Yes it happens for GANGA, but not for HC. Different versions of dq2 and panda clients? They are coupled with GANGA version
Pablo volunteered to help with configuration and building rpms.
How code is committed and released will be decided offline. Luca kindly agreed to follow it up with ATLAS colleagues.
HC evolution
Julia suggested to assess current implementation of the HC components in order to understand possible improvements for scalability, sustainability and new use-cases.
Johannes thinks that there is a working system in place and he is confident about scalability.
There was a discussion whether GANGA should stay a mandatory component of the HC or it can be just one of the optional backends. The whole discussion during the meeting could not bring to any conclusion and required deeper investigation
Next meetings
------------------------------------------------------------------------------------------------------------
It was decided to get in touch via mail. Next f2f meeting is planned during ATLAS sw week in June