FTS Steering meeting

Europe/Zurich
31/S-023 (CERN)

31/S-023

CERN

22
Show room on map
Remote
- Catalin Condurache
- Eric Vaandering
- Joaquin Bogado
- Martin Barisits
- Stephane Jezequel
- Chris Haen

Local
Andrea, Maria, Eddie, Dirk, Mario, Martin, Cedric, Michael Davis,
Giuseppe, Alessandro
 
Minutes: Eddie
-----------------------
 
Discussion on news and plans for 2018
 
Giuseppe: Tape migration will increase the db load? Back-pressure is already in Castor
we need to avoid clashes with the one implemented in FTS
 
Andrea: there will be more queries to retrieve the file to be monitored

Martin: FTS-REST-API package will it be python3? yes.
 
Discussion on Scalability and Scheduler improvements
 
Martin: would be good to compare also the performances of the DB when doing stress testing with different schema versions
 
All: we agreed to perform new tests including also staging and check DB performances as well.
 
ATLAS requirements discussion
 
Mario/Cedric: fix to avoid duplicates transfers submitted  it's important.
—> Will come in 3.9  (~April) Need to agree on error.  (+
transfer-duplicate for Rucio monitoring)

Cedric: Are you sure it's not a server picking up work from another
server and resubmitting?
Andrea: describing the issue with RAL - > 3.8.3 maybe fixed this.

Ale: Analytics studies-> intelligence. automatic selection of the best
strategy. it is not that it is not working and we need to fix. it's to have
the intelligence. We need to work on this together
—> Analytics person will join FTS team (not full-time on FTS) (starting
April)

Mario: All we need  to know is why is there a queue of transfers:
—> FTS team has a plan to assess the performance and will provide
updates on the developments/improvements. Overall goal is to converge to
a single instance per VO.


Ale: Which is the best deployment model, 1 instance? 3? we need to agree on deployment strategies
Joaquin how to measure the impact of config changes : like the limit of the storage
can have a huge impact
Andrea: Changes on config limits are not frequent and they are stored on our audit logs
 
Mario: auto session reuse discussion, currently not usable in prod. maybe we can
do what we will do in data carousel. Send a huge bulk to us (--> group
by dataset).
Andrea: 200 in bulk from FTS side (configurable).

Mario: OpenId in Rucio will be implemented

Martin: QoS -> are you doing QoS transitions to the storage?
Andrea: this is CDMI. Integrating CDMI in FTS.


Andrea: we can schedule a new meeting before or after summer. At some
point after 3.9

CMS/Eric: tape checking is the highest priority. Andrea: Once it is
implemented for CTA, it will be straightforward to contact srm by
changing the protocol. Eric: is there a ticket we can follow? Andrea: We
will create a document with a study with design of the system and
analysing what needs to be changed and then we will create tickets.
There are minutes attached to this event. Show them.