Date
Attendees
Fabrice Jammes, Vaikunth Thukral, Serge Monkewitz, Andy Salnikov, Fritz Mueller, Unknown User (npease), Andy Hanushevsky, Jacek Becla
Discussion items
Qserv software on in2p3 cluster
- unify how we install/manage qserv @in2p3, see newly opened DM-3199
Large scale tests
- debugging, latest problems look like a bug in log4cxx
Ingest
- almost done, dealing with complexities of aliasing in afw table code
mysql-based kv interface
- flat key/value structure
- children - implement through a separate table that keeps info about relationships
- ok to keep key length under 512 bytes
managing credential files for integration tests
- config files in home dir? AndyS not convinced. Need to find "better" solution
Integration tests and CI
- already have dockr file that runs single-node, reuse that and develop a script that could be used by CI system
- later support multi-node tests (harder to do, but useful to have)
Planning discussion
- distributed loader
- Can't test without infrastructure (fast shared fs). Discuss w/NCSA at Bremerton
- do exploration and plausible design, postpone implementation until we have hardware to test it on
- do we need epic for loading images? Discuss w/K-T
- non-partitioned tables
- maybe duplicate most frequently columns, use shared fs for the rest
- simplest would be to duplicate all non-partitioned tables
- revisit if we could duplicate all non-partitioned tables on all nodes
- need infrastructure (scalable shared fs). Discuss w/NCSA at Bremerton
- BTW, we still have to support L3 shared fs
- dbserv
- POST related story is for async queries, move to S16
- near neighbor optimizations
- neighbors-of-neighbors idea won't work (e.g. won't find neighbor if it is far away and there are no objects in between that could be used for chaining)
- worth exploring how to optimize, not high priority
- explore through XSWAP