at IN2P3 cluster: need to migrate metadata because we switched from zookeeper to mysql
what happens when we lose CSS info
can't reproduce based on data
rely on backup / replicas
will need tools to check consistency data < – > css
Docker
problems with starting cmsd inside docker, AndyH provided fix
docker work - deliverable / milestone: recipe explaining how to run integration tests inside docker on open stack using vagrant + containers
authorization
NCSA team asked Brian for standalone dax service for testing authentication
dax for sui
need to enable ports
need to load / make available some consistent set of data (images + catalog)
image search
Serge checking out memsql - in memory, supports spherical geometry. But not free, startup. Unclear what the cost is
create story for that work, attach to epic DM-1648
greenplum
open source now, worth checking out
need good benchmark to really compare
can use the queries we run for large scale tests this summer
also, might want to checkout hsqldb (research version of voltdb)
image archive
need to plan epics / metaepics
writing images from AP ... local disk? shared fs? object store? Prefer shared fs over object store. Probably don't want to serve images to users from the location where AP writes. So writing to local disk might be best
when images saved, need to save provenance. It is either via butler "put", or separate step. Better via butler, it is more atomic that way
in butler put, it should be possible to turn off writing metadata, we have a usecase where someone might want to write just images say on a laptop
need to understand how to hook provenance to butler
butler probably does not need to add to provenance, hopefully not
querying provenance: preexisting "query templates", user fills bits and pieces, we expand it to internal queries that know how to query provenance. That can be implemented through plugin to butler
usecase: we push alerts, but fail to push images and/or catalog data. Later we reprocess that thing that failed. Obey observing order, also, have to remember processing order! So, provenance needs to keep track of processing order for AP (this is not the case for DRP, as Jim Bosch mentioned in the thread on community.lsst.org)
archive manager
access via buter, webserv, ingestion daemon that ingests data from pipeline. Maybe exposed to users for ingesting images
also, see picture of whiteboard related to that discussion: