Skip to end of metadata
Go to start of metadata

Date

Attendees

Goals

Discussion items

TimeItemWhoNotes
Extending these meetings to include the deployment of the SV instance
  • Brought this up at DMLT call
  • Confirmed: we will use this meeting to cover both the integration/PDAC and science validation/SV/Kubernetes-Commons instance of the Science Platform. I will think about whether a name change is needed (e.g. "LSP Integration and Deployment" meetings?).
 Kubernetes Commons
  • Configured and set up. Adam has "kicked the tires". Persistent storage has been set up and configured. Jellybean home directories are available. Anyone with NCSA credentials can log in. Jellybean home is available from lsst-dev. Includes a Docker registry. The registry should already be accessible from the PDAC nodes.
  • Did not end up using Samba for the home directory setup. Just plain NFS.
  • Not planning to bring up Qserv during this cycle.
  • Brian Van Klaveren interested in bringing up imgserv on the commons, able to see Butler repositories on /datasets. Can also bring up dbserv looking at a database on lsst10.

PDAC Kubernetes status
  • Andrew Loftus has been working on this. All of the node rework looks like it will be done by next week. Will probably federate the cluster and then roll out the configuration to the rest of the DB nodes. See IHS-940 - Getting issue details... STATUS .
  • Existing Docker-deployed DAX services will continue to run for now. Will announce when it is time to switch to the Kubernetes-deployed services.
  • Note that the Kubernetes Commons registry is available for PDAC use.
  • Loi Ly has a pending action item of responding to progress made by Andrew Loftus.
  • Michelle Butler reports that staff have suggested using the same head node as for the Kubernetes Commons. Currently it's a bare metal machine. Apparently a node could be shared. Discussion leads to decision that we'll spin up a new VM for this instead in the LSST VM infrastructure.
    • Fritz Mueller will make a ticket with the requirements for a Kubernetes head node VM for PDAC - DM-14146 - Getting issue details... STATUS
  • Discussion of how to provide Jellybean / Notebook Aspect services in PDAC: minimum requirements are for a pod for the JupyterHub service and a pod per user. Will not need a storage pod in PDAC if the same approach to GPFS/NFS integration is taken that was used in the Kubernetes Commons. Additional pods needed for logging and for container image pre-loading.
    • Tentative decision: Use SUIT load-balancer node for now for the necessary pods; request a VM if this is found not to be adequate

Longer-term hardware requirements
  • Looks like it may be desirable to move the DAX services to hardware with SSD storage. Lots of files are written as part of serving requests.
  • Discussion of whether a more in-memory model can be adopted (Kian-Tat Lim and Brian Van Klavern).
  • Same issue arises for the SUIT services.
  • Michelle Butler Next provisioning round is in October/November. Can this wait? Fritz Mueller: Can we do this sooner for dax01? Unknown User (xiuqin): We can wait the 6 months. Brian Van Klavern: for DAX nodes, favor speed over capacity.
    • Fritz Mueller: (step 1) Submit a ticket requesting SSD storage for dax01

Integration of PDAC with authentication, and exposure to public Internet
  • Group membership micro-service is now up and running and being used by KC's Jellybean
  • Have not yet all signed off on group naming conventions - see ITRFC-12 - Getting issue details... STATUS
  • Will discuss the overall plan for getting PDAC services out from behind the VPN ( a: security signoff; and b: meaningful login and authorization) two weeks from today at  PDAC meeting in preparation for May DMLT and F18 cycle planning.

Action items

  •