Date

Attendees

Notes from the previous meeting

Discussion items

DiscussedItemNotes
(tick)Qserv at USDF

Igor Gaponenko 

John Gates: 

(tick)Using qserv-ingest  at UkDF for ingesting Gaia_EDR3

Igor Gaponenko: present a context

  • UkDF colleagues have been trying to ingest a small-scale (1 TB, 1k chunks, single director table gaia_source) catalog into Qserv (in Kubernetes) using qserv-ingest.

  • The metadata file for the ingest: http://wsa.roe.ac.uk/gaiaFull/metadata.json

  • They have run into trouble with the ingest. It looks like the problems are starting with occasional timeouts (after 5 seconds) on POST requests sent to the Replication Controller or worker services. This results in aborting the transactions and the inability to start new ones.
  • Details can be found in this thread: https://lsstc.slack.com/archives/C2B70NXFD/p1683562658021569

Fabrice Jammes :

  • will contact the UkDF colleagues to see what help they might need with the problem
  • will try ingesting this catalog into IDF
(minus) Postponed due to lack of timeFile-based result delivery, Qserv lockups

Igor Gaponenko the context:

John Gates any news on the lockup during query cancellation?

(tick)(new problem) Alleged memory leak in the Replication Controller REST services

Igor Gaponenko the context:

  • the leak is happening when processing large monitoring objects (JSON) received from Qserv workers.
  • at least one REST service has been affected
  • the problem is seen while monitoring worker tasks when Qserv is processing Full-Sky catalogs (150k catalogs GAI_DR2  or wise).

Fritz Mueller suggested testing the life time of the relevant request process objects within qhttp 

Action items

  •