2021-08-27 QC meeting notes

Investigation update from MolSSI on optimization stalls, COMPLETE ↔︎ INCOMPLETE cases

Ben

• BP: can submit an optimization, creates a task and an optimizationresult

  • when task is completed, gets deleted

  • but if you resubmit, a task will get created even if the optimizationresult is complete

• BP: have about 40,000 tasks that are associated with COMPLETE optimizations

  • will clear these from the task queue

  • DD: will add guardrail to qca-ds-sub lifecycle automation to only call compute on non-submitted optimizations

    • check object_map for each entry, if doesn’t exist, submit it

• BP: other issue of tasks that say they are running, but manager is not active

  • only about 300 cases of this at the moment

  • can happen when manager starts up and is immediately killed

    • manager will ask for tasks, they get assigned by the server, but manager goes down before it receives them

User questions, issues, new optimizations

• PB: MP2 (#217) insufficient progress; need 100% torsiondrives in

  • DD: set compute tag to openff-mp2; will spin up workers on PRP specific to this

Infrastructure needs/advances

Simon

• SB: QCElemental torsiondrive models in; thank you Ben and Lori!

  • as a follow-up, need the models added to QCElemental removed from QCFractal, since they are now duplicated

  • now QCEngine torsiondrive procedure unblocked

• BP: looking to change how nested models are in QCSchema, so these may get adjusted in the future

  • QCElemental#264 for that discussion

• SB: fastapi author just put out sqlmodels, perhaps worth a look to see if that suits needs for next QCFractal

  • BP: there’s also a library called marshmallow that has some nice features, but we have some esoteric needs in some places that not sure that satisfies either

• DD: is there anything else needed from Ben or me on QCEngine#305?

  • BP: is that blocked by QCElemental#264?

  • SB: once new QCElemental conda-forge package up, can update QCEngine#305 with use of the QCEl models; then ready for final review from Ben

MD snapshots for many molecules coming

Simon

• John and Peter Eastman planning to calculate a bunch of single point calculations on a collection of conformers from MD

  • would be in the millions of calculations

  • BP: next branch could prioritize depending on timeline