2020-08-14 QCA Submission Meeting notes

QCA conda deployment

• Pinning MM dependencies

• Create a separate prod env on the basis of harness we are running to avoid conda dependency conflicts between programs

  • OpenMM (off-tk and friends)

  • Psi4 (no off-tk?)

  • ANI (no off-tk?)

• Also, can we put a manager config somewhere (with creds substitutible)?

  • Perhaps write a script that clears old env, installs new env, pulls manager template, injects credentials?

lifecycle handling for compute.json

Josh

• should we decouple compute from molecule submission?

  • what’s the convention we could use?

  • would give us flexibility on adding compute after a submission

• David: given usage pattern we want for submission, need to then think how we can make automation handle this deterministically.

  • will report back on #131 proposed path forward for automation support

GHA restarting

Trevor

• Restarts perhaps not working for TorsionDrives based on #119; would expect to see ERRORs turn back into INCOMPLETEs, but we may need to wait until the queue is less full of running TDs

• David: error cycling last night failed because the comments were too big. Added truncation if error tracebacks more than 60,000 chars collectively

New datasets

David

• Disacharride submission

  • Josh: need Cerutti to add molecules

  • David: I will ping

• Protein Fragments Dataset V2

  • Josh: need Cerutti to answer clarification question; then ready to go

  • David: I will ping

• Add additional compute to OpenFF-Benchmark-Ligands

  • David: will chew on this approach of decoupling the molecule submission from the compute; need to make sure we can automate what we decide on doing

  • David: will report back on that PR proposed approach

QCSubmit questions

Trevor

• QCSubmit might benefit from some minimal usage docs and what programs, basis, methods it supports for specs

  • This is not necessarily straightforward, since QCSubmit relies reactively on what the tools it calls can support