2020-09-25 QCA Submission Meeting notes

QCFractal release, QCArchive upgrade

Ben

QCFractal release on track for next week

• modify and regenerate tasks

• maybe have the ability to set defunct status

• changes the modify_tasks
  

• Considering upgrading Postgres 10 → 13

Reworking fractal interface is going to be a challenging task

• TG: We need to change some data where the status is incomplete but the data exists

• BP: The new modify_tasks will create new tasks, and this will allow the job to be restarted. The problem could be the underlying task was already deleted, and couldn’t restart.

• We should use the create task functionality on a debugging basis only

There is some talk about QCSchema changes

Entry creation

Trevor

• TG: Entries aren’t created until tasks are created

  • dataset doesn’t show up until you call compute

• BP: One of many things that needs a rework

Protein Fragments

David

• Need to go through constraints with fine-tooth comb

Enamine REAL

Trevor

• Working on performance optimizations for QCSubmit to handle a dataset of this size

PhAlkEthOH dataset

Trevor

• Small dataset that was used for initial fit

  • finally getting into QCArchive

  • useful sandbox to test typing schemes

  • ~40k conformers (different from the conformers used in the original dataset)

STANDARDS-based versioning

Trevor

• Coupled with QCSubmit development

QCEngine Release

Ben

• QCEngine release coinciding with Fractal next week

QCElemental Identifiers

Trevor, Josh

• TG: it is implemented, present

• JH: We pushed for using Identifiers for CHMILES early on, but because this isn’t necessarily stable information if the connectivity changes over the course of the optimization

  • conclusion: we didn’t want to put information that was searchable and meant to reflect the molecule as is but could very well be inaccurate

  • the compromise was to put this information into extras since that is free-form and not considered to be as much of a source of truth

• TG: is it possible to guess connectivity at every optimization step?

  • or do we have a postprocessing step at the end that guesses connectivity after an optimization?

  • The drawback to the current implementation in QCElemental is that it does not perceive bond order.

PCM calculations

Josh

• Can do PCM calculations, need to restrict symmetry

  • latest conda build for Mac!

• Just need to update QCSubmit with corresponding options

Data Access layer in service of future advancements

David, discussion

• DD: value streams, upcoming needs a data access layer would need to meet

  • SB: really agree with the approach; partial to fewer packages

  • TG: agree, will be working on QCSubmit more to include extraction bits, make sure these line up with input standards

  • JH: agree, and like this approach; have some minimal data access bits currently to support bespoke-workflow

  • TG: have a lot of existing functions that I can put into QCSubmit

QCFractal managers

TG: Using a pre-empt queue causes a lot of jobs to be killed, which are sent back as ERRORS, and need to be restarted manually. Can this be detected and have those jobs automatically restarted?

DD & BP: This will be challenging as the exceptions are adapter specific

TG: Short term solution is a loop that scans the dataset from a client and restarting tasks with specific exception strings in the output