Cylc Project VC 5? June 2020
(Meeting notes in italics)
Admin
Cylc-7
- 7.8.6 and 7.9.1 released (bugs: cycle point format conversion; spurious
warnings on restart)
Cylc-8.0a2
- Need to make the release before platforms changes go in
- Agreed last meeting to wait for data-store push on delta updates, but those
are still not in
- Relitigate that decision, or wait just a bit longer?
- Release 8.0a2 this month
- alpha-3 release should be SoD and n=1 window
Cylc-flow now using GH Actions for CI
Cylc 8 MVP
We are probably at least a month behind schedule, partly due to the pandemic,
partly unexpected difficulties.
Spawn on Demand
HO
- Simplification on realizing I can avoid keeping finished tasks to prevent
conditional reflow, without any additional DB access:
- No need for “parents-finished” housekeeping of finished tasks
- No need for spawn-on-completion, which also requires “parents-finished”
housekeeping of additional waiting tasks generated
- (There is still potential for parents-finished housekeeping of waiting
tasks downstream of AND triggers, but let’s consider nicer ways of doing it
later - for now, at worst it remains a niche case for suicide triggers -
much less than in SoS)
- Still updating the document damn it, almost done. Some details are difficult
to nut out because there isn’t an obvious right answer - we just have to
decide which way to go and document the reasons and consequences … but it
is all better than SoS! Example: should we treat handled and not-handled
failed tasks differently.
Current status:
- All supported functionality works (all test passing on CI)
- All advertised advantages have proved to be true
- New functionality seems to work nicely: trigger with and without reflow, flow
merge, flow stop
- we can refine new functionality later
Decided:
- DM: Keep agree fail handling for now
- Need a follow-up meeting with DM and OS to go through the details
Back-end: delta updates
DS
- a few changes since last month, but mainly just waiting on HO review (sorry)
- (OS and BK have had a good poke around - what’s the verdict?)
Ready to be merged
UI
BK
(UK team)
- TP: platforms changes very close - 1 week? test batter has been trouble; may
not be fully functional at first merge, e.g. cylc-7 compat
- RD: isodatetime efficiency changes (OS: lru_cache; then make data classes
hashable and therefore cacheable)
- MH stripping out
cylc submit
and cylc jobscript
and tests that use
them
- key generation changes are good to go (private keys don’t travel now)
- next: suite installation changes
AOB?
- BOM: will ask JR if she will join us soon
Next Meeting?
- TP to arrange for early July?
- July 7