andycasey / astra / 34414741904

21%

main: 21%

push

github

Merge pull request #115 from andycasey/mini_survey_test

Merge mini_survey_test: DR21 bulk run preparation and hardening

101 commits (2026-05-01 to 2026-09-09), ~6,200 insertions across 45
files. The bulk of this work hardens the ASPCAP/FERRE pipeline against
hangs and timeouts at DR21 scale, adds the ARJL coadd pipeline, fixes
APOGEE flag and migration handling, and bumps paths/version for the
DR21 bulk run.

FERRE robustness and watchdog
- Resume the remainder of a FERRE execution after a hang, and track
  outstanding objects by name rather than by FERRE's own counter.
- Add max_t_communicate_first_result to allow a longer wait for the
  first result, addressing large dispatch times on big grids.
- Increase the grid load time allowance (real failures seen at 1000s)
  and check it before communication time.
- Kill when all threads exceed max_t_elapsed, plus a max sigma outlier
  case; default to max_t_communicate when the watchdog is drained.
- Retune watchdog thresholds and nprocs/nthreads from load-balancing
  tests (#144); apply a 0.25 fractional threshold when n < 10 samples.
- Catch errors in pre_process (#144), dropped or corrupted FERRE rows,
  and unset PBS or other OSError in the debugger.
- Add relocate_into so reprocessing places sub-groups in separate
  element directories; exclude hanging objects on list-mode resume.
- Fix n_processes logic so it cannot exceed max_workers on reprocess.
- Fix a missing import that was silently suppressing FERRE timeout
  flagging.

Flags and data quality
- Add flag_affected_by_timeout to the param and abundance stages so
  rows that are NaN only because of a batch timeout are tracked.
- Create distinct names for coarse flags and set them correctly in
  ASPCAP (#145).
- Implement APOGEE visit and star flags; take visit flags from
  rv_visit during migration, including objects with no RV (#118).
- Implement the m_h - fe_h > 0.1 flag (#137) and fix the ebv flag bit
  definition in the model (#126).
- Source... (continued)

305 of 921 new or added lines in 31 files covered. (33.12%)

7012 of 32762 relevant lines covered (21.4%)

0.21 hits per line