Repository navigation
Persist ngmix epoch-cut counts; tile cleaning no longer needs module logs - #955
Open
cailmdaley wants to merge 4 commits into
Open
cailmdaley wants to merge 4 commits into
cailmdaley wants to merge 4 commits into
Conversation
Capture each ngmix summary before temp cleanup, aggregate the full chunk set at make-cat, and publish the manifest beside final_cat. Refuse reclamation without validated durable counts. This changes completeness params across stages and the make-cat/clean shell fingerprints; deploy between campaigns, not as an unreviewed resume. Co-Authored-By: GPT-6.1 Sol via Claude Opus 5.5 <noreply@anthropic.com>
Keep clean_tiles enabled by default, remove the stale false-default comment, and document the ngmix summary prerequisite, persistent schema, backfill procedure and campaign-boundary deployment. Co-Authored-By: GPT-6.1 Sol via Claude Opus 5.5 <noreply@anthropic.com>
Use frozen run defaults, require persisted final_cat and complete successful chunk evidence, publish counts atomically, then reuse clean_tile.reclaim. Dry-run uses the same whitelist traversal and never follows links. smk-b01 dry-run: 25 tiles, 53,290,513,588 logical bytes (49.631 GiB), no skips or refusals; nothing written or deleted. Co-Authored-By: GPT-6.1 Sol via Claude Opus 5.5 <noreply@anthropic.com>
…kfill Cover missing/ambiguous summaries, full chunk coverage, zero counts, duplicate log copies, stable manifest mtimes, cleanup refusal, read-only byte accounting, publication-before-cleaning and idempotent retries. Include the exact PR #922 producer line. Container verification after restacking: 199 passed, 1 skipped across targeted, structural and ngmix producer suites; ASTRA validation passes. Co-Authored-By: GPT-6.1 Sol via Claude Opus 5.5 <noreply@anthropic.com>
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Tile cleaning can reclaim scratch without losing ngmix's epoch-cut counts.
This PR is stacked on #922 (
feat/masked-pixels), which writes theepoch cuts:summaries.Production currently sets
clean_tiles: falseonly to keep those log lines, leaving roughly 2.1 GB per finished tile on/scratch(about 50 GiB per 25-tile batch).Changes
tile_make_cat.jsonand publish it beside the persistentfinal_cat-<tile>.fits.Its
epoch_cutsrecord holds the schema version, chunk count, per-chunk counts and tile totals forconsidered,masked_fraction,central_veto,failedandobjects_emptied.clean_tileon a validated persistent manifest; missing, malformed or ambiguous summaries refuse cleanup rather than becoming zeros.workflow/scripts/backfill_epoch_cuts.pyfor idle existing campaigns, with a read-only dry run and atomic publication before optional cleanup.These are chunk/tile diagnostics, including rejected epochs and objects that never reach the final catalogue.
A manifest preserves that scope and the complete chunk coverage without repeating tile totals in catalogue columns or changing the catalogue schema.
No scientific cuts or catalogue values change.
Validation
The earlier smk-b01 dry run covered 25 finished tiles (400 chunks) and would free 49.6 GiB; nothing was written or deleted.
It counted 3,110,928 considered epochs, 131,428 masked-fraction rejections (4.2%), 65,427 central vetoes (2.1%), 3 failures and 10,257 objects emptied of epochs.
This is consistent with the single-tile check on 202.301 (2.0% central veto; 6.1% total rejection).
After restacking, the targeted, structural and ngmix producer tests pass in
shapepipe-develop-240b37e4.sifwithsrc/shadowed throughPYTHONPATH: 199 passed, 1 skipped.The parser regression includes the exact #922 line:
epoch cuts: considered=8181 masked_fraction=235 central_veto=271 failed=0 objects_emptied=16.ASTRA validation and
git diff --checkpass.Deployment
Deploy at a campaign boundary.
The shared helper hash changes completeness parameters across stages, and the make-cat/clean outputs and shell commands change, so resuming an existing campaign can trigger reruns.
Use the standalone backfill for existing campaigns instead of resuming their DAG against this code; inspect
--dry-runfirst.The real backfill has not been run.
Usage and the persistent schema are documented in
workflow/README.md.GPT-6.1 Sol via Claude Opus on behalf of Cail.
🤖 Generated with Claude Code