Skip to content

24h full-campaign summary (2026-08-23)

One-round live re-verification of all 24 workflow repos against the latest engine main (post-v0.14.1) on tx-ubuntu + bioinfo-wsx. Every verdict below is a REAL CLI run, not a dry-run — checkpoints were cleared to force real execution (see the checkpoint no-op lesson in the failure catalog).

The matrix (24/24)

repo coverage this round verdict evidence
ampliseq DADA2 full chain + SBDI-GTDB taxonomy 16/0/14 exit 0 site 1bfad3c
atacseq default bwa arm full chain 29/0 exit 0 site 85b6d1f
auto-sra-rnaseq-pipeline 4-sample 2v2 real SRA, STAR full-GRCh38 + DESeq2 exit 0 (mini21) site 5f37dbd
bgcflow prokka + antismash 7 full DB set + gtdb 14/0/13 exit 0 site 5d6d1bf
chipseq bwa→macs3→FRiP→homer full chain 143/0/11 exit 0 site def8453
circrna 4 callers real; find_circ planted-BSJ hit 9/0 exit 0 site 39ee9f3
clindet RNA port 26 rules (arriba BSJ detected) 26 real/0 failed site f1aae9c
eager nfcore/eager:2.5.3 single-container chain 22/0/6 exit 0 site f67f54a
enrichment GSEApy + GREAT + LOLA (Fisher real) 43/12/2 + 6/51/0 site 87b5fdb
fetchngs 3 real ENA ids, md5-guard verified 7/0/9 exit 0 site fe8a5c9
genome-tracks full-line-sc: sc + bulk + ucsc_hub exit 0 (5-fix chain) site f458708 + PR #2
mag 295 rules; 5 binners + QUAST/BUSCO + ALE 224/32/4 → resume green site 4fcde83
methylseq bismark default chain 19/0/3 exit 0 site 2fe1170
mixscape full chain 9/9 exit 0 site 9d5b90e
nanoseq minimap2+medaka default chain 29/0 exit 0 site 0da8165
rnaseq STAR + tximport + DESeq2 (heavy) 68 real/0 failed site 40ec98d
rnaseq-star-deseq2 STAR+GeneCounts + live biomaRt 3-step 56/0/48 exit 0 site 31212a0
sarek GATK haplotypecaller + vep germline 165/0 + 1/38/0 site 97f25b2
scrnaseq cellranger S1+S2 real + cellbender 22/0/10 exit 0 site 6466481
snparcher gatk intervals + fastq default 21/21 exit 0 site f458708
tcasia rmats + spladder + MAJIQ/voila 24/0/10 exit 0 site 31f0d2e
unsupervised full chain 27/27 exit 0 site 9d5b90e
varlociraptor vg giraffe pangenome chain 17 + 38/50/0 site f9f7045
viralrecon illumina-amplicon chain, 27-env build 7+76 exit 0 site 6d3ac37

Repo fixes pushed this round

repo commit fix
genome-tracks 2ae877e..fdfeec6 sc-branch runtime fixes ×5 (index fixtures, setuptools<81, header-only BAM, symlink idempotence, plot gating)
genome-tracks 80adcbc test instance list follows the mini fixture (PR #2 CI)
mag 2554bde build_ale.sh GCC≥14 legacy-C compatibility
fetchngs 0a118df id.txt concurrency race (shared workdir)
fetchngs 9a9f09f wget -c corruption → md5-first idempotence guard
enrichment c7f194c LOLA fixture index filename header
varlociraptor 6fda195 delly download ghfast.top mirror fallback

Engine-side work

Merged: #152 (CLI↔docs consistency ×6), #154 (macOS .app icon + launcher + ad-hoc codesign), #155 (Linux deb/rpm/AppImage menu entry + icon), #156 (env-cache hit re-verifies on disk; vanished envs invalidate + rebuild). Filed: #159 (env identity — same-name cross-workflow collisions), #162 (singularity URI→IMG %3A naming).

Web lane

149 (dev-mode role derivation fix + serve.md correction) + #151

(deployment + 3-role simulation matrix, 450 lines: RBAC, ownership isolation, SQLite restart persistence, base-path). Local gate green, both merged.

Methodology lessons (in the failure catalog)

Checkpoint no-op trap (force real runs), honest coverage labeling (chr21-subset reference, gtdbtk=false, DRAFT constraints), China-network DB pre-staging pattern (sha256-accepted), conda post-link proxy pathologies (two endgames), docker storage exhaustion, env-cache family (three dimensions).

Resource evidence (honest: peak RSS not recorded — the engine does

not emit per-rule peak RSS; wall-clock is run start→end)

tx-ubuntu (4 vCPU / 3.7GB, clamp constant): nanoseq 13:55, circrna 3:24, chipseq 31:45 (docker --memory 3723M + 1881 clamps), methylseq 16:00 (bismark 12→4 threads), rnaseq-star-deseq2 3:41 (+biomart env ~1.5h proxy battle), eager 0:56, scrna-seq 11:14 (cellranger 72GB declared → clamped, runs anyway), sarek ~40min, rnaseq ~1.1h, ampliseq 2:36 (+dada2 env ~2h). bioinfo-wsx (64c/1.4TB, ZERO clamps): bgcflow ~1.4h (DB staging dominated), mag ~2h (26 envs ~50min), tcasia ~30min (rmats 3s + spladder 339s parallel), enrichment ~1.8h (env battle), varlociraptor ~15min, auto-sra mini21 (STAR ~6min/sample, full-index build 48 threads ~32min — the heaviest scheduling point).

Unified observations: (1) tx-ubuntu clamp + "run alone" serialization is CORRECT degradation, cost = wall-clock; (2) bioinfo-wsx zero clamps but network is the bottleneck — env/DB downloads are 40-60% of total wall time; (3) conservative -j (2 / 4-8) was already throughput- saturated by env+network constraints. Engine candidate (post-campaign): per-rule peak RSS in the run report.