Repository navigation
Phase 0: land and harden v2, repoint everything off v1 #194
Description
Activity
Progress 2026-10-07 evening:
- feat(cf): Cloudflare-native control plane (v2), wire-compatible with v1 #182 and feat(hpc): cluster parity on v2 — Alpine on R2, ADRs 0007/0008, cutover plan #198 merged;
cf/is onmain. First real batch: one cluster claiming from v2 #178 can close. - Worker hardened and deployed (chore(cf): production hardening — reset off, CORS pinned, task-log guards #199):
ALLOW_RESET=false, CORS pinned to cmgd.cancerdatasci.org + localhost, task-log guards (400 on run_name null, NUL stripped), POST /task-logs auth-exempt like /runs/{run}/event. - Dashboard reads v2 (feat(frontend): dashboard reads the v2 Worker (#191) #200, deployed): 501 panels say 'not available yet', sign-in/write actions hidden until Decide the auth model for the dashboard and for operators #176, Runs page fixed (/runs/ → /runs).
- Fleet-health cron repointed to v2 (in place on onclappc02; tolerates 501s). Next timer run 08:00 MDT.
- Pipeline 2.2.2 (api_url + weblog.url → v2, task-log upload guard) in progress on release/2.2.x.
- curation repo
build_manifest.pyrepoint: PR in progress. - Demo cohort registered on v2 via
add-cmd: ZhangM_2023 (24), AsnicarF_2017 (24), WibowoMC_2021 (22) → 70 pending → 3 runs submitted on Alpine (33512907-9). Alpine daemon had died with the login node (tmux gone since ~13:00Z); restarted. Alpine PM reservation blocks acpu until 2026-10-08 06:30 MDT, so runs start then. - GitHub var
NF_TELEMETRY_URLstays on v1 until the catalog service exists (Phase 1: readset ids and the Postgres catalog seam #195):submit-studyneeds/submissions.
- feat(cf): Cloudflare-native control plane (v2), wire-compatible with v1 #182 and feat(hpc): cluster parity on v2 — Alpine on R2, ADRs 0007/0008, cutover plan #198 merged;
- Pipeline 2.2.2 tagged on
release/2.2.x(api_url + weblog.url → v2; task-log upload skipped for ad-hoc runs; NUL stripped). Record PR release: 2.2.2 (telemetry URL → v2 Worker, task-log upload guard) curatedMetagenomicsNextflow#103 (merge with a merge commit so main keeps its unreleased HUMAnN work). v2cmgd_nextflow 2.2.1revision set to2.2.2via PATCH /workflows/6/revision; runs claimed from now on use it. The 3 runs already queued (33512907-9) were rendered with 0623be7 and still post task logs to v1. - ADRs: docs(adr): 0009 catalog/control-plane split; amend 0006; accept 0004 #201 merged (0009 new; 0006 amended; 0004 accepted; 0005 header; 0007 conforming impls; 0002/0003 notes).
- curation repo: Read processing status from the v2 telemetry Worker waldronlab/curatedMetagenomicDataCuration#172 open (build_manifest.py and nf_get_completed_run.R on v2). Left for Sean to merge (external org). Note: 'processed' there still means 'registered in telemetry', as it did against v1; the Studies view (Phase 3: metadata phase 1 — studies view in users' terms #197) derives status from completed-under-active-version instead.
- Decide Decide the auth model for the dashboard and for operators #176.
- Pipeline 2.2.2 tagged on
Anvil on v2 (#203, merged):
- Anvil runs
-profile anvil,r2: user-space Temurin 21 at$NF_TEL_DAEMON/jdk, Nextflow 25.10.8 (launcher re-fetched; the 23.x launcher can't boot 25.x),~/.nf_tel.r2from GSM, checkout moved off stalefeat/cluster-paritytomain, nf-client reinstalled. Rendered submit script passessbatch --test-only; a 10 s probe job started within ~2 min. Daemon restarted and polling v2. - Pipeline 2.2.3 (seandavi/curatedMetagenomicsNextflow
release/2.2.x, tag2.2.3): 2.2.2 had nor2profile (it only existed onmain, Phase 1: stable API hostname via CNAME (nf-telemetry.cancerdatasci.org) #80), so any<site>,r2run claimed under 2.2.2 would fail at config load. Backportedconf/profiles/r2.configunchanged. v2cmgd_nextflowrevision →2.2.3. Pulled on Anvil; on Alpine the asset's tags were fetched with git (no Java on the login node during PM). feat(frontend): add Requeue Dead-Letter admin action #103 (release/2.2.x → main) now carries 2.2.3 too. - The 3 queued Alpine runs (33512907-9) are unaffected: rendered with 0623be7, which has
r2.
- Anvil runs
- added a commit that references this issue
on Oct 8, 2026
Phase 0 of #193 (plan:
docs/plan-v2-cutover-and-metadata-phase1.md). Goal: v2 onmain, production-safe, the only thing clusters, pipeline and dashboard talk to.main; rebase/mergefeat/cluster-parity. Close First real batch: one cluster claiming from v2 #178.ALLOW_RESET="false",CORS_ORIGINS= dashboard origins, test assertingPOST /admin/reset→ 403. Rotate operator token if not done.params.api_url→ v2 origin;afterScripttask-log upload no-op whenrun_nameis null; strip\x00. Release patch 2.2.2 (ADR-0004: revision bump, no new epoch);POST /workflows/{pk}/revisionon v2.NF_TELEMETRY_URL(after the catalog service exists, see Phase 1: readset ids and the Postgres catalog seam #195); curation repobuild_manifest.py; delete/repointnf_get_completed_run.R.Exit:
cmgd.cancerdatasci.orgshows v2 data; nothing referencesnf-telemetry.cancerdatasci.org/apifor dispatch; v1 container read-only.