Update Datasets

Update Datasets

Build and publish Bactopia's version-pinned datasets to Cloudflare R2.

bactopia datasets downloads bundles from https://datasets.bactopia.com/datasets/. Most bundles are shared across releases, but a few are pinned to the pipeline version. The only version-pinned dataset today is AMRFinder+ (conf/params.config: amrfinderplus_url = ".../datasets/v${params.bactopia_version}/amrfinderplus.tar.gz"), so this skill implements that path concretely; the name stays generic for future version-pinned datasets.

The publish version is versions.yml's bactopia value, and the R2 layout is datasets/v<version>/amrfinderplus.tar.gz. The tool version that builds the database must equal the version the pipeline pins (an outdated binary can produce a database the pinned binary cannot load), so the skill hard-gates on the amrfinderplus module being current before building. This skill is the sole owner of the build recipe — the old modules/amrfinderplus/update/ module that carried it was removed.

Steps

1. Currency gate (hard stop). Confirm the module tool is the latest bioconda release before building:

   bash .agents/skills/update-module/scripts/run-bactopia-update.sh \
       --bactopia-path /home/rpetit3/repos/bactopia/bactopia --module amrfinderplus --json --silent

   Parse the entry whose tool == "ncbi-amrfinderplus".

   • needs_update == true -> STOP. Tell the user to run /update-module (bumps modules/amrfinderplus/run/module.config), then re-run this skill. Do not build against a stale version.
   • latest_version == null (API failure) -> STOP and report the failure.
   • needs_update == false -> proceed.

2. Build. Rebuild the database in the module's pinned container:

   bash .agents/skills/update-datasets/scripts/build-amrfinderplus-db.sh \
       --bactopia-path /home/rpetit3/repos/bactopia/bactopia

   Add --runtime singularity if the user asks or docker is unavailable. This downloads the latest NCBI database and produces the tarball; expect it to take several minutes and hundreds of MB. The script reads the container image from modules/amrfinderplus/run/module.config, so version bumps flow automatically.

3. Report the build. Show the script's summary fields (tarball path, sha256, tool version, database version). Read the publish version:

   awk '/^bactopia:/{print $2}' /home/rpetit3/repos/bactopia/bactopia/versions.yml

   The R2 key is datasets/v<version>/amrfinderplus.tar.gz.

4. Upload (gated behind explicit confirmation). The rclone destination base is bactopia-r2:bactopia (remote bactopia-r2, bucket bactopia); $BACTOPIA_R2_DEST overrides it if set. The public URL https://datasets.bactopia.com/datasets/... maps to bactopia-r2:bactopia/datasets/..., so the full key is <base>/datasets/v<version>/amrfinderplus.tar.gz.

   • Verify reachability: rclone lsd bactopia-r2:bactopia. On failure, report it and fall back to printing the manual command below — do not upload.
   • Show the exact command and ask for confirmation before running it:

     rclone copyto "<tarball>" "bactopia-r2:bactopia/datasets/v<version>/amrfinderplus.tar.gz" --s3-no-check-bucket --progress

   • --s3-no-check-bucket is required: R2 API tokens cannot CreateBucket, which rclone otherwise attempts before the first upload and fails with a 403 AccessDenied. The bucket already exists, so skip the check.
   • Only on an explicit yes, run it. copyto overwrites an existing key, so re-publishing a version is idempotent.

5. Verify publication. After upload, confirm the public URL resolves:

   curl -s -o /dev/null -w '%{http_code}' -I "https://datasets.bactopia.com/datasets/v<version>/amrfinderplus.tar.gz"

   200 -> published (this is exactly /release-checklist check 15's sentinel). Report the HTTP code.