Release from dev: Annotate and Home tabs rebuilt on the hypeR-native client, staging tooling, React app removed #154

Activity

1. Cvicnaire

   added

   enhancementNew feature or requestNew feature or request

   readyStatus: next step is stated; anyone could pick it up.Status: next step is stated; anyone could pick it up.

   on Oct 7, 2026
2. Cvicnaire commented on Oct 13, 2026

   

   Cvicnaire

   on Oct 13, 2026

   CollaboratorAuthor

   More actions

   Validation pass, 2026-10-13 against montilab: both checkouts at dev HEAD (server e02aff8, client 76a78c5), SIGREPO_IMAGE_TAG=dev, today's 12:04 deploy log for the pull and the API smoke test, the Shiny checks through the tunnel with chromote.

   Ticked

   • Publish montilab/sigrepo:dev when dependencies change on dev #102, Select the sigrepo image tag from SIGREPO_IMAGE_TAG #104: staging .env has SIGREPO_IMAGE_TAG=dev and today's deploy pulled montilab/sigrepo:dev; docker-compose-local.yml reads ${SIGREPO_IMAGE_TAG:-latest}; production's docker-compose-vm.yml pins :latest and the droplet runs it (b65374fddca7, built 2026-10-07). The dev build gate works as described: the four dev merges today ran the workflow but pushed nothing, and the tag is still the 2026-10-07 build.

   Already ticked, re-checked

   • Restyle the Shiny Home tab to match the retired React dashboard #153: Home shows 287 signatures, 7 active users, 2 organisms, 2 assay types; COUNT(*) on staging is 287.
   • Withdraw the MCP run_enrichment tool until it is rebuilt on the new runHypeR() #139: staging tools/list is compare_signatures, get_signature_context, list_vocabulary, rummagene_enrich, search_collections, search_features, search_geneset_entries, search_geneset_resources, search_signatures.
   • Let a SigRepo client change rebuild the staging image #122: :dev has OmicSignature 1.4.1 (hypeR 2.11.1, fgsea 1.32.4). The client-triggered path itself has never run: this repo has no repository_dispatch run, only the 2026-10-07 workflow_dispatch, because client #241 is unmerged. The image bakes client d6dd4d88 (the #262 merge), two merges behind client dev.
   • Install hypeR before SigRepo and fail the build when an install step leaves no package #145: the Dockerfile installs hypeR before SigRepo, each step guarded by stopifnot(requireNamespace()); library(SigRepo) in :dev attaches 112 exports including the seven the rebuilt tabs call.
   • Remove the retired React web app #134, Keep only code, docs and the testthat suite in the repository #136: no web/ on dev and no sigrepo-web in any compose file. Staging still has an orphan sigrepo-local-web container (today's deploy warned about it) and production still runs sigrepo-web on 8050; both are docker rm chores at deploy time.
   • Publish the staging Shiny interface on SHINY_BIND so BU-network users need no tunnel #143: unticked for now; see the note I sent you directly.

   Not ticked

   • Rebuild the Shiny Annotate tab on SigRepo::runHypeR() and open it #148: LLFS_Aging_Gene_2023 is not a repository signature on staging, so it went in as an uploaded .rds. Hypergeometric, kstest and fgsea against Hallmarks each produced the results table (100 / 100 / 49 rows), dot plot, enrichment plot, hypeR table and a 200 Excel export, and the R code tab's script reproduces each run exactly (same rows, same 5 / 7 / 4 genesets at FDR ≤ 0.05) when run against staging. The geneset status reads "H: 50 genesets (Homo sapiens, MSigDB cache 2026-09-24)", so no cache warning, but the cache is not in the image: neither :dev nor :latest has /SigRepo_Server/data/msigdb_genesets; staging (and production) read the 28 files /init_db_genesets wrote into the bind-mounted checkout. Mouse: the Organism control is a required select that defaults to Homo sapiens. Picking Aging_Ms_male_Liver_SIRT6_KO_vs_WT_Kanfi2012 leaves it there and readiness blocks the run with "A run tests one organism: '…' (Mus musculus) is not Homo sapiens. Change the organism in Step 2 or run it separately." That is the behaviour agreed on 2026-09-16, not "species is picked from the signature". With the organism set to Mus musculus, Aging_4mosc1_Old_vs_Young_MontiLab2024 ran fgsea (50 rows, 21 at FDR ≤ 0.05) and kstest (100 rows, 31) with every tab, plot and export working, and the SIRT6 signature's hypergeometric run stops cleanly with "No query had genesets left to test" because its two 14-gene groups find no Hallmark genes, which is the intended error path. The box needs rewording to what was built, or a follow-up issue for the auto-pick.
   • Add an API smoke test for the staging instance #106, Add a headless Shiny sign-in check for staging #108: today's smoke_test.sh passed api docs, search (287), gene sets (50 Hallmark, source=cache), enrichment and compare; its one failure is containment: published off 127.0.0.1: 0.0.0.0:8051, which is Publish the staging Shiny interface to the BU network so lab members need no SSH tunnel #142's deliberate SHINY_BIND, so the script exits 1 on every montilab deploy until it gets an allowance. smoke_test_ui.R through the tunnel: "PASS signed in as devadmin" with every tab present.
   • Add a one-command deploy script for the montilab staging instance #110, Add a reseed script for the staging database #112: the deploy ran today from a running stack rather than a clean state, and that copy printed "Running" for api and mcp instead of recreating them (the --force-recreate version is on the host now, md5 matches the Mac copy). The only api/ change since the last recreate is Move the Shiny sign-in, register, forgot-password and profile flows into an auth module #162's user.R. Reseed was not run (sudo).
   • Run dev's test suite in the dev image #118, Scope the containment check to our own containers #120: dev CI's testthat-integration ran in montilab/sigrepo:dev with the client checked out at dev (15:53 today, green); the containment check matches the exact sigrepo-local-* names. "Green on the release PR" waits for the PR.
   • Apply and record schema migrations with a runner #151: open and DIRTY against dev. It makes the API refuse an unmigrated database, so if it goes in, the release steps need a scripts/migrate.sh line before the production recreate. Without it no migration is needed: nothing in master..dev touches mysql/.
   • Document the dev branch, the release pull request and the staging tunnel #114, Say when Closes #NNN actually fires under the dev model #117, Close linked issues when a pull request merges into dev #147: the CONTRIBUTING text is still accurate; close-linked-issues fires on merges into dev only.

   For the release steps

   • dev → master merges clean (git merge-tree); the client release PR does not (NEWS.md), see Release 1.1.0 from dev: hypeR-native client, searchExternal(), LINCS script writer SigRepo#267.
   • Rollback baseline on the droplet: /SigRepo_Server at c24c6ee (master), image b65374fddca7, and the running API container already carries OmicSignature 1.4.1 with client master 80e36475. docker-compose-vm.yml changes in this release (sigrepo-web removed), so the deploy is a pull plus up with the new file, and sigrepo-web becomes an orphan to remove. The droplet had 346 MB available at the time of checking, so recreate the services one at a time.
   • A server master merge rebuilds :latest with whatever client master is at that moment, and a client master merge triggers nothing, so merge the client first, or run workflow_dispatch on master after both are in.
3. Cvicnaire commented on Oct 14, 2026

   

   Cvicnaire

   on Oct 14, 2026

   CollaboratorAuthor

   More actions

   #151 (schema migration runner) is left out of this release by Cameron's decision; it stays an open PR against dev and goes in a later release. Nothing in master..dev touches mysql/, so this release needs no migration step.
4. Cvicnaire commented on Oct 14, 2026

   

   Cvicnaire

   on Oct 14, 2026

   CollaboratorAuthor

   More actions

   Release PR opened: #163 (dev -> master, checklist in the body). Note from montilab/SigRepo#267: the client's repository_dispatch from #241 is dropped by GitHub until this release puts the trigger on master, because dispatches go to the default branch's workflow. A manual workflow_dispatch on dev is building :dev now.
5. Cvicnaire commented on Oct 14, 2026

   

   Cvicnaire

   on Oct 14, 2026

   CollaboratorAuthor

   More actions

   #143 reworded and ticked: 8051 is public on purpose as of 2026-10-14, Shiny's login is the gate, the other SigRepo ports stay on loopback. The containment allowance is #165 (--public-ports, into dev; it joins the release PR #163 when merged). The montilab deploy script in sigrepo-ops now passes the flag from SHINY_BIND and has been copied to the host, and ~/check_staging.sh there passes it too, so the next staging deploy should end green.
6. Cvicnaire commented on Oct 14, 2026

   

   Cvicnaire

   on Oct 14, 2026

   CollaboratorAuthor

   More actions

   Released 2026-10-14. #163 merged into master at 16:07 UTC (merge commit f53606d; the README bot then added a2afda4), one minute after the client's #274, so no rebuild happened between the two merges. dev was fast-forwarded to master the same day.

   Image. montilab/sigrepo:latest was rebuilt twice by the two master pushes; the later build (run 38595697369) pushed at 16:29 UTC and bakes SIGREPO_BRANCH=master = SigRepo 1.1.0 at 6c3b7d2 (the v1.1.0 tag), OmicSignature 1.4.1 and hypeR 2.11.1, read from the build log.

   Staging dispatch now works. With the repository_dispatch trigger on master, the first client-triggered build ever ran here (run 38596663461, event repository_dispatch, fired by gh workflow run notify-staging-image.yml --ref dev on the client) and republished :dev at 16:36 UTC with the same client. The back-merge push itself did not dispatch because the client's notify workflow has a paths: filter and the merge commit changed no package file.

   Boxes. #148 reworded to what was built (Organism control with a mismatch guard, cache read from the bind-mounted checkout) and ticked from the 2026-10-13 validation. #151 struck through: out of this release by decision.

   Left for the production deploy (needs a shell on the droplet):

   cd /SigRepo_Server
   docker compose -f docker-compose-vm.yml pull sigrepo-api        # one image; mcp and shiny share it
   git pull
   docker rm -f sigrepo-web                                        # retired by #134, frees memory first
   docker compose -f docker-compose-vm.yml up -d sigrepo-api
   docker compose -f docker-compose-vm.yml up -d sigrepo-mcp
   docker compose -f docker-compose-vm.yml up -d sigrepo-shiny
   docker exec sigrepo-api Rscript -e 'cat(as.character(packageVersion("SigRepo")), as.character(packageVersion("OmicSignature")), "\n")'
   bash scripts/smoke_test.sh   # and scripts/smoke_test_ui.R, against sigrepo.org
   docker image prune -f        # drops the 2026-10-07 image

   One service at a time: the box had 373 MB available before the deploy. Then tick the last two boxes and close this issue.
7. Cvicnaire

   closed this as completedon Oct 14, 2026