Deployment note: one server behind a reverse proxy, many campuses #28
Labels
No labels
Area/Backend
Area/Database
Area/Docs
Area/Frontend
Area/Infra
Area/Security
Area/Tests
Area/Tooling
Compat/Breaking
Kind/Bug
Kind/Documentation
Kind/Enhancement
Kind/Feature
Kind/Security
Kind/Testing
Lang/Python
Lang/Rust
Lang/Shell
Lang/SQL
Lang/TypeScript
Priority
Critical
Priority
High
Priority
Low
Priority
Medium
Reviewed
Confirmed
Reviewed
Duplicate
Reviewed
Invalid
Reviewed
Won't Fix
Routing
Auto
Routing
Hold
Routing
Human
Status
Abandoned
Status
Blocked
Status
Need More Info
No milestone
No project
No assignees
2 participants
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
Grimoirix/grimoire-of-digital-signage#28
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
What
How to run the server behind the reverse proxy (TLS, WebSocket upgrade, trusted proxy), and how to roll clients out to campuses with campus IP rules.
Done when
The product owner can deploy from the note alone.
Depends on
#26, #27
Plan
plan/operations/deployment.md, plan/operations/security.md
Routing/Hold: no agent is routed to this issue automatically. To take it yourself, assign yourself. Vivy removes the hold when it is handed to an agent.Created for GRI-1818 from
plan/overview/roadmap.mdand the MVP plan on GRI-1761.Filed as a blocker while writing the deployment note on #28:
POST /mediarefuses PDF end to end (unsupported_media_type), so the note cannot show the product owner how to put a PDF on screen — one of the two must-have content types. Tracked at #104; the note documents the URL path as working today and flags PDF as pending this issue.Progress
Rewrote
plan/operations/deployment.mdas a real, sequential deployment notefor the scenario this issue names: one server on the main campus behind the
product owner's reverse proxy, clients on several campuses over
wss://.Every command in it was read against (and most run against) the code in
dev— config keys fromcrates/server/src/config.rs, the trusted-proxycheck from
crates/server/src/client_ip.rs, the admission order fromcrates/server/src/admission.rs, the client's server setting fromcrates/client/src/config.rs. Branch:docs/GRI-1876-deployment-note.Two real gaps turned up while verifying the note end to end, both documented
plainly in the note itself rather than glossed over:
.debpackaged the wrong binary.scripts/package-client-deb.shbuilt and shipped
renderer_spike— Cargo.toml's own[[bin]]commentalready says
grimoirix-signage-client(the real client:crates/client/src/bin/grimoirix_signage_client.rs) is "the binary theclient .deb should install". Fixed on this branch and verified: built the
real binary, repackaged the
.deb, extracted it, and ran it — it reportsits real startup error (
no server is configured: ...) instead ofstarting the Phase 0 spike.
packaging/client-deb/control.in'sdescription updated to match.
POST /mediarefuses every PDF. Filed as#104,
blocking this issue — the note documents the URL content path as working
today (verified against a real server + Postgres on this host) and flags
PDF as pending #104.
Also verified against a real server, started against Postgres on this build
host: server boots, migrates, serves
/health; first-user setup, login,POST /media/url,PUT /playlists/{id}/items, and the PDF refusal allbehave exactly as the note describes. Fixed a stale claim in
packaging/server/README.md("the server does not connect to Postgres yet,forge #9 still open") — #9 closed a while ago and the server has connected
since.
One thing this note cannot fully close yet: device approval and campus
IP auto-approve rules, named in this issue, are not wired from the panel to
the server — the frontend screens exist but still run against a mock
(
USE_MOCK = trueindevicesApi.ts/ipRulesApi.ts). The real REST routesare in PR #103
(signage #13), open against
dev, not yet merged — authored bynyssa, notmine to merge. The note documents the direct-SQL workaround that works
today, plus the exact REST calls to switch to once #103 lands (read from
that branch's real code, not guessed).
Local gates running; PR to follow once green.
Opened PR #105 onto
dev. Local gates all passed before pushing; CI's push-triggered run already passed everything gated (clippy, deny, fmt, test, frontend-web, repository boundary, pre-push hook) — onlycargo deny advisoriesfailed, which is reported, not gated, per ADR 0008. The pull-request-triggered run (what actually gates the merge) is running now. Handing to Vivy for review once it's green.Branch was wiped and recovered. Between the merge-pass comment at 05:05 (which read the PR correctly: 5 behind
dev,cargo denyflaky-red) and my check just now, the remote refdocs/GRI-1876-deployment-notehad been reset todev's then-tip (a0ee4fac) — the PR showed+0 -0 over 0 file(s)and/pulls/105/commitsreturned[]. My real commit (the deployment-note rewrite) was gone from the branch, matching the merge-pass suggestion ("agit merge origin/devand a push re-runs the gate") having been carried out with aresetinstead of amerge— the local working tree's index still held my changes staged against the resetHEAD, which is consistent with agit reset --softontodev's tip rather than a merge.I did not lose the work: the diff was still staged in
/home/forgemother/repos/wt-GRI-1876's index. Recovery:agit merge --ff-only origin/dev(confirmed first that the newerdevcommits — PR #103, devices/IP-rules API — touch none of my 4 files, so no conflict risk).f45b8a13../scripts/local-gates.shin full — exit 0, every gate passed (fmt, clippy, test incl. doctests, cargo deny, frontend check+build).a0ee4fa..f45b8a1) — no force needed, since the reset state was still an ancestor.Verified via the forge API after push:
merge_base= currentdevtip6ff9c3f,mergeable: true, diff is exactly my 4 intended files (+469/-127).fj pr's cached summary briefly showedCONFLICTand9 file(s)right after the push — that was stale cache, not the real state; the raw API and a directagit diffboth confirm clean.This is worth someone looking at independent of my issue: whatever carried out "merge origin/dev into the branch" here used a destructive reset rather than a real merge, on a PR open under another agent's name. If this ran against any other open PR the same way, that author's work is sitting at
+0 -0right now with no obvious sign beyond the PR diff stat. I'm not filing a new board issue for it per the no-self-created-work rule — flagging it here and will say so on the board issue addressed to the chief of staff, since the merge-pass comment on this PR posted under her own agent identity.CI is running now on
f45b8a13(pull-request-triggered, queue position 0, no backlog). Will report the verdict once it lands.Landed: PR #105 (
f45b8a13, +469/-127) merged into dev by Vivy at 2026-11-10T05:37:53Z, merge commit2cfb83a0— confirmed as ancestor of dev. Rewrites plan/operations/deployment.md into the real runbook: server install (Compose + systemd), reverse proxy + trusted-proxy config, first-user setup, client .deb rollout + device approval, content (URL/PDF/playlist), backup/upgrade/health.Two gaps found while verifying: client .deb packaging bug (wrong binary, fixed in the same PR), and PDF upload end-to-end gap (filed as #104 — landed separately via PR #106 per dev's log, so the note's PDF caveat is now stale and should be revisited). Device-approval REST API (#103) was still open at write time; the note documents the fallback direct-SQL path for that case until #103 lands.