Mislukte mails door een configuratieprobleem (auth_failed, rejected_sender, tls_failed, not_configured) worden
als Error gelogd zodat Sentry een alert stuurt; timeouts en ontvangerproblemen blijven Warning.
Documentatie: uitrol per host met de hand in shared/env, rollback, monitoringplan, checklist; ASF-productie op
IIS als open punt voor een latere feature.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
De route-guard las elke fout (429, 500, 503, 404) als 'niet geinitialiseerd' en stuurde naar de
setup-pagina. Nu geen gok: bij een onleesbare status geen doorverwijzing. De statuslezing valt onder de
ruimere limiter in plaats van de strikte login-limiter (die 5 per minuut deelt met echte inlogpogingen).
De loginpagina meldt als het CMS uitgeschakeld of in onderhoud is.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Een uitgeschakeld CMS gaf een kale 503 op de publieke website en op /admin, omdat de poort elk pad
stopte dat niet op de bypass-lijst stond. Sinds de host website en admin zelf serveert, viel dat verkeer
eronder en kon de admin-app de status niet lezen om de uitgeschakeld-kaart te tonen. Nu worden alleen
paden onder /api/ beoordeeld; alle andere paden gaan door.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
De SMTP-verzender gaf alle stappen samen een tijdsbudget; nu krijgt elke stap (verbinden met STARTTLS,
inloggen, versturen) zijn eigen volledige timeout. Bij een timeout noemt de code de stap
(timeout_connect, timeout_auth, timeout_send) en toont het log per stap de verstreken milliseconden.
TimeoutSeconds blijft instelbaar (standaard 10), bereik nu 1 tot 120. Infra-doc met veiligheidsmodus,
timeout en leeswijzer voor het journal.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Oorzaak van de onbehandelde fout 'window is not defined' in de CI: ResetPasswordPage (en dezelfde
constructie in InviteCompletePage en SetupPage) startte een setTimeout voor de doorverwijzing naar
de loginpagina en liet die doorlopen na het verlaten van de pagina; in de testrunner vuurde hij nadat
de testomgeving was opgeruimd en navigeerde een router die er niet meer was. Nieuwe hook
useDelayedAction ruimt de timer op bij het verlaten.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
SMTP: bij een geweigerd commando logt de verzender fase, statuscode en de eerste regel van het serverantwoord
(adressen vervangen, max 120 tekens); specifiekere foutcodes (rejected_sender, rejected_recipient, tls_failed,
connection_failed) die de beheerder in het venster ziet; opstartlog met SMTP-host, poort en standaard
afzenderdomein. Uit de property-test: besturingstekens worden eerst verwijderd, anders kon een adres weer
aan elkaar plakken.
Nieuwe uitnodigingslink: regenerate-invite mailt alleen met sendEmail=true (standaard false); eerst een
keuzevenster 'Alleen link kopieren' / 'Link kopieren en mailen' (zonder mail alleen de link).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Een echt access token heeft geen naam-claim, dus User.Identity.Name was null en elke aanroeper gold als
Beheerder: een Eigenaar kreeg 403 bij het toekennen van de Eigenaar-rol, en de controles voor 'jezelf'
(niet zelf verwijderen of deactiveren) sloegen nooit aan. Nu via het id van het token. Regressietest met
echte tokens en de echte pipeline.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
De schil was min-h-svh, dus zodra de inhoud iets hoger was dan het scherm scrolde het hele document
en de zijbalk mee. Nu is de schil precies een scherm hoog zonder eigen scroll en scrolt alleen het
inhoudsgebied; het menu in de zijbalk scrolt zelf op een heel laag scherm.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Eén bron (IMailAvailability) bepaalt of mail aan staat: SMTP-host gezet EN de hele mailconfiguratie geldig.
Anders start de host gewoon, gebruikt de Null-verzender, meldt emailEnabled=false en logt eenmalig
mail_configuration_invalid met de namen van de ontbrekende of ongeldige instellingen (nooit waarden of
geheimen). Getest: halve config, elke soort fout, geen geheim in het log, volledige config werkt als
voorheen, geen mailinstelling geeft geen enkele waarschuwing, alle verbruikers zien mail als uit.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Oorzaak 502 op ASF test: Email__Smtp__Host stond al in de omgeving maar App__PublicBaseUrl nog niet;
de startvalidatie weigert zo'n halve mailconfiguratie. Een host zonder enige mailinstelling start
gewoon: nu voor master, slave, SlpSoftware en ASF getest. Website__LogLevel vervangt de
Logging__LogLevel__SlpModularCms.Website-variabele, waarvan systemd de naam met punt negeert.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
POST auth/forgot-password antwoordt altijd hetzelfde (202) en handelt de mail na het antwoord
af; POST auth/reset-password zet een nieuw wachtwoord met de link uit de mail. Alleen een
voltooide reset trekt alle refresh tokens in (en beeindigt eerdere access tokens via de
security stamp); het aanvragen raakt niets. Link eenmalig en tijdsbeperkt (standaard 60 minuten),
cooldown van 5 minuten per adres in het geheugen, beide endpoints onder de login-limiter.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Kolommen Users.PreferredLanguage en Invitations.Language (additieve migratie). IdentityMailService
verstuurt uitnodiging, nieuwe link en meldingen bij deactiveren, heractiveren en rolwijziging;
een mailfout laat nooit een actie falen, het resultaat gaat mee in de respons naast de
kopieerbare link. Nieuwe actie resend-invite. De security stamp zit in het access token en wordt
vergeleken in de actief-controle: een wachtwoordwijziging beeindigt eerder uitgegeven access tokens,
de refresh geeft stil een nieuw token.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
IEmailSender met MailKit-implementatie (StartTls, timeout, fouten als waarde), opties met
startvalidatie, nl/en sjabloon, taalkeuze, tabel SiteSettings (additieve migratie),
SiteSettings-API (Beheerder leest, Eigenaar wijzigt) en emailEnabled/defaultLanguage in de
capabilities. Zonder SMTP-host verandert er niets.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Een Beheerder mag alleen gebruikers met rol User verwijderen, deactiveren, van rol wisselen
of hun uitnodiging intrekken, en kent nooit de Owner-rol toe. De regels staan op een plek
(UserManagementRules) en worden in de API afgedwongen. Nieuwe actie om een openstaande of
verlopen uitnodiging een nieuwe link te geven (oude link vervalt).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
De naam uit het uitnodigingsformulier werd niet opgeslagen (profiel en verwijder-dialoog
toonden het e-mailadres). Een geaccepteerde uitnodiging bleef na het verwijderen van de
gebruiker staan en blokkeerde het adres; het conflict gaf een 500. Nu 409 met een
specifieke melding en opruimen van verouderde uitnodigingen.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Het JWT werd alleen op handtekening en looptijd gecontroleerd, dus een gedeactiveerde
gebruiker kon tot het verlopen van het token lezen en schrijven. Nu controleert de
bearer-validatie per request of het account bestaat en actief is, en weigert de refresh
een inactieve gebruiker.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
De frontend bouwde origin + inviteLink; op een host waar de root een aparte website is
leidde de link naar die website. Nu origin + BASE_URL + pad via buildAppUrl.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
De catch-all 404-route van de website werd ook voor /admin/... gekozen, waardoor de
static-file-middleware (die routes met endpoint overslaat) de adminbestanden niet
serveerde. Een routeconstraint houdt de gereserveerde prefixes buiten die route.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- Core: opt-in ForwardedHeaders:KnownProxies en rate limiters per clientadres
(zonder configuratie ongewijzigd gedrag)
- Identity: optioneel setup-token (X-Setup-Token), beide setup-endpoints onder de
login-limiter, beveiligingsgebeurtenis setup_token_rejected (5007)
- Nieuw client-project SlpModularCms.Api.AndreaSlopFotografie (Identity + Availability)
met bijbehorende tests
- Frontend: thema asf (VITE_THEME), VITE_DEFAULT_LANGUAGE, tokenveld op de setup-pagina
- CI: jobs publish/deploy-asf-test achter ASF_TEST_DEPLOY_ENABLED
- Fix U1: 301-Location wordt nu percent-encoded (gevonden door de property-test)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- Pure WebsitePathResolver (canonieke 301's, echte 404, dotfiles geweigerd, gereserveerde prefixes)
- PrerenderedSiteMiddleware, 404-fallback na de API-routes, gzip, Cache-Control, NoIndex/robots.txt
- Beleid StrictSite met eigen AllowedSite*Origins; volledige headerset op alle websiteresponses
- Tests: voorbeelden, FsCheck-properties en integratietests op een in-memory server
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Consistent met de confirm/cancel-knoppen ernaast (ghost icon buttons)
in plaats van de tekst-linkknop.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Op verzoek: "Slug: gegenereerde-slug [Aanpassen]" direct onder het
titelveld, waarbij de preview meebeweegt met wat je typt in Titel
(client-side mirror van de backend Slugifier) zolang de slug niet
handmatig is overschreven. Zodra de admin 'm expliciet aanpast en
niet leeg laat, stopt het meebewegen met de titel — consistent met
hoe OfferingsService.UpdateAsync de slug al behandelde.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
De CMS-admin toonde het nieuwe Slug-veld nog niet — dat bestond alleen
in de API-respons. Voegt een read-only slugweergave onder het
titelveld toe met een potlood-icoon om 'm inline te bewerken.
Backend: Create/UpdateOfferingRequest accepteren nu een optionele
Slug-override. Zonder expliciete waarde blijft het bestaande gedrag
(auto-genereren uit titel, regenereren bij titelwijziging). Met een
expliciete waarde wordt die genormaliseerd en op uniekheid
gecontroleerd; een botsing geeft 409 Conflict (RFC 9457 ProblemDetails,
naar het patroon van MasterControlledAvailabilityException).
Frontend: de slug wordt alleen meegestuurd als de admin 'm daadwerkelijk
bewerkt heeft (dirtyFields.slug), zodat een ongemoeide slug bij een
titelwijziging gewoon blijft auto-regenereren.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Het Offering-ID is een GUID en niet geschikt om aan gebruikers te tonen.
Voeg een op de titel gebaseerde, unieke slug toe die via de public en
admin API wordt meegegeven, zodat de frontend iets leesbaars kan tonen
in plaats van het ruwe ID. Het ID blijft ongewijzigd de echte identifier.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
A new push to a PR previously left the prior CI attempt running
independently instead of superseding it. Scoped strictly to
pull_request events, since deploy-test/deploy-production only ever
run on a push to master or a manual workflow_dispatch.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FWyStNL2ZsjrS7FLd7xvvN
Confirms existing tag-based Sentry alerting and domain-based UptimeRobot/
Umami monitoring already cover the Offerings module and the Api.SlpSoftware
cutover with no new configuration. Closes the SECURITY-13 audit-trail open
item and records the final production-readiness traceability, completing
the slpsoftware-api feature's AI-DLC lifecycle.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FWyStNL2ZsjrS7FLd7xvvN
Implements Unit 2 "Offerings" (backend module, admin CRUD UI with
drag-and-drop reordering, public GET /api/v1/offerings endpoint) and
executes the feature's D-15 CI/CD cutover, switching the deploy
pipeline's build/publish target from SlpModularCms.Api to
SlpModularCms.Api.SlpSoftware.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FWyStNL2ZsjrS7FLd7xvvN
The services: block was silently broken on this runner: job and service
containers both use host networking here, so the ports: mapping was
ignored and the mariadb:11 service ended up sharing the host's own port
3306 -- which something else on this runner already answers on (root auth
was rejected with a password nothing but this job ever set). Replaced with
an explicit `docker run` on host port 3307, bypassing that collision, plus
a readiness loop. Also fixes the readiness command itself: mariadb:11 does
not provide a `mysqladmin` alias, the correct binary is `mariadb-admin`
(verified locally against the exact image before pushing this).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FWyStNL2ZsjrS7FLd7xvvN
The new pipeline regression tests boot the real Api host via
WebApplicationFactory, which unconditionally runs the startup database
migration -- unlike every other test project here, which mocks or uses EF
Core InMemory. CI had no MariaDB service at all, so those tests failed on
the connection itself. Adds a mariadb service container to the backend-test
job plus a ConnectionStrings__DefaultConnection override for that step,
which overrides the appsettings.Development.json placeholder via standard
config layering without touching any committed appsettings file or
affecting local test runs.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FWyStNL2ZsjrS7FLd7xvvN
Unit 1 of the slpsoftware-api feature (FR-1/FR-2/FR-3): a new Client project
in the Clients solution folder, intended to eventually become the deployed
API for test.slpsoftware.nl/slpsoftware.nl, hosting the same four modules as
SlpModularCms.Api plus a future Offerings module.
- Extracts SlpModularCms.Api/Program.cs's hosting-pipeline composition into
SlpModularCms.Core.Hosting.CmsHost (ConfigureServices/ConfigurePipeline),
shared by both Client projects so they cannot drift apart
- Moves StaticContentExtensions.cs + WebsitePlaceholder.html from Api into
Core, since CmsHost cannot live in Api but Core cannot depend on Api
- Adds SlpModularCms.Api.SlpSoftware with its own isolated local dev database
and dev ports (5286/7223, distinct from Api's and Api.Slave's)
- Adds SlpModularCms.Api.Tests with WebApplicationFactory-based pipeline
regression tests (security headers, health check, SPA fallback, rate
limiting), scoped to Api per NFR Design
- Adds a frontend dev:slpsoftware pnpm script mirroring dev:slave
- Fixes GlobalExceptionHandler logging routine 401s (e.g. an expired/missing
refresh token) as unhandled errors -- pre-existing, unrelated to this
feature's own scope, found while testing the new instance
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FWyStNL2ZsjrS7FLd7xvvN
Groups auth, setup, invitation, profile, users, cms, availability and
system code (services/hooks, components, schemas, mocks, pages) under
src/features/<name> instead of splitting by technical layer (api/,
components/, lib/schemas/, mocks/, pages/). Renames the old
connection-oriented `api` layer to `services` per feature, and splits
the monolithic api/types.ts into per-feature types.ts files (with
ProblemDetails/ApiResult merged into lib/api-client.ts as shared
infra). Layout-agnostic code (ui primitives, app shell, i18n, test
utils, lib) stays at the top level.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Gitea's UI has no reliable way to see a past run's dispatch input values, so the config job now echoes them into the log. publish-test previously always built and uploaded the test artifact regardless of the deploy_test/deploy_production selection, wasting a build on manual runs that were never going to deploy to test; it now mirrors publish-production by skipping when workflow_dispatch didn't check deploy_test. Push and pull_request runs are unaffected.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VhySVFARKAP89WqQ3Kff8M
Gitea Actions only implements always() from GitHub Actions' status-check functions; success(), failure(), and cancelled() aren't supported. publish-test/publish-production relied on !failure() && !cancelled() to let a run through when an unaffected gate job was skipped by the path filter while still blocking on an actual failure — rewritten to check needs.<job>.result explicitly instead, which Gitea does support.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VhySVFARKAP89WqQ3Kff8M
Push-to-master keeps auto-deploying to test unchanged. A new deploy_test input (default false) lets a manual workflow_dispatch run opt into a test deploy instead of it firing by default. A new changes job (dorny/paths-filter) gates the backend and frontend build/test/lint jobs on whether their side of the repo actually changed, shortening runs that only touch one side; publish jobs still run whenever either side changed, and manual runs always run everything since there's no commit to diff against.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VhySVFARKAP89WqQ3Kff8M
The aside had no explicit height, so it stretched to full height on
desktop only because its flex-row parent applies align-items:stretch.
In the mobile SidebarOverlay it sits in a plain (non-flex) h-full div,
so it shrank to fit its nav items, pushing settings/profile up under
the nav instead of staying pinned to the bottom.
actions/setup-node@v4's Post step attempts a cache-save that fails hard on
this self-hosted Gitea Actions runner (no GHES cache API support), marking
frontend-prepare as failed and blocking its dependents even though the
actual install/build work succeeds. v3's older cache implementation doesn't
hit this. Also cleans up two stale comments left over from the /admin
MSBuild-target removal.
Also updates the documented backup script to skip cleanly (exit 0) when the
target database doesn't exist yet (MariaDB error 1049) rather than failing
the whole deploy - expected on a brand-new environment's first production
run, before migrations have ever had a chance to create it. Still fails hard
on any other error, so a real backup failure against an existing database
still blocks the deploy as intended.
The pull_request trigger was a temporary convenience to test the deploy
pipeline itself against the real test environment on every PR push. Now
that the pipeline is verified working end to end, going back to only
deploying on a master push or manual dispatch avoids concurrent PRs
silently overwriting the single shared test environment.
5000ms was occasionally too tight on the self-hosted Actions runner for
lazy-loaded routes gated behind an async setup-status check (observed on
SettingsPage.test.tsx, passed reliably locally). Bumps to 10000ms across all
nine test files using that pattern, and raises vite.config.ts's global
testTimeout from 15000 to 20000 to keep headroom above it.
The BuildAndCopyAdminFrontend MSBuild target (BeforeTargets=Publish) never actually
worked: files it created after project evaluation were silently absent from the
publish output, and forcing them in via an explicit Content item collided with the
SDK's own static-web-asset resolution. Verified locally with a clean obj/bin and a
fresh node_modules - every deploy so far genuinely shipped without wwwroot/admin.
Moves the frontend build + copy into its own CI step ahead of dotnet publish, so the
SDK's ordinary wwwroot handling picks it up with no custom MSBuild involved.
Also fixes unreadable <code> badges on the website placeholder page (no explicit
text color, relying on inherited body color pairing unreliably with the badge
background).
ModuleOrchestrator already discovers SlpModularCms.Modules.*.dll from disk at
startup via reflection, so this needed no code change — only a persistent
location (mirroring shared/wwwroot-web) that survives release swaps, and a
deploy-scp.yaml step to symlink its contents into each new release before
restart.
The PI_MAIN_USERNAME/PASSWORD debug step served its purpose (found the
shell-metacharacter password bug fixed in 51488d6) and is removed now
that deploy-test gets past authentication.
"Link persistent website content" then failed with `ln: failed to
create symbolic link '.../wwwroot/web': No such file or directory` -
dotnet publish only emits a wwwroot/ folder when the source project's
own wwwroot has content, so a release built before any frontend/website
content exists can land with no wwwroot/ directory at all, not merely
an empty wwwroot/web/. `ln -s` can't create the missing parent
directory itself. Added `mkdir -p $RELEASE_DIR/wwwroot` before the
existing rm/ln pair to guarantee the parent exists either way.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015FffvxxJp5wG34Ru48GBig
Every ssh/scp step interpolated ${{ secrets.PI_MAIN_PASSWORD }} and
${{ secrets.PI_MAIN_USERNAME }} directly into the run: shell text.
Gitea/GitHub Actions substitutes that as literal text before bash ever
sees it, so any shell-metacharacter in the password ($, `, ", \) gets
re-interpreted by bash instead of passed through - silently changing
what sshpass actually receives. This plausibly explains a persistent
"Permission denied" even after the username and password length were
both confirmed correct via the debug step (which used env: and so
never hit this).
All five steps now receive SSH_USER/SSH_PASS/SSH_PORT/SSH_HOST via
env:, which bash treats as opaque values with no re-parsing.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015FffvxxJp5wG34Ru48GBig
Username is now confirmed correct (14 chars, exact match). This checks
whether the password secret carries leading/trailing whitespace or a
stray newline picked up when it was pasted into Gitea - the length
differs from the trimmed length if so, without ever logging the
password's actual characters.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015FffvxxJp5wG34Ru48GBig
So PR pushes deploy straight to the test environment, without a manual
workflow_dispatch each time. Only one test environment exists, so this
means concurrent PRs would overwrite each other's deploy - acceptable
with a single active PR, worth revisiting once that's no longer true.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015FffvxxJp5wG34Ru48GBig
Monitoring Setup: operations/plans/monitoring-setup-plan.md and
operations/monitoring/monitoring-instructions.md, covering Sentry alert
rules on the security_event tag, UptimeRobot's 6 liveness monitors, and
the two new Umami website entries for the admin SPA.
deployment-instructions.md gains a missing Observability__Environment
host var (without it, both environments would tag Sentry events as
"Production"), the nginx client_max_body_size fix for the 413 seen on
publish-test/production artifact uploads, and two troubleshooting notes
on Gitea Actions re-run behaviour: re-running deploy-test/production
alone loses the run's uploaded artifact, and re-running all jobs on an
existing (rather than a brand new) run can replay stale secrets.
deploy-scp.yaml: step names no longer show literal unresolved
${{ inputs.* }} text (Gitea doesn't interpolate that context in step
names), and a temporary debug step logs PI_MAIN_USERNAME/PASSWORD
length plus a username equality check to diagnose a persistent
Permission denied during the SSH steps, without ever logging the
secret values themselves.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015FffvxxJp5wG34Ru48GBig
The mocks re-derived VITE_API_BASE_URL with their own fallback instead of
reusing getAppConfig().apiBaseUrl, so the two silently diverged whenever the
env var was unset (as in CI, where no local .env.local is present) — every
MSW-intercepted request then had no matching handler.
Pre-existing file from the original frontend setup - had allowBuilds
and confirmModulesPurge but no packages list, which pnpm 9+ requires
the moment a workspace file exists at all, even for a single package.
Surfaced by the first real pipeline run (frontend-prepare failed with
"packages field missing or empty").
backup-slpmodularcms-db.sh named the CMS; this backs up slpsoftware
specifically, same as every other instance-scoped path already does.
Hardcoded in deploy-scp.yaml itself, so this is the actual pipeline
code, not just docs describing it.
Four contexts got tangled together across the doc as it grew: pi-main
as root, pi-main as gitea-workflow, the separate proxy Pi, and Gitea's
own web UI. Added a legend up front and tagged every single command
block so it's no longer something the reader has to infer.
The backend already serves /sentry-tunnel itself with its own rate
limit, payload cap, and DSN-derived destination - exactly what that
nginx block exists to avoid needing. Proxying the same path at nginx
intercepts it first and silently drops all three protections.
nginx and certbot were designed assuming they lived on the same host
as the app. They don't - a dedicated proxy Pi terminates TLS and
forwards plain HTTP over the LAN. Kestrel now binds 0.0.0.0 instead
of localhost, the whole certbot procedure moved to the proxy Pi's
side, and pi-main gets a firewall rule restricting the backend ports
to just the proxy Pi's address - otherwise binding all interfaces
would let anything on the LAN skip the proxy's TLS entirely.
New section walks through adding any domain from nothing - plain
HTTP block first (certbot's nginx plugin needs one to validate
against), then the certbot command, then a verification step. The
existing slpsoftware.nl blocks stay as the worked result of that
procedure rather than a separate one-off.
Traced live on the actual Pi: su - gitea-workflow left XDG_RUNTIME_DIR
unset even with the user manager already running and the bus socket
already there. None of su/sudo -i/sudo -u reliably go through
pam_systemd on Debian - only a real SSH login or sshd itself does,
which is what the deploy workflow already uses.
Enabling lingering doesn't start the user manager retroactively, and
sudo -u often drops XDG_RUNTIME_DIR - both produce the same "Failed
to connect to bus" error when setting this up by hand.
The target Pi only has MariaDB, and SQL Server has no ARM64 build at
all - not a config problem, a real gap discovered during deployment
setup. Swapped the EF Core provider, regenerated every migration,
updated connection strings and the backup script everywhere they
appear.
Took two tries to land on a provider that actually works: Pomelo
builds fine against this project's EF Core 10 packages but fails at
runtime (it's compiled against 9's internal API surface, which moved
in 10 wherever Identity/DataProtection force the newer packages).
Oracle's official provider builds and migrates fine but has a real
MariaDB bug in its own migration-lock code, reproduced against a live
database. Kept Oracle's provider and worked around just that one
broken method - everything else it does is correct - rather than
give up more of the stack to chase a workaround.
Verified against a real local MariaDB end to end: all three
migrations applied, both hosts start clean, full suite still green.
AllowedConnectOrigins had a sentry-ingest-origin placeholder that
was simply incorrect - the Sentry tunnel exists precisely so that
exception is never needed. Both SecurityHeaders fields are Umami-only
now, with inline comments on when to set them versus leave them out,
and a note that MasterModule__MasterUrl degrades to a warning rather
than a startup failure when unset.
SlpModularCms is the product; slpsoftware.nl is the customer running
it - the first one, and a fine stand-in for what a real deployment
looks like. Renamed the directories, systemd units, and backup paths
accordingly. Left the backup script's own filename alone since the
already-committed deploy-scp.yaml calls it by that exact name.
Applies to both the app's own release tree and the website upload
path - slpmodularcms/test and slpmodularcms/production side by side
under one parent, not two separately-named directories. Service unit
names stay hyphenated; those aren't folders.
No reason to mint a new system account for every project on the
same Pi - one shared, host-wide account works, namespaced by
directory per project underneath it.
webadmin turned out to be FTP-only and belongs to whoever uploads
customer websites, not to this pipeline. Switched to a separate
deploy account with its own home-directory paths, well away from
where the other sites live, and made the persistent website symlink
cross into webadmin's territory instead of the other way around.
Host setup checklist, real domains and ports, the database backup
script the deploy workflow only ever invokes, and a rollback plan.
Also drafted the FTPS switch procedure for whenever production moves
off the Pi, with a note that shared hosting is likely IIS-based -
that's a bigger change than swapping the transport.
Unit suites already passed individually; this is the full solution
together plus what none of them could check alone: two real hosts
talking to each other, a slave API key encrypted by one process and
decrypted by the next one after a restart, and headers landing on
actual static files instead of a simulated response.
U7 was the last one. State doc corrected: it had U7 down for an
infrastructure design pass that never actually happened, because by
the time units were carved up U7 had shrunk to documentation only.
WEBSITE_WORKSPACE.md, plus the parts of the README that were still
describing the old layout or a manual step the code doesn't need
anymore (the key-ring paragraph, mainly - that one was actively
wrong now, not just stale).
continuous_integration.yaml: six blocking checks, then a separate
publish per environment so a Vite build never gets tagged for the
wrong one, then a call into last commit's deploy workflow. Along the
way: the lint list had drifted (two problems not in the requirement,
one already fixed), and the Umami-origin gate needed a variable pair
of its own since the backend's side of that comparison lives on the
host, not in CI. Pinned the two vulnerable packages while at it.
deploy-scp.yaml: download, back up the production database before
touching anything, upload into a timestamped release, symlink the
persistent website in, switch current and restart, verify /health
with retries, prune old releases only once that check passes. No
sudo, no container actions, one shared interface U5 will call next.
Infrastructure design for U6: single Pi split by directory not port,
systemd --user services instead of sudo, a releases/current/shared
layout that keeps the customer website outside every atomic switch,
and a health check through the existing reverse proxy before pruning
anything. Flags that user-level systemd needs enable-linger or the
service dies the moment the deploy SSH session closes.
Records the stage as complete and logs the two defects that local testing
turned up after U3 and U4 were generated: ciphertext predating the key-ring
move, and an integrity check that could not tell an unreachable slave from
one that does not recognise the master.
Both were fixed in this branch rather than filed. The second is master/slave
domain behaviour rather than deployment work, so the state file says so
plainly — it sits here by decision, not because it belongs to the feature.
Verified at close: build 0 errors, 372 backend tests, 237 frontend tests.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HHoJpxYXzHACSQguHrC5fw