The integration cascade stored its top layer differently from the two below
it: org (L2) and user (L3) plugin config lived in PocketBase, in a
pluginSettings field, while the global (L1) layer sat in a plugins.json
next to the binary. That split was accretion rather than design - the file
was the whole store in the v1 MVP, and the per-tenant layers were later
built on PocketBase and layered on top of it instead of replacing it.
It also cost something real. plugins.json was a second state store with
different durability from pb_data: its own volume, its own ownership, its
own backup. Losing pb_data is unmissable; losing api_data was silent, which
is how "every plugin comes back disabled after a redeploy" happened.
L1 now lives in the app_settings collection - one record keyed "global",
holding its settings in a pluginSettings field, the same mechanism and the
same field name the layers below use. The documents still differ in shape,
because only L1 carries enable state and the registration of external
plugins, but the storage is no longer a special case.
The Manager grows a Store seam (PocketBase in production, file for the
import, memory for tests) and, more importantly, a loaded gate. Settings in
a database mean the store can be unreachable at boot - a cold stack, or a
service account still to be set from the panel. That must not read as "no
plugins configured", or the first save would write emptiness over real
settings. So until a read succeeds the Manager stays unloaded, every
mutation is refused, /api/admin/plugins* answers 503, and a background
retry backs off to two minutes. The same gate covers a document that will
not parse: it is never replaced by one built from an empty map, which is a
stronger guarantee than the .corrupt backup it replaces.
Writing to a store also revealed a hole in the previous fix. Classifying a
save failure as errPersist was left to each Store, and a store that
returned a plain error would fall through to the "saved, but the plugin
failed to start" branch and be reported as a 200 - the same silent-success
bug through a different door. The Manager now classifies, whatever the
Store returns; a test pins it.
Upgrades are automatic: on the first boot that finds no settings in the
database, an existing plugins.json is imported and renamed to
plugins.json.migrated. The import is refused if the store is merely
unreachable, or if the file does not parse, so a stale or broken file can
never overwrite live settings. /data is still needed - the panel rewrites
.env there when it retargets PocketBase - but plugin settings no longer
depend on it.
21 tests in internal/plugins cover both stores, including the production
path against a fake PocketBase: create-then-update of the singleton,
round-trip across a restart, an outage that leaves settings intact, a
missing collection reading as not-ready rather than empty, and the import
running exactly once. go build, go vet and go test ./... pass. Schema
changes are mirrored into scripts/setup-pocketbase.mjs as that file
requires. Not verified: no Docker CLI here, so no image was built and the
bootstrap of app_settings against a real PocketBase is untested outside the
fake.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Reported symptom: every plugin comes back disabled after redeploying the
image, having been enabled before it. The persistence design was already
right - each compose file mounts api_data:/data and points PLUGINS_FILE at
/data/plugins.json - so the fault was that a failed write to that file was
invisible. Three defects, each confirmed with a test before being fixed:
A failed write was reported as success. Upsert set rec.Enabled before it
persisted, and the handler folded the resulting error into the same
200-with-warning used for "saved, but the connector failed to start". The
panel reloaded, read the in-memory record and showed the plugin enabled;
only a restart revealed that nothing had reached the disk. A save that
fails now rolls back in memory and returns 500, so the panel row shows the
error instead of "Saved".
A corrupt state file silently wiped the rest. Load returned an error,
main.go logged it and carried on with an empty record set, so the next
toggle overwrote plugins.json and took every other plugin's config with
it. An unreadable file is now moved aside to plugins.json.corrupt, and
persistLocked writes through a temp file + rename so an interrupted write
cannot produce that corrupt file in the first place.
A state file holding "null" panicked the server with "assignment to entry
in nil map" on the next save, and a null entry nil-dereferenced in Load.
Both now decode to "nothing configured".
Two changes make the next such failure loud rather than silent.
StartPlugins probes writability at boot and warns that plugin changes will
not survive a restart. And the API Server image gains the root entrypoint
the AIO image already had - chown /data, then drop to app via su-exec -
because a host bind mount (API_DATA=/srv/...) or a volume created before
/data existed arrives root-owned, and the unprivileged process cannot
write to it.
Not addressed here: a deployment that never reuses the named volume
(docker compose down -v, a renamed compose project, an anonymous volume
from a bare docker run) loses the file whatever the code does. The new
boot warning tells the two apart - writable but empty means the volume is
the problem, not permissions.
go build, go vet and go test ./... all pass. The Dockerfile change is
reviewed but not built: there is no Docker CLI on this machine, so the
su-exec privilege drop follows standard Alpine practice rather than an
observed run.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The all-in-one image builds from the project root, and Docker only reads
.dockerignore from the context root — so the ones under "API Server" and
"Web App" never applied to it and every AIO build shipped the whole tree,
"Phone App/build" included. A root .dockerignore allow-lists the paths that
build actually copies.
The dev split stack passed neither PB_BOOTSTRAP nor the SUPERADMIN vars, so
it created the schema and then no user to log in with. It passes them now,
and .env.example says so.
WEBAPP_URL was never set anywhere, leaving the panel status page probing
localhost:8090 — itself — and always reporting the Web App as down. Each
compose file now points it at wherever the Web App really is, and the BFF
grew a real /healthz instead of letting the SPA fallback answer probes with
index.html and look healthy no matter what.
In the AIO, PocketBase and the API Server drop to an unprivileged user;
only nginx stays root to bind :80. The entrypoint takes ownership of the
two volumes first, so data written by the old root-only image stays
writable. All three images carry a HEALTHCHECK, every compose file declares
one too (so depends_on still gates against an older pulled image), and
web-app waits for the API Server to be serving rather than merely started.
Also: pinned alpine/golang/node and PocketBase 0.39.11, so a rebuild months
from now produces the same image; nginx forwards WebSocket upgrades instead
of stripping them, with the map in http.d where Alpine actually reads it;
and a .gitattributes keeps entrypoint.sh on LF, because a CRLF shebang from
a Windows clone fails at container start with "no such file or directory".
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Audited every Dockerfile, compose file and .env.example against the code
they deploy. Four things had drifted:
Persistence. The API Server writes plugins.json and rewrites .env (the
panel's retarget-PocketBase flow) relative to its working directory,
which was a root-owned /app while the process runs as the app user - so
both writes failed, and no volume was declared to keep them anyway. The
binary moves to /usr/local/bin and the working directory becomes a /data
volume owned by app. The AIO image gets the same via directory=/data on
its supervisord program.
OCPP. Charger control was undeployable: OCPP_REQUIRE_TLS defaults to true
and appeared in no Docker file, so a charger dialling the plain-HTTP
/ocpp/{serial} was rejected with nothing explaining why. Both OCPP vars
are now threaded through the compose files and env examples, with the
reasoning (the per-charger control token rides in a Basic-auth header).
CORS. API Server/docker-compose.yml defaulted to localhost:5173, the Vite
dev port, where every other file uses 8090.
Env names. .env.example has called PB_URL/PB_ADMIN_*/PORT legacy for a
while, but the Docker layer still used them. Container-side names are now
POCKETBASE_*/API_ADDR; the .env keys operators set stay PB_ADMIN_* so
existing .env files keep working.
Left alone deliberately: alpine:latest stays unpinned (cannot verify
current tags or test the build from here), and the golang/node bases
already match go.mod and Vite 8's floor.
Validated as YAML only - there is no Docker CLI on this machine, so no
image was built and the /data ownership fix follows standard volume
semantics rather than an observed run.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
READMEs: correct the auth model (PocketBase token relay, not JWT/sessions),
document the full feature set (technical checks, fuel, maintenance, documents,
reminders, attachments, integrations, OCPP charging control), the shipping
built-in connectors (toyota, anker-solix), and the current endpoint surface.
Docker: build against the current repo layout — Go 1.26, cmd/server entry
point, Web App source under web/. Add the missing Web App Dockerfile (Go BFF)
and .dockerignore, drop the obsolete AUTH_SECRET, modernise CORS var naming,
and standardise on drivervault-* naming.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
