main
10
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 tajniak81andClaude Opus 5
|
fe1e314df9 |
The store comes apart into the four roles it always had
`weed server -s3` was never one thing. Master, volume, filer and the S3 gateway ran as four goroutines under one process, on one volume, sharing one fate. Splitting them into four containers changes nothing a client can see — the same bucket answers on the same port — but it makes three things possible that were not: the SeaweedFS admin UI, which wants a cluster to look at; a restart or an upgrade of one role without the others; and, eventually, a second volume server somewhere else. A fifth container carries the panel itself. The single-process files stay exactly as they were. These are `.split.` twins beside them, four in all, one per folder per shape, each with the .env example of the same name that both READMEs already promise. Identities are the part that could not simply be copied across. SeaweedFS picks its credentials from one source, in order: an -s3.config file, the filer's IAM store, then AWS_ACCESS_KEY_ID and its secret — and a higher source replaces a lower one rather than adding to it. The existing files use the env pair, which is fine precisely because nothing else writes identities there. Hand somebody a panel that can, and the first user they create lands in the filer's store, the store outranks the environment, and PocketBase's key stops existing — with the first failed upload as the notification. So the gateway here is started with no config file and no AWS_* at all, and the init container seeds PocketBase's identity into the filer's store instead: the same store the panel writes. One source of truth, PocketBase's key sitting in Object Store → Users beside every other, keys minted there picked up without a restart, and a rotated PB_S3_SECRET re-applied in place on the next boot rather than added as a second identity. That seeding is now allowed to fail. The bucket-create it grew out of was best-effort — `|| true`, on the reasoning that the API Server's own S3 check would report a gateway that was genuinely unreachable. That reasoning does not survive the change: a gateway whose IAM store is empty does not refuse anyone, it serves everyone, and the bucket would be wide open rather than unreachable. So the step ends by grepping the configuration back for the access key, the gateway waits on it completing successfully, and a seed that did not land stops the stack instead of opening it. The prod files publish the gateway and the panel, both on loopback, and nothing else. Port 8080 on the volume server hands out file content by file id with no authentication of any kind — the S3 credentials have no bearing on it — so publishing it would publish every attachment in the stack, and the panel shows what that port and the master's would. The panel's own password is required rather than defaulted, because weed serves it with authentication switched off entirely when it is empty, and a page that mints bucket credentials is the bucket. It is passed as WEED_ADMIN_PASSWORD rather than a flag so it stays off the process command line, and SEAWEED_ADMIN_BIND is the knob a remote host needs, named after PB_BIND and API_BIND for the same reason. Master, volume and filer share one /data mount rather than taking three of their own. That is precisely the layout `weed server -dir=/data` writes — the master's raft state, the volume's .dat and .idx, the filer's filerldb2, no two of them naming the same file — so a stack can move between the single-process file and its twin in either direction with nothing to migrate. A second volume server would need its own, and the files say so where somebody would go looking. The dev files map the volume server to 8081 on the host: 8080 there is already the API Server, and in the all-in-one it is the API Server inside the image. Unexercised: written on a machine without Docker, so none of the four has been brought up. Every flag, health path and env name was read out of the pinned 4.45 source rather than recalled — -mdir, -volumeSizeLimitMB, -defaultStoreDir, -max, admin's -master and -dataDir and WEED_ADMIN_*, the filer's and gateway's /healthz, the panel's unauthenticated /health — and the four files were parsed, interpolated against their examples, and checked for duplicate host ports. A `docker compose config` on the target host is still the first thing to run. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
tajniak81andClaude Opus 5
|
9a2a4ab72e |
The files leave the volume the database sits on
Every attachment — a document scan, a fuel receipt, a workshop invoice, a
photo of a part's box — has lived inside pb_data, in a directory beside the
SQLite file. One volume held both, so neither could be sized, backed up or
moved without the other. PocketBase can keep those bytes in an S3 bucket
instead, and now it is told to.
Nothing on the way to a client changes, because an attachment was never a
storage URL to begin with: it is fetched from GET /api/{records}/{id}/file,
which re-checks car access and asks PocketBase for the bytes as the service
account. PocketBase streams from the bucket through that same endpoint rather
than redirecting to it, so the web app, the phone and the plugin cannot tell
which side of the switch they are on.
The bootstrap that already creates the collections now writes PocketBase's
files-storage settings too, from PB_S3_*, on every boot and only when they
differ from what is already there — then asks PocketBase to prove it can reach
the bucket, and says so in the log either way. Two asymmetries are deliberate.
A read of the settings masks the stored secret, so a rotation of the secret
alone is invisible from here and needs another PB_S3_* to move with it. And it
never turns S3 back off: files already written to a bucket are reachable only
while PocketBase still points at it, so dropping the configuration would strand
them rather than undo anything.
Each deployment shape is one compose file with an .env example of the same
name, not a base plus an overlay to remember — six of each per folder, for
Docker and Docker-AIO alike: the plain one, .seaweedfs, .s3, and the three prod
twins. The SeaweedFS files run master, volume, filer and gateway as one process
and a one-shot init container beside it, because PocketBase never issues a
CreateBucket and SeaweedFS will not conjure one on first upload. The credentials
do double duty there — the gateway's only identity is also what PocketBase
authenticates with. In the all-in-one that gateway is a second container rather
than a fourth process under supervisord: keeping the object store inside the
image, on the volume the files are being moved off, would have defeated the
point and would have meant rebuilding.
Files uploaded before the switch are not carried across; PocketBase copies
nothing, and both READMEs say so where an operator will read it.
The TLS overlay and its Caddyfile go. The section they served stays, without
them: nothing in the stack terminates TLS any more, so it now names the four
variables to set in front of whichever proxy already does — TRUST_FORWARDED_PROTO
being the one that decides whether a charger is believed about how it arrived.
Unexercised: this was written on a machine without Docker, so the pinned
SeaweedFS image, the bucket-create and the settings write have not been run
against a live stack. The Go side builds, vets and tests clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
||
|
tajniak81andClaude Opus 5
|
9f5c8dc49a |
The address chargers are given is now an address that answers
The API Server tells a charger to dial the host it was itself asked on. The panel asks through the Web App, so the address handed out is the Web App's — which proxied /api/ and nothing else, and answered the WebSocket handshake at /ocpp/ with index.html. A charger pointed at the endpoint the screen showed could never connect to it, and the screen went on saying "Not connected" without a hint as to why. Both front doors now carry /ocpp/ through to the API Server: the BFF via the same reverse proxy, which relays the 101 by hijacking, and the all-in-one image's nginx via a location of its own, with timeouts long enough for a session that is idle between heartbeats. The proxied hop also has to say how the client arrived, since the API Server reads X-Forwarded-Proto to decide a charger reached it over TLS. That header is set from this server's own connection and overwrites whatever came in: believing a client on that point would let a plaintext charger claim wss and walk past OCPP_REQUIRE_TLS. TRUST_FORWARDED_PROTO opts into the inbound value for the one deployment where it is true — TLS ending at a proxy in front of the stack. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
tajniak81andClaude Opus 5
|
9258532952 |
Docker: give the panel's settings screens a permanent home in .env
|
||
|
tajniak81andClaude Opus 5
|
ee4ac441be |
Plugins: drop the plugins.json migration, and the volume it needed
The project has no public installs, so there is nothing to migrate from. MigrateLegacyFile, the file-backed Store it read through, PLUGINS_FILE and the legacy path threaded through the Server all go. What is left is one store, PocketBase, and a plugins package that touches no filesystem at all. That was the last thing keeping api_data alive, so the volume goes too. All four compose files now declare exactly one volume, pb_data, and the standalone API Server compose declares none - it talks to an external PocketBase and has nothing of its own to keep. Backing up the stack is backing up one path again. Both images get simpler for it. The API Server image loses VOLUME /data and the su-exec entrypoint that existed only to fix a mounted volume's ownership, so it goes back to a plain USER app; its working directory is now /app and holds nothing. The AIO image loses its second volume and chowns only /pb/pb_data. One consequence worth stating plainly, because it is a small regression rather than a no-op. The panel's Settings -> PocketBase and Settings -> Web App screens write .env in the working directory, which is now ephemeral. In the multi-container stack that changes nothing: compose sets all five of those keys as container environment, and loadDotEnv only applies a key that is not already set, so the file could never win a restart there anyway. In the AIO image it did win for POCKETBASE_ADMIN_EMAIL/_PASSWORD, which are not in that container's environment - so a service account fixed from the panel now lasts only until the container is recreated. Both READMEs say so. Moving those two screens into the app_settings singleton would close it properly; the PocketBase URL and credentials cannot follow, since they are how the database is reached in the first place. go build, go vet and go test ./... pass; the compose files parse and each resolves to a single pb_data volume. Not verified: no Docker CLI here, so neither image was built. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |
||
|
tajniak81andClaude Opus 5
|
660af5736a |
Plugins: create the settings collection instead of waiting for it forever
|
||
|
tajniak81andClaude Opus 5
|
01a8fecf40 |
Docker: stop telling operators to turn off the bootstrap that upgrades them
Every deployment file advised setting PB_BOOTSTRAP=false "once the database
is established". That was harmless while the schema was static. It stopped
being harmless in
|
||
|
tajniak81andClaude Opus 5
|
9bd5c523c4 |
Plugins: the global layer moves into the database, beside the other two
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> |
||
|
tajniak81andClaude Opus 5
|
f08849e50c |
Docker: a build context that isn't 3.6GB, and health you can see
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> |
||
|
tajniak81andClaude Opus 5
|
9487de84b0 |
Docker AIO: a folder name without a space in it
The all-in-one folder is now Docker-AIO, so -f Docker-AIO/Dockerfile resolves without quoting. Every path that pointed at the old name follows it: the compose build stanza, the documented build commands, and the links from the two READMEs. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> |