**Production:** `docker compose -f docker-compose.prod.yml --profile editor-tools up -d languagetool libretranslate` publishes **127.0.0.1:8010** and **127.0.0.1:5000** (loopback-only). Proxy those paths from Apache/nginx to the SPA origin, and bake the client with `LANGUAGE_TOOL_URL=/api/languagetool` and `TRANSLATE_URL=/api/translate` when running `./scripts/build-and-push-prod.sh`.
**Production:** `docker compose -f docker-compose.prod.yml pull && docker compose -f docker-compose.prod.yml up -d` starts the full stack (including LanguageTool on **127.0.0.1:8010** and LibreTranslate on **127.0.0.1:5000**). Run `bash scripts/ensure-libretranslate-dirs.sh` once on the server for LibreTranslate volume permissions. Proxy `/api/languagetool` and `/api/translate` from Apache/nginx to those ports, and bake the client with `LANGUAGE_TOOL_URL=/api/languagetool` and `TRANSLATE_URL=/api/translate` when running `./scripts/build-and-push-prod.sh`.
**Notes:** LanguageTool’s JVM image often needs **~1–2GiB** RAM. LibreTranslate **does not listen on port 5000 until models are ready**; without **`LT_LOAD_ONLY`** it may pull **many gigabytes** first, so the Vite proxy can show **`ECONNRESET` on `/translate`** while booting. Compose defaults **`LT_LOAD_ONLY`** to **ten** widely used codes (**en, de, es, fr, it, pt, ru, zh, ja, ar** — see `libretranslate` in `docker-compose.dev.yml`). Override with **`LT_LOAD_ONLY`** to add or remove codes; first start downloads packs for every listed code. **`LT_UPDATE_MODELS`** defaults to **`true`** so if you **expand**`LT_LOAD_ONLY` later, a **recreated** container still **installs missing** Argos packages into the bind-mounted `.local-libretranslate` tree (otherwise an older en/de-only cache sticks). Set **`LT_UPDATE_MODELS=false`** after everything is installed if you want faster routine restarts. Models are stored under **`.local-libretranslate/share`** and **`.local-libretranslate/cache`** (gitignored) with **bind mounts** so they survive **`docker compose down`**, image updates, and container recreate. **`scripts/ensure-libretranslate-dirs.sh`** (run automatically by **`npm run dev:all`**, **`npm run stack:remote`**, **`npm run docker:editor-tools`**, etc.) creates those dirs and **`chown`s them to UID 1032** via a short **Alpine** container so the LibreTranslate user can write. If you start **`libretranslate` by hand**, run **`npm run docker:prep-libretranslate`** once first. First download can still take **several minutes**; use **`docker logs -f jumble-libretranslate`** until **`curl http://127.0.0.1:5000/languages`** returns JSON. If logs show **`Cannot update models`** / **`Unavailable language codes: …`**, one bad token in **`LT_LOAD_ONLY` aborts the whole install** (you stay on whatever was already on disk, often en/de only). **Norwegian** must be **`nb`** (Bokmål), not ISO **`no`**. After you shrink **`LT_LOAD_ONLY`**, run **`npm run docker:prune-libretranslate-packages`** to remove leftover Argos package dirs under **`.local-libretranslate/share/argos-translate/packages`** (and unused **MiniSBD**`.onnx` files); the script briefly stops **`jumble-libretranslate`** or **`imwald-libretranslate`**. Override codes with **`LT_LOAD_ONLY=…`** on the same command if they differ from the compose default.
**Notes:** LanguageTool’s JVM image often needs **~1–2GiB** RAM. LibreTranslate **does not listen on port 5000 until models are ready**; without **`LT_LOAD_ONLY`** it may pull **many gigabytes** first, so the Vite proxy can show **`ECONNRESET` on `/translate`** while booting. Compose defaults **`LT_LOAD_ONLY`** to **ten** widely used codes (**en, de, es, fr, it, pt, ru, zh, ja, ar** — see `libretranslate` in `docker-compose.dev.yml`). Override with **`LT_LOAD_ONLY`** to add or remove codes; first start downloads packs for every listed code. **`LT_UPDATE_MODELS`** defaults to **`true`** so if you **expand**`LT_LOAD_ONLY` later, a **recreated** container still **installs missing** Argos packages into the bind-mounted `.local-libretranslate` tree (otherwise an older en/de-only cache sticks). Set **`LT_UPDATE_MODELS=false`** after everything is installed if you want faster routine restarts. Models are stored under **`.local-libretranslate/share`** and **`.local-libretranslate/cache`** (gitignored) with **bind mounts** so they survive **`docker compose down`**, image updates, and container recreate. **`scripts/ensure-libretranslate-dirs.sh`** (run automatically by **`npm run dev:all`**, **`npm run stack:remote`**, **`npm run docker:editor-tools`**, etc.) creates those dirs and **`chown`s them to UID 1032** via a short **Alpine** container so the LibreTranslate user can write. If you start **`libretranslate` by hand**, run **`npm run docker:prep-libretranslate`** once first. First download can still take **several minutes**; use **`docker logs -f jumble-libretranslate`** until **`curl http://127.0.0.1:5000/languages`** returns JSON. If logs show **`Cannot update models`** / **`Unavailable language codes: …`**, one bad token in **`LT_LOAD_ONLY` aborts the whole install** (you stay on whatever was already on disk, often en/de only). **Norwegian** must be **`nb`** (Bokmål), not ISO **`no`**. After you shrink **`LT_LOAD_ONLY`**, run **`npm run docker:prune-libretranslate-packages`** to remove leftover Argos package dirs under **`.local-libretranslate/share/argos-translate/packages`** (and unused **MiniSBD**`.onnx` files); the script briefly stops **`jumble-libretranslate`** or **`imwald-libretranslate`**. Override codes with **`LT_LOAD_ONLY=…`** on the same command if they differ from the compose default.
# First-time LibreTranslate bind mounts: bash scripts/ensure-libretranslate-dirs.sh (see scripts/README-deploy.md).
# Images built/pushed by ./scripts/build-and-push-prod.sh (app, monitor, piper-tts-proxy); others pull from Hub.
#
#
# Apache (unchanged on your host) should keep (order: specific paths before catch-all /):
# Apache (unchanged on your host) should keep (order: specific paths before catch-all /):
# ProxyPass /sites/ http://127.0.0.1:8090/sites/
# ProxyPass /sites/ http://127.0.0.1:8090/sites/
@ -14,10 +15,8 @@
# - Cron service `jumble-nip66-monitor` (Imwald NIP-66 monitor image) uses NIP66_MONITOR_NSEC to publish 30166/10166; nsec never goes to the client.
# - Cron service `jumble-nip66-monitor` (Imwald NIP-66 monitor image) uses NIP66_MONITOR_NSEC to publish 30166/10166; nsec never goes to the client.
# - Set NIP66_MONITOR_NPUB (npub1... derived from the same key) so the relay info page shows the monitor's avatar and handle in the NIP-66 liveliness section.
# - Set NIP66_MONITOR_NPUB (npub1... derived from the same key) so the relay info page shows the monitor's avatar and handle in the NIP-66 liveliness section.
- **Piper HTTP proxy:**`silberengel/imwald-piper-tts-proxy` (Wyoming Piper still pulls `silberengel/wyoming-piper` and `silberengel/wikistr` on the server.)
Registry paths keep the historical `imwald-jumble` name; retagging to e.g. `silberengel/imwald`is optional and requires updating `docker-compose.prod.yml` and pull scripts.
Registry paths keep the historical `imwald-jumble` name; retagging is optional and requires updating `docker-compose.prod.yml`.
## Remote server: one-time setup
## Remote server: one-time setup
@ -36,17 +37,38 @@ This builds both images and pushes two tags each (`latest` and the version from
NIP66_MONITOR_NPUB=npub1...
NIP66_MONITOR_NPUB=npub1...
```
```
4. **Once per machine** (LibreTranslate bind mounts need UID 1032 on the host dirs):
```bash
bash scripts/ensure-libretranslate-dirs.sh
```
## Remote server: pull and run
## Remote server: pull and run
Use a **current**`docker-compose.prod.yml` from this repo (`git pull` in the clone). If `docker compose pull` only lists two images, the file is outdated and **LanguageTool / LibreTranslate will never start**.
After you’ve pushed from local:
After you’ve pushed from local:
```bash
```bash
cd jumble
cd jumble
git pull
bash scripts/ensure-libretranslate-dirs.sh # once per host
docker compose -f docker-compose.prod.yml pull
docker compose -f docker-compose.prod.yml pull
docker compose -f docker-compose.prod.yml up -d
docker compose -f docker-compose.prod.yml up -d
```
```
The app is on **port 8089**. Both services use `:latest`; to pin a version, set the image in `docker-compose.prod.yml` to e.g. `silberengel/imwald-jumble:17.0.0` and `silberengel/imwald-jumble-nip66-monitor:17.0.0`.
That starts the **full** stack in `docker-compose.prod.yml` (app, NIP-66 monitor, OG proxy, Piper, LanguageTool, LibreTranslate). The SPA is on **port 8089**; see `docker-compose.prod.yml` header for Apache paths.
**Grammar + translate in the browser** also require the SPA to be built with API paths baked in, for example:
and Apache (or nginx) must proxy `/api/languagetool` → `127.0.0.1:8010` and `/api/translate` → `127.0.0.1:5000`.
**Shared host:** if you already run another `og-proxy` on `127.0.0.0.1:8090` or another Wyoming Piper on the same ports, `docker compose up -d` can fail with a port or name conflict. Either stop the duplicates or start only the pieces you need, e.g. `docker compose up -d jumble jumble-nip66-monitor languagetool libretranslate` (and point `PIPER_TTS_HOST` / Apache at your existing Piper stack if applicable).
Equivalent one-liner: `npm run stack:remote`
Images default to `:latest`; to pin a version, set image tags in `docker-compose.prod.yml`.
# Build main app and NIP-66 monitor images locally; push to silberengel/imwald-jumble and silberengel/imwald-jumble-nip66-monitor as :latest and :<version from package.json>.
# Build and push first-party production images (same tags :latest and :<version from package.json>):
