fix false missing Hermes CLI detection after deployment
This commit is contained in:
@@ -33,6 +33,23 @@ npm run build
|
||||
- `npm run test:e2e` runs Playwright browser tests against a mocked BFF API.
|
||||
- `npm run build` type-checks and builds both client and server.
|
||||
|
||||
## Hermes CLI Detection
|
||||
|
||||
The installation probe runs `hermes --help` in the Web UI service environment.
|
||||
Some Hermes versions synchronously check upstream updates during `--version`, so
|
||||
version output is not a reliable installation/liveness probe. A timeout does not
|
||||
mean Hermes is absent and must not send users to the installation guide.
|
||||
|
||||
`GET /api/hermes/cli-status` reports `available`, `bin`, and `reason` (`ready`,
|
||||
`missing`, `permission_denied`, `timeout`, or `failed`). `?refresh=1` bypasses the
|
||||
completed-result cache while sharing an in-flight probe. Deploy validation must
|
||||
check this endpoint as well as HTTP health; a running Node process alone does not
|
||||
prove that its service user can execute Hermes.
|
||||
|
||||
When investigating, reproduce with the service's `HERMES_BIN`, `PATH`, `HOME`, and
|
||||
`HERMES_HOME`. Inspect the complete launcher/interpreter path before changing
|
||||
permissions. Do not use recursive permission changes to treat a probe timeout.
|
||||
|
||||
## One-Click Deploy
|
||||
|
||||
`deploy.sh` deploys a checkout to a host that runs the built server under
|
||||
|
||||
Reference in New Issue
Block a user