Docs
Open console

Intercepts

An intercept makes one service's traffic go to your workspace. Everything in the environment keeps dialling api:8080; the bytes arrive at the process you are running, on the port you choose. No rebuil

An intercept makes one service's traffic go to your workspace. Everything in the environment keeps dialling api:8080; the bytes arrive at the process you are running, on the port you choose. No rebuild, no redeploy, no proxy.

Set an intercept

The workspace must be in a session on the environment.

API
curl -sS -X POST https://dev.kloudlite.io/v1/environments/$ENV/intercepts \  -H "Authorization: Bearer $KL_TOKEN" -H 'content-type: application/json' \  -d '{    "service": "api",    "workspace": "ws-7f3a…",    "ports": [{ "service": 8080, "workspace": 3000 }]  }'

ports maps a port callers dial on the service to the port your process listens on in the workspace. Then run your dev server:

bash
pnpm dev --port 3000

Every request another service sends to api:8080 now reaches it.

What happens in the environment

  • The real service is scaled to zero, so a queue consumer in the old pod cannot eat messages your workspace never sees.
  • The service's address and DNS name are unchanged; callers change nothing.
  • Its endpoints point at your workspace. Traffic is routed by endpoints, never by a proxy in the path.

Release

API
curl -sS -X DELETE https://dev.kloudlite.io/v1/environments/$ENV/intercepts/api \  -H "Authorization: Bearer $KL_TOKEN"

The real service scales back up and its endpoints are restored.

Wish and effect

The intercept you set is a wish on the environment. What is in force is each service's intercepted_by, and the console shows only that. The two differ in one case: your workspace stops answering (stopped, deleted, node down, or its pod restarts and stays down past a grace period). Then the real service comes back on its own, the wish stays, and it takes hold again when the workspace returns. Only the DELETE above removes a wish.

Rules

  • One workspace per service. A second request is refused naming the holder.
  • An ordinary pod restart in the workspace does not bounce the service; only an outage past the grace period does.