fix: drop the Alloy healthcheck that could only ever fail

[cnmaia]

Found while validating the deploy: datamap_alloy had been unhealthy for twenty hours.

exec: "wget": executable file not found in $PATH

The image carries a shell but no HTTP client, so the check was running a command that does not exist. It never had a chance to pass.

It was working the whole time

alloy      http://datamap_alloy:12345/metrics        up
gatekeeper http://datamap_gatekeeper:9095/metrics    up
gatekeeper http://datamap_gatekeeper_b:9095/metrics  up
loki       http://datamap_loki:3100/metrics          up
prometheus http://localhost:9090/metrics             up

And Loki holds the logs of all eleven datamap_* containers, and none of the three neighbouring projects' — so the discovery, the filter and the shipping are all fine.

Why I missed it

Loki had exactly this problem and I fixed it when its image turned out to be distroless. I did not check Alloy the same way: the local run only ever showed health: starting before I moved on to the next thing, so the check never had time to fail in front of me.

The lesson is the narrow one — health: starting is not healthy, and waiting for the difference is the whole point of looking.

What replaces it

Nothing new. The Prometheus scrape was already the real signal, and TargetDown was already the alert — which is why a container reporting unhealthy for a day changed nothing about whether the logs arrived.

🤖 Generated with Claude Code

https://claude.ai/code/session_01EQda9NZvkbStEeNU54Tqgh