Inspect prometheus.yml for a scrape timeout longer than its interval, rule files with no Alertmanager, inline credentials and jobs that find no targets.
Waiting for prometheus.yml
No Prometheus server is contacted and no target is scraped.
The findings appear here
A scrape timeout longer than its interval, rule files loaded with nothing to notify, a credential readable on the /config endpoint, and jobs that find no targets.
This reads prometheus.yml as YAML. It does not contact a Prometheus server, reach any scrape target, or verify that service discovery resolves — so a configuration that reads cleanly may still scrape nothing.
Prometheus Configuration Inspector runs entirely on your device — credential values are reported by name only, and no Prometheus server or scrape target is contacted.
You’re on the Free plan. Compare plans
The finding that matters most is the quiet one: rule files loaded with no Alertmanager configured. The alerts evaluate, become firing, appear in the Prometheus UI, and notify nobody. There is no error anywhere, which is why this goes unnoticed for months — and is usually discovered during the incident the alert was meant to catch.
The timeout-versus-interval check is one Prometheus itself enforces: a scrape timeout longer than the interval would let one scrape still be running when the next begins, so Prometheus refuses to start. It is checked at both the global and per-job level, and a per-job interval inherits the global timeout, which is where the mismatch usually hides.
Credentials written into the file are worth flagging even on a private repository, because Prometheus serves its own resolved configuration on `/config`, which is unauthenticated by default. A basic_auth password in the file is readable by anything that can reach the Prometheus HTTP port. The `_file` variants read from disk at scrape time instead, and the value itself is never read here.
Missing `external_labels` gets a note because the consequence only appears later: without them, federating or remote-writing two Prometheus instances into one place produces series that collide and silently overwrite each other.
Nothing is contacted. No Prometheus server, no scrape target, and no check that service discovery resolves — so a configuration that reads cleanly may still scrape nothing at all.
Almost certainly no Alertmanager is configured. Prometheus evaluates the rules and marks them firing regardless, and shows them in its own UI — there is no error to notice. This reports rule files loaded without an alerting block as a warning for exactly that reason.
A common cause is a scrape timeout longer than the scrape interval, which would allow overlapping scrapes. It is easy to miss because a job can set its own interval while inheriting the global timeout, so neither value looks wrong on its own.
Yes, because Prometheus publishes its resolved configuration on the `/config` endpoint, unauthenticated by default. Anything that can reach the HTTP port can read it. `password_file` keeps the secret on disk and out of both the repository and the endpoint.