Uptime checks probe your HTTP endpoints from multiple AWS regions and alert you when an endpoint stops responding successfully. The probes run inside your own AWS account — Stacktape provisions a small prober Lambda in each monitoring region, so there is no external monitoring vendor and no per-check subscription fee. The Console shows live status, availability and latency history, and an incident timeline for every check.
When to use
Add an uptime check to every public endpoint whose downtime you want to hear about before your users do: production APIs, customer-facing web apps, and webhook receivers. Checks are cheap (a few Lambda invocations per minute), so the practical default is one check per public-facing service in every production stage.
When NOT to use
- Verifying business logic — an uptime check asserts that an endpoint responds; it does not run flows like sign-in or checkout. Use a synthetic test for that.
- Monitoring internal-only services — probers reach endpoints over the public internet. A VPC-internal service is not reachable.
- Catching application errors — error tracking groups runtime errors from logs; an endpoint can serve 200s while a background job fails.
Configuration
An uptime check is a standalone resource. The only required property is url:
A full example with assertions, evaluation thresholds, and notifications:
Without assertions, a probe succeeds when the response status is 2xx or 3xx. Without regions, Stacktape probes from three regions: the stack's region plus two distant ones.
notificationChannels accepts three kinds of destinations, none of which puts a credential in your configuration:
console-channel: a channel created on the Console's Channels page, referenced by name. Use it for Slack with your own bot token, Microsoft Teams, Discord and webhooks. See alert channels.slack-app: a channel of the Slack workspace connected through the Stacktape Slack app, such as#alerts.email: asenderand arecipientaddress.
Inline slack, ms-teams, discord and webhook channels, which alarms accept, are rejected on uptime checks. The Stacktape Console decides when a check is down and sends its notifications, so those channels would have to send their bot token or webhook URL to the Console. A $Secret() reference does not help: the Console cannot read secrets in your AWS account. Create the channel in the Console once and reference it instead:
Without any channel, incidents still appear in the Console but nothing is delivered.
Set enabled: false to pause a check without deleting its history.
How it works
- On deploy, Stacktape provisions one shared prober Lambda per monitoring region in your account and stores each check's definition there. Probes run every 30 or 60 seconds.
- A check is marked down only when a quorum of regions (at least two, when the check probes from two or more) is failing at the same time — a region counts as failing for up to ten minutes after its last failing report — and the failure has persisted for
consecutiveFailuresconsecutive evaluations. This prevents alerts from single-region network blips. - Recovery follows the same rule symmetrically:
consecutiveSuccessesconsecutive successful evaluations. - Down and recovery events open and resolve incidents, deliver to your notification channels, and are retried on delivery failure.
- If probing itself stops reporting (for example, the prober was deleted manually), the Console raises a monitoring-silent incident so a dead monitor never masquerades as a healthy endpoint.
Deleting the stack (or the check) removes the check's probing; the shared prober infrastructure stays idle in the account and costs nothing meaningful.
What leaves your AWS account
The probes run in your account, but the Console decides when a check is down, so it receives:
- On deploy: each check's definition: name, URL, method, interval, timeout, redirect setting, assertions, regions, failure and recovery thresholds, and its
notificationChannels: theconsole-channelname, theslack-appchannel name and ID, or the email sender and recipient. No channel credential is sent, because uptime checks accept only channels without one. - From every probe: the project, stage, check name and region, up or down, the HTTP status, the total latency, the DNS, connect, TLS and time-to-first-byte timings, a failure reason of at most 500 characters, and the TLS certificate's expiry date.
Response bodies are read only inside the prober to evaluate body-contains assertions and are not sent.
Limits
- Probed methods are
GETandHEAD. Checks with request bodies are not supported. - At most 100 uptime checks per stack, and at most 5 regions per check. A region's shared prober has a bounded probing budget per minute — with many slow or timing-out endpoints in one region (across all stacks), some probes can be skipped for that tick; keep high-
timeoutSecondschecks few. A probe that cannot get its full configured timeout is dropped rather than reported as down, and a check that stops getting probed entirely raises a monitoring-silent alert. - Response bodies are read up to 512 KB for
body-containsassertions; the asserted text itself is capped at 2000 characters. - Uptime checks are skipped in dev mode — ephemeral dev stacks would only produce noise.
Viewing results
The Console's Uptime page lists every check with its live status. Opening a check shows availability and latency history per region and the incident timeline.