Status

Website

Our public-facing website is a Ruby on Rails application. We monitor this with a number of Outbound Probes, covering our homepage, a health check endpoint, and key HTTP protocol and subdomain redirects:

API Server

Our API server is written in Crystal and lives at https://api.heiioncall.com, receiving check-ins and state changes from your applications and APMs (see our docs). We monitor this with Outbound Probes as well:

Outbound Prober

Our Outbound Prober is a sharded cluster of Crystal processes. We use an Inbound Liveness trigger to make sure the entire shard space is being probed at all times. We also monitor a diverse control group of popular hosting providers so we can suppress false-positive alerts when it's more likely that our own connectivity is intermittent:

Background Job Queue

We use Sidekiq as a job queue for asynchronous and scheduled background jobs. We have a WorkerHeartbeatJob that hits an Inbound Liveness trigger to verify our scheduler and workers are running continuously:

ToolServ Microservice

We have an internal Crystal microservice that our Rails code calls for use in one-off queries. There are different approaches and tradeoffs to monitoring non-exposed internal services. Since we care that this microservice is accessible from Rails, we're monitoring this internal microservice by pointing an Outbound Probe at a Rails endpoint that makes a call to the ToolServ health check:

Deployed Version Convergence

Our production Rails app, API server, Sidekiq workers, Outbound Prober, and ToolServ are all built and deployed from one Git monorepo. While they will be temporarily running different versions during a rolling deployment, all production processes should eventually all be deployed from the same commit ID. We have a job that checks in with an Inbound Liveness trigger if there is exactly one commit ID currently running in production, ensuring we're alerted to any incomplete or inconsistent deployments that don't converge within a few minutes:

Postgres Database

We monitor some housekeeping tasks on our primary Postgres database, such as ensuring successful logical database backups, and verifying the correct state of schema migrations, primarily using Inbound Liveness triggers which get checked-in on success:

Miscellanous Background Jobs

Many of our on-call schedule rotation features are time-dependent, such as updating who's currently on call. These tasks run as scheduled jobs in Sidekiq. We monitor the completion of these important jobs with Inbound Liveness triggers:

External APIs

We monitor critical third-party APIs with scheduled jobs that check in with Inbound Liveness triggers once they verify connectivity and authentication success:

External Monitoring

We're in a special situation: we can't rely on Heii On-Call to monitor Heii On-Call itself, because we might not get alerted if something in the critical alerting flow breaks! Therefore, we monitor Heii On-Call with three external services running on distinct infrastructure in geographically distributed locations. We then monitor these external services with Inbound Liveness triggers so that we can get alerted if they stop probing our special health check endpoints. As long as one of these is up, we should be notified when any downtime occurs: