CheckPulse
Multi-region uptime, DNS, and SSL monitoring for agencies. Deployed, public, and handling real traffic.
Every figure above is measured or read from the running system.
Technology Stack
What it does
CheckPulse monitors uptime, DNS resolution, and SSL certificate health for agencies that are contractually on the hook for a portfolio of client websites. Each client gets a brandable public status page; alerts fan out to Slack, Discord, Telegram, email, or a webhook.
It is deployed and publicly reachable at checkpulse.dev, running as seven containers behind a Cloudflare tunnel with no inbound ports open on the host.
The design decision I care about: consensus before alerting
Single-probe monitors cry wolf. A blip between one probe and one target produces a 3am page for an outage that never happened, and after enough false alarms people stop reading the alerts, which is worse than having no monitoring at all.
CheckPulse runs probes from three regions and only transitions a monitor to DOWN when at least two regions independently agree. That trades a small amount of detection latency for alerts that are worth waking up for. The check workers are Celery tasks; region agreement is resolved in the incident detector rather than in the probe, so adding a fourth region is a config change rather than a rewrite.
Running seven containers in 1.9 GB
The whole stack lives on one small VPS: API, worker pool, beat scheduler, Postgres, Redis, the tunnel connector, and the dashboard. Memory is the binding constraint, not CPU.
That constraint drove real choices: Postgres tuned down from its defaults, worker concurrency capped so Celery prefetch cannot balloon resident memory, and no per-service observability sidecars. It is a useful exercise in sizing a system for the box you actually have instead of the box you would like to have.
Incident: the signup form became someone else's email validator
Between April and July 2026, roughly 1,191 accounts were created by a bot walking an alphabetical list of corporate email addresses. It was not trying to use the product. It was using my verification email as an oracle: submit an address, see whether the mail bounces, learn whether the mailbox is real.
The damage was not CPU or storage, it was sender reputation. Three months of unsolicited verification mail went out from the transactional domain, with my DKIM signature on it.
The fix was a Cloudflare Turnstile gate on every endpoint that can trigger an outbound email, and a split between the transactional sending domain and any other mail. The fake accounts are quarantined rather than deleted so the abuse pattern stays auditable.
The general lesson: any unauthenticated endpoint that sends mail to an attacker-supplied address is an email validation service you are operating for free, and you will not notice from your own dashboards, because the traffic looks like growth.
Honest status
Three real user accounts, zero paying customers. Stripe is wired up with Free / $12 / $49 tiers and one checkout was started and abandoned.
I am listing it because the engineering is real and deployed, not because it has traction. It does not.