Galaxy

Galaxy is the dashboard we run our own infrastructure from. It exists because the honest answer to "how many servers do you have?" used to be a pause. Everything a fleet accumulates quietly — a box nobody logs into any more, a certificate three weeks from expiry, a domain whose auto-renew was turned off a year ago, a DNS record still aimed at a machine that was cancelled — is the kind of fact that is obvious the day after it costs you something. Galaxy's job is to know it the week before.

One inventory, checked rather than remembered

A single data file is the source of truth: every production machine, its role, what it carries, what it costs and when it renews, and every name pointed at it. The page is one HTML file with no build step — a map, a table and a set of panels rendered straight from that data — because a dashboard that needs a toolchain to open is a dashboard you stop opening. Probe scripts log into each machine and dump live state (disks, services, virtual hosts, certificates) in a form that can be reconciled against the inventory, so the file is audited against reality instead of drifting away from it.

The watcher, and why it stays quiet

A scheduled job re-checks the things that expire or leak: registrar expiry pulled from the registries themselves rather than from memory, dated hosting bills, certificate expiry and hostname mismatch, names still resolving to retired boxes, configuration and repository files reachable from a public webroot, and database ports listening to the internet. It sends mail only when a finding is new or urgent — an alert that arrives every week is an alert nobody reads by the third one.

Once a month it sends a digest whether or not anything is wrong, and that is deliberate: the digest is the heartbeat. A quiet watcher and a dead watcher look identical from the outside, so the schedule makes them look different. If it crashes it mails the traceback to itself.

Audits are dated files, not a feeling

Every sweep is written up and committed — what was checked, what was found, what was fixed the same day, and what was consciously accepted. That turns "we should tidy the servers up" into a list with dates on it, and it makes drift visible: an item that appears in three consecutive audits is a decision nobody has made rather than a task nobody has done.

What it refuses to guess

Each report names its own blind spots — machines the dashboard has no key for, settings that live in a vendor's control panel rather than on a box, and the checks it cannot run against the host it runs on. A monitoring tool that quietly skips what it can't see is worse than no monitoring at all, because it produces confidence instead of coverage.

It is also where this site's numbers come from

The fleet figures on our home page — how many boxes, how many metros, how many live names — are counted out of Galaxy's ledger rather than rounded up from memory, which is why they occasionally go down. A box in Florida died this year and the number on the page moved accordingly.