ServiceLens — the mesh, observed - Showcase Project

Search goes red at 3:07. The page says Search Service. Inventory stopped answering a few minutes earlier, and Search is just the one your users can see. You redraw who calls whom from env vars while someone is already asking for an update. ServiceLens starts from the GitHub repos and draws that mesh itself. It checks each service, and when one fails it pages the person on your on-call sheet, writes the cause from the health, the neighbors, and the logs, and opens a draft fix PR. The on-call engineer does not need an account. The email link is enough, and if nobody acknowledges, it escalates. When the service comes back, the incident closes and leaves a short note, so the next one is not a blank page. A service can register itself from CI, so the graph does not go stale the week after you set it up. After a deploy it calls the GET routes in the code and fails the run if production is not serving them. Reads each repo at a pinned commit and maps routes plus the env vars services use to call each other. Edges it is not sure about stay dashed until you confirm them. Health checks every minute over HTTP, TCP, Postgres, and Redis. Services it cannot reach send a heartbeat. Rules for down, slow, and error-heavy. One bad check does not page anyone, and a service only has one open incident. On-call from a sheet, then email and Slack, with acknowledge, escalation, quiet hours, and per-person severity. Log ingest, search, and a live tail. Warnings and errors from the open stay on the incident. A write-up from the last half hour of health, neighbor status, those logs, failed checks, and how the last similar incident on that service was closed. A draft fix PR from the real source files. It does not merge, one PR per repo per hour, and it stops if the file changed underneath it. Contract checks on the GET routes found in the code, by hand, on a schedule, or from CI. Drills that take a service down so you can watch the page, the write-up, and the notifications fire. A live graph, an in-app feed, owner / editor / viewer roles, and an audit log. An API and SKILL.md so CI or an agent can onboard the mesh.