Skip to content

Overview and the Map

Overseer opens on the whole stack, kept live, in one of three views (the toggle at the top, kept in the address as ?view=):

View Shows
Cards A card for each part, in the sidebar's order: its status and health line, and with containers each one's CPU and memory over the last 15 minutes and the way to its Container and Logs. Trouble first (the default) puts the parts not well at the top; Only trouble hides the rest
Map The Twin drawn: each process a node in its tier, the connections between them as lines, and what runs in the webserver as chips in the Server's node
Flow The stack's processes and stores left to right, the way work moves (through the queue to the worker), each in its shape

Overseer Overview, Cards view: a card per part with its health line and container charts

Overseer Overview, Map view: the Twin drawn, each process a node in its tier

Overseer Overview, Flow view: processes and stores left to right, through the queue to the worker

The Map

Each process is a node at a fixed place, so the map draws the same every time and nothing overlaps. A node links to its page, with its state and its containers' CPU and memory. Lines take the tone of what they lead to: dashed for queued work, dotted for a key Secrets holds. Data is drawn in its shape (a store, the queue, object storage).

Zoom. Open the Server to see what runs inside it, what each part reaches, and the outside providers whose key is not set (each with its key's dialog one click away). Open a service to see its pieces (router, facade, domains) and what each reaches. Every level reads the same way, the way a C4 diagram does.

Traffic. The Traffic chip tones each node by how much ran through it instead of its health: every call along the lines touching it, Overseer's own reads included, counted as it happens with no tracer.

What the code should fix. Opened up, a part or a piece that breaks the service standard is marked with how many places it should fix. Hover for what to do and each place, linked to its line in Code. The mark is about the code; the node's tone stays its health.

Replay a request

Replay a request (/overseer/replay) picks one of the last 200 requests the server answered and plays it on the map, a level at a time: the Server opened up where the request went past one service, then each service it ran in, each drawing only what the request went through.

Overseer > Replay: recent requests by route, with their count, slowest time and hops

Each hop it made (a service calling another, a query and its tables, a provider called, work queued) lights as it ran, at its real timing slowed down to watch, the slowest held and a failure flashed, and is said in words under the controls. The Hops view lists them in the order they ran, each at its line in Code.

A request's id rides every answer as X-Request-Id and every line it logs as request_id, so Routes' Execute answer and an error logged during a request link straight to its replay. The same requests are GET /api/v1/requests and Illiana's requests tool. Overseer's own pages are not kept.

Was this page helpful?