Telemetry readers
Something polling /api/v1/telemetry too often shows up here. The owner's /system page has a Telemetry readers section, and the terminal's System dialog (palette) shows the same rows under the history lines. Both are for a forge admin who is a person; nobody else gets the section.
- What a row says. Route (
flow,system, …), reader (the account or token name), reads in the window, and the mean interval between the seconds in which it read. Most reads first. A reader with a single read second has no interval (—). - The window. The last 10 minutes, counted per process in one-second slots (at most 600 per reader and route); older reads drop out, and a restart empties it.
- The 5 s rule. A reader on
flowwhose mean interval is under 5 s is flagged ("faster than the flow copy changes" on the page,TOO FASTin the terminal): the flow view does not change that often. - The 200-name cap. At most 200 route/reader pairs are tracked; a new pair beyond that is counted under reader
other, so no read goes uncounted. - Not changed.
/metricskeepsforge_telemetry_reads_total{route,reader}; the window is the page's and terminal's alone. The terminal lists the top six rows and says how many more. - Agents. The same rows (
route,reader,reads,meanIntervalMs,tooFast, most reads first) come asreadersin the MCPsystemtool and inGET /api/v1/telemetry/system, under the same gate as the rest of that answer: a forge admin, or an agent grantedtelemetry:read. A read is counted after its answer is built, so it is not in its ownreaders.