Liveness from Vector scrapes, request rate from each service's /metrics, container CPU/memory from cadvisor. Up % is computed against the last hour of metrics_obs.service_up.
| Service | Up · 1h | Samples | RPS | Err % | CPU % | Mem · MB |
|---|---|---|---|---|---|---|
| gateway | no data | 0 | — | — | — | — |
| portal | no data | 0 | — | — | — | — |
| executor | no data | 0 | — | — | — | — |
| clickhouse | no data | 0 | — | — | — | — |
| cadvisor | no data | 0 | — | — | — | — |
Total active time-series in metrics_obs.metrics_raw. Budget < 20,000 — sized to cover MAIVEN Warehouse's native /metrics (~3.2k names) plus cadvisor per-container series. See scripts/verify/ops.py for the hard limit. If this climbs sharply, look for unbounded labels (request_id, raw path, user).