Before you begin: Your container must be deployed and emitting metrics. Grafana access is provisioned by your Index Representative on request, so ask for it early rather than on the day you need it.
Once the rollout is complete, monitoring is yours. Index confirms your container started and that its health and readiness endpoints report healthy, but it does not judge whether your logic is producing the outputs you want. That call is yours, and you make it from the metrics your container emits.
Index collects your Prometheus metrics from /metrics and your logs from STDOUT and STDERR, and surfaces both in a hosted Grafana instance.
What to watch
| Signal | Why it matters | Where |
|---|---|---|
Latency |
| Grafana, Explore or your own dashboard |
Error rate | Any status code other than | Grafana, by |
Your business metrics | Whether your logic is doing what you intended: how often you return a mutation, which branch fired, how your model is scoring, your segment match rate. Only you can define these. | Grafana, from your own custom metrics |
Deployment state | Whether your latest change actually landed everywhere it should have. | Deployment Tracker dashboard, Index Managed folder |
Note: Prefer metrics over verbose logging. Metrics are what you should use to capture container performance, and logs are currently retained for a short window only. You can expose metrics beyond the required histogram, as long as total cardinality stays under 10,000.
Optimizing
Monitoring is also where the optimization loop lives. Once your container is stable and you can see how it performs against real traffic, you and Index iterate on it before settling on a target QPS
Queries Per Second (QPS). The number of bid requests a DSP processes per second. Also known as impressions per second. for production. There are three levers:
Model data. Iterate on the data your logic reads without rebuilding or redeploying the image. See Updating models in containers without rebuilding.
The thresholds you are optimizing against. See the Service contract and Circuit breaker logic sections in Build a gRPC RTD service.
Your service architecture. Changes here mean a rebuild, so they are worth getting right early. See the Architectural patterns section in Building a Docker container.
Topics in this section
Access your Grafana data. Requesting access, signing in for the first time, and finding your metrics, dashboards, and logs.
