Component(s)
Dashboards, Panels, Documentation
What is missing? Please describe.
Perses Community Mixins does not currently include a dashboard for operating an AI gateway. I would like to contribute one AISIX AI Gateway Overview dashboard based on the Prometheus metrics exposed by the open-source AISIX gateway.
The dashboard would help Perses users answer a small set of operational questions without requiring a custom exporter or recording rules:
- What is the request rate and success ratio?
- What are the P95 end-to-end latency and streaming time to first token?
- How many input and output tokens are reported by model?
- Are cache, rate-limit, and guardrail policies taking effect?
- Is OTLP telemetry delivery failing or dropping data?
I propose keeping the first contribution to one dashboard with eight panels and job and instance filters. It would be implemented with the Perses Go SDK and generated in both native Perses and Perses Operator formats, following the repository's existing component dashboards.
The source metrics are documented in the AISIX metrics reference. I have already exercised the proposed PromQL against AISIX 1.2.0 and Prometheus 2.53.0 using synthetic traffic. Before opening a PR, I would rerun the queries with the exact Perses variables, import the generated dashboard into Perses 0.54.0, and include a synthetic-data screenshot.
Would an AISIX-specific dashboard be in scope for this repository? If so, is one overview dashboard the right initial size?
Disclosure: I work at API7.ai, which maintains AISIX. AISIX is an Apache-2.0-licensed independent open-source project and is not an Apache Software Foundation project.
Describe alternatives you've considered.
AISIX already has a Grafana dashboard, but submitting that JSON would not provide a native or maintainable Perses asset. I also considered adding only static YAML, but the repository uses Go SDK sources and generated dashboard resources. A generic AI gateway dashboard is not a good first contribution because metric names and policy semantics are not standardized across gateways.
If product-specific dashboards are outside the repository's scope, I would stop rather than introduce a new collector, exporter, or generic abstraction solely for this contribution.
Environment Information
Perses Version: 0.54.0
Community Mixins commit: 2c848c3c5eb5d9d9389522b22c4531995a5fa5e1
Go Version: 1.26.5, as pinned by the repository
AISIX Version: 1.2.0
Prometheus Version used for the existing query validation: 2.53.0
Component(s)
Dashboards, Panels, Documentation
What is missing? Please describe.
Perses Community Mixins does not currently include a dashboard for operating an AI gateway. I would like to contribute one
AISIX AI Gateway Overviewdashboard based on the Prometheus metrics exposed by the open-source AISIX gateway.The dashboard would help Perses users answer a small set of operational questions without requiring a custom exporter or recording rules:
I propose keeping the first contribution to one dashboard with eight panels and
jobandinstancefilters. It would be implemented with the Perses Go SDK and generated in both native Perses and Perses Operator formats, following the repository's existing component dashboards.The source metrics are documented in the AISIX metrics reference. I have already exercised the proposed PromQL against AISIX 1.2.0 and Prometheus 2.53.0 using synthetic traffic. Before opening a PR, I would rerun the queries with the exact Perses variables, import the generated dashboard into Perses 0.54.0, and include a synthetic-data screenshot.
Would an AISIX-specific dashboard be in scope for this repository? If so, is one overview dashboard the right initial size?
Disclosure: I work at API7.ai, which maintains AISIX. AISIX is an Apache-2.0-licensed independent open-source project and is not an Apache Software Foundation project.
Describe alternatives you've considered.
AISIX already has a Grafana dashboard, but submitting that JSON would not provide a native or maintainable Perses asset. I also considered adding only static YAML, but the repository uses Go SDK sources and generated dashboard resources. A generic AI gateway dashboard is not a good first contribution because metric names and policy semantics are not standardized across gateways.
If product-specific dashboards are outside the repository's scope, I would stop rather than introduce a new collector, exporter, or generic abstraction solely for this contribution.
Environment Information