TSDB query backends

pkg/tsdb/ hosts the Go-side query implementations for the datasource plugins that ship with Grafana. Each pkg/tsdb/<name>/ is paired with a frontend plugin in public/app/plugins/datasource/<name>/.

Tour

Backend	Frontend plugin	Notes
prometheus/	prometheus (in packages/grafana-prometheus)	The most-used datasource. Streams instant + range queries via the Prometheus HTTP API.
loki/	loki	LogQL queries. Includes log line iteration, labels, log volume.
tempo/	tempo	Distributed traces (via the Tempo API and TraceQL).
jaeger/	jaeger	Jaeger traces.
zipkin/	zipkin	Zipkin traces.
cloudwatch/	cloudwatch	AWS CloudWatch metrics, logs, and X-Ray.
azuremonitor/	azuremonitor	Azure metrics, logs, traces, resource graph.
cloud-monitoring/	cloud-monitoring	Google Cloud Monitoring (Stackdriver).
mysql/	mysql	MySQL/MariaDB SQL queries.
mssql/	mssql	Microsoft SQL Server.
grafana-postgresql-datasource/	grafana-postgresql-datasource	PostgreSQL.
influxdb/	influxdb	InfluxDB v1 (InfluxQL) and v2/v3 (Flux/SQL).
opentsdb/	opentsdb	OpenTSDB.
graphite/	graphite	Graphite.
parca/	parca	Parca profiles.
grafana-pyroscope-datasource/	grafana-pyroscope-datasource	Pyroscope continuous profiling.
grafana-testdata-datasource/	grafana-testdata-datasource	Test data generator (used in dev / e2e).
grafanads/	"-- Grafana --" datasource	Built-in query types: random walk, dashboard list, annotations.

In addition, pkg/expr/ registers itself as the __expr__ datasource so that expression queries can flow through the same orchestration pipeline.

Standard structure of a backend

A typical backend has:

pkg/tsdb/<name>/
├── plugin.go            # ResourceHandler / DataHandler / StreamHandler registration
├── service.go           # ProvideService(...) constructor
├── client.go            # Upstream client (HTTP/SDK)
├── querydata.go         # QueryData implementation (the hot path)
├── healthcheck.go       # Datasource health check
├── resource.go          # Custom REST resources (e.g. metric metadata)
├── kinds/               # CUE schemas (for query types)
└── *_test.go            # Unit tests

The plugin SDK (grafana-plugin-sdk-go) provides the gRPC plumbing; backends only need to implement the handlers.

Cross-cutting tools

• Macros / variables — datasources that support templating (Prometheus interval/range, SQL ${date}, …) implement macro expansion before sending queries upstream. Helpers in pkg/tsdb/grafana-postgresql-datasource/sqleng/ and pkg/tsdb/prometheus/intervalv2/ are the typical reference points.
• Auth helpers — Azure SDK credentials, AWS SDK config, GCP service-account JWT all live near their datasources. Shared OAuth refresh helpers in pkg/services/oauthtoken/ are reused for upstreams that need Grafana's signed-in OAuth identity.
• Streaming — Loki and Live use the SDK's StreamHandler to stream live tail / live updates over gRPC and on to the WebSocket.

SQL datasources

mysql, mssql, grafana-postgresql-datasource, and influxdb (v1) all use a shared SQL engine helper that handles row → DataFrame conversion, time formatting, and macro expansion. Look for the sqleng package within each datasource for the shared base and per-dialect overrides.

Adding a new built-in backend

1. Create pkg/tsdb/<name>/ with service.go, plugin.go, and your handlers.
2. Add a ProvideService constructor and a wire set; reference it from pkg/server/wire.go.
3. Register the plugin id in pkg/services/pluginsintegration/pluginstore/ (it's auto-registered for built-ins, but you may need to add it to a list of "core plugins").
4. Create the matching frontend plugin under public/app/plugins/datasource/<name>/.
5. Add e2e and unit tests.

In practice, new datasources are created as external plugins these days — they don't need to live in this repo unless they ship with Grafana itself.

See also

• Datasources & queries — orchestration above this layer.
• Plugin host — how plugins are loaded and dispatched.
• Built-in plugins — the matching frontend plugins.