grafana/pkg/operators/provisioning
Dominik Prokop e266c88505 Merge main into feat/faro-tracing-integration-part1
- Sync feature branch with latest main branch changes
- Resolve conflicts in dashboard scene serialization imports
- Include both existing and new profiler functions for compatibility
- Refactor dashboard performance profiling architecture
- Create shared performanceUtils.ts with memory utilities and grouped logging
- Convert observer methods to arrow functions in analytics and logger classes
- Add type-safe performance.memory access with proper TypeScript interfaces
- Standardize performance logging with writePerformanceGroup utilities
- Convert duration logging from strings to numeric values for better analysis
- Fix localStorage usage to use @grafana/data store for consistency
- Update dashboard scene integration to use improved profiling system
2025-10-07 22:28:34 +02:00
..
README.md Provisioning: Remove temporary logic to test clients in jobs operator (#110758) 2025-09-08 12:15:13 +00:00
config.go Merge main into feat/faro-tracing-integration-part1 2025-10-07 22:28:34 +02:00
repo_operator.go Merge main into feat/faro-tracing-integration-part1 2025-10-07 22:28:34 +02:00

README.md

Provisioning Controllers

Git sync has two different controllers: the jobs controller and the repo controller.

Jobs Controller

Behavior

  • Watches provisioning Jobs and emits notifications on job creation.

  • Optionally cleans up HistoricJobs after a configurable expiration. Disable when job history is stored in Loki.

  • Queueing and claiming:

    • Creating a Job enqueues work. Drivers “claim” one job at a time under a time-bound lease so only one worker processes it at once.
    • If a driver crashes or loses its lease, cleanup makes the job eligible to be claimed again. This yields at-least-once processing.
    • New job notifications reduce latency; periodic ticks ensure progress even without notifications.
  • Processing and status:

    • A supporting worker processes the job, renewing the lease in the background. If lease renewal fails or expires, processing aborts.
    • Status updates are persisted with conflict-aware retries. Progress is throttled to avoid excessive writes while still providing timely feedback.
    • When processing finishes, the job is marked complete and a copy is written to history.
  • Historic jobs role:

    • Historic jobs are a read-only audit trail and UX surface for recent job outcomes, progress summaries, errors, and reference URLs.
  • Retention is implementation-dependent: this controller can prune old history objects periodically, or history can be stored in Loki; when using Loki, disable local cleanup with --history-expiration=0.

This binary currently wires informers and emits job-create notifications. In the full setup, concurrent drivers consume notifications and execute workers to process jobs using the behavior above.

Flags

  • --token (string): Token to use for authentication against the provisioning API.
  • --token-exchange-url (string): Token exchange endpoint used to mint the access token for the provisioning API.
  • --provisioning-server-url (string): Base URL to the provisioning API server (e.g., https://localhost:6446).
  • --history-expiration (duration): If greater than zero, enables HistoricJobs cleanup and sets the retention window (e.g., 30s, 15m, 24h). If 0, cleanup is disabled.

TLS Configuration

  • --tls-insecure (bool): Skip TLS certificate verification. Default: true (for development/testing).
  • --tls-cert-file (string): Path to TLS client certificate file for mutual TLS authentication.
  • --tls-key-file (string): Path to TLS client private key file for mutual TLS authentication.
  • --tls-ca-file (string): Path to TLS CA certificate file for server certificate verification.

How to run

  1. Build grafana:
    • make build
  2. Ensure the following services are running locally: provisioning API server, secrets service API server, repository controller, unified storage, and auth.
  3. Create a operator.ini file:
[database]
ensure_default_org_and_user = false
skip_migrations = true
[operator]
provisioning_server_url = https://localhost:6446
tls_insecure = true

[grpc_client_authentication]
token = ProvisioningAdminToken
token_exchange_url = http://localhost:6481/sign/access-token
# Uncomment to enable history cleanup via Loki. First ensure the Provisioning API is configured with Loki for job history (see `createJobHistoryConfigFromSettings` in `pkg/registry/apis/provisioning/register.go`).
# history_expiration = 24h  
  1. Start the controller:
  • GF_DEFAULT_TARGET=operator GF_OPERATOR_NAME=provisioning-jobs ./bin/darwin-arm64/grafana server target --config=conf/operator.ini

TLS Configuration Examples

  • Production with proper TLS verification:
[operator]
provisioning_server_url = https://localhost:6446
tls_insecure = false
tls_ca_file = /path/to/ca-cert.pem

[grpc_client_authentication]
token = ProvisioningAdminToken
token_exchange_url = http://localhost:6481/sign/access-token
  • Mutual TLS authentication:
[operator]
provisioning_server_url = https://localhost:6446
tls_insecure = false
tls_ca_file = /path/to/ca-cert.pem
tls_cert_file = /path/to/client-cert.pem 
tls_key_file = /path/to/client-key.pem

[grpc_client_authentication]
token = ProvisioningAdminToken
token_exchange_url = http://localhost:6481/sign/access-token
  • Development with self-signed certificates (insecure):
[operator]
provisioning_server_url = https://localhost:6446
tls_insecure = true

[grpc_client_authentication]
token = ProvisioningAdminToken
token_exchange_url = http://localhost:6481/sign/access-token

Expected behavior

  1. Create a repository and enqueue a job (note that the repository must be marked as healthy):

export ACCESS_TOKEN=$(curl -X POST http://localhost:6481/sign/access-token \
  -H "X-Realms: [{\"type\":\"system\",\"identifier\":\"system\"}]" \
  -H "X-Org-ID: 0" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer ProvisioningAdminToken" \
  -d '{
    "namespace": "*",
    "audiences": ["provisioning.grafana.app"]
  }' | jq -r '.data.token')

curl -X POST https://localhost:6446/apis/provisioning.grafana.app/v0alpha1/namespaces/default/repositories/test6/jobs \
  -H "Content-Type: application/json" --insecure \
  -H "X-Access-Token: Bearer $ACCESS_TOKEN" \
  -d '{
    "action": "pull",
    "pull": {
      "incremental": false
    }
  }'
  1. The controller emits a notification on job creation.
➜  job-controller git:(feature/standalone-job-controller) ✗ ./bin/job-controller --token-exchange-url=http://localhost:6481/sign/access-token --token=ProvisioningAdminToken --provisioning-server-url=https://localhost:6446
{"time":"2025-08-21T14:27:03.789337+02:00","level":"INFO","msg":"job create notification received","logger":"provisioning-job-controller"}

  1. In a full setup with the concurrent driver, workers claim and process jobs, updating status and writing history.
  2. Entries move to HistoricJobs; if cleanup is enabled, older entries are pruned based on --history-expiration.

[WIP] Repository Controller

This controller is responsible for watching repositories. It will eventually do health checks, queue sync jobs, and create/delete github hooks.

To run locally, run GF_DEFAULT_TARGET=operator GF_OPERATOR_NAME=provisioning-repo ./bin/darwin-arm64/grafana server target --config=conf/operator.ini