Fix observability stack

[Norbiox]

What was changed

✅ Loki 3.0 is running without errors and properly configured
✅ Prometheus successfully scrapes metrics from all Temporal services
✅ Grafana 12.2.1 is deployed with proper datasource configuration
✅ All 4 dashboards are provisioned and display Temporal metrics correctly
✅ Consistent YAML formatting across all provisioning files (camelCase field names)
✅ No compatibility or configuration errors in logs

Why?

Because of the breaking changes in Loki 3.0 mentioned in this issue. I've started with fixing the Loki configuration but it didn't helped and I still haven't seen any dashboards in Grafana.

Checklist

1. Closes 215

2. How was this tested:
   docker compose -f docker-compose-multirole.yaml up and then go to Grafana to ensure dashboards are working

3. Any docs updates needed?
   No

Detailed description of changes made and their explanations

1. Loki 3.0 Configuration Update

File: deployment/loki/local-config.yaml

Changes Made

1.1 Updated Schema Version (v9 → v13)

• Reason: Loki 3.0 requires schema v13 or newer for proper support and modern features like structured metadata and OTLP ingestion
• Change: Modified schema configuration from v9 to v13

1.2 Changed Index Store (boltdb → tsdb)

• Reason: BoltDB index type is deprecated. Loki 3.0 requires TSDB (Time Series Database) as the index type for modern features
• Change: Updated store: boltdb to store: tsdb in schema_config

1.3 Updated Storage Configuration

• Reason: Storage backend needs to match the new index type
• Changes:
  • Removed boltdb storage section with directory: /tmp/loki/index
  • Added tsdb_shipper configuration with:
    • active_index_directory: /tmp/loki/tsdb
    • cache_location: /tmp/loki/index_cache
    • cache_ttl: 24h

1.4 Optimized Index Period

• Reason: Smaller index periods improve performance and make index management more efficient
• Change: Updated index period from 168h to 24h

1.5 Added Compactor Configuration

• Reason: Loki 3.0 requires compactor configuration for proper operation. Previous version was missing this
• Changes: Added compactor section:

  compactor:
    working_directory: /tmp/loki/compactor
    compaction_interval: 10m

1.6 Removed Deprecated Fields

• Reason: These fields are no longer valid in Loki 3.0
• Removed:
  • enforce_metric_name from limits_config
  • Entire chunk_store_config section (with max_look_back_period)
  • Entire table_manager section (only used with DynamoDB)

---

2. Prometheus Configuration Fix

File: deployment/prometheus/config.yml

Changes Made

2.1 Updated Scrape Targets from host.docker.internal to Container Names

• Reason: Prometheus runs inside Docker and the Temporal services are also running as Docker containers on the same network. Using host.docker.internal (which points to the host machine) was incorrect.
• Changes: Updated targets in temporalmetrics job:
  • host.docker.internal:8000 → temporal-history:8000
  • host.docker.internal:8001 → temporal-matching:8001
  • host.docker.internal:8002 → temporal-frontend:8002
  • host.docker.internal:8003 → temporal-worker:8003
  • host.docker.internal:8004 → temporal-frontend2:8004

Result: Prometheus can now successfully scrape metrics from the Temporal services, and dashboards have access to the required metrics (service_requests, service_errors, etc.)

---

3. Grafana Datasource Configuration Fix

File: deployment/grafana/provisioning/datasources/all.yml

Changes Made

3.1 Fixed isDefault Field Format

• Reason: Grafana expects isDefault: true (camelCase) not is_default: true (snake_case)
• Change: Updated Prometheus datasource from is_default: true to isDefault: true

3.2 Standardized Field Naming (org_id → orgId)

• Reason: Consistency with Grafana's modern conventions and matching the dashboards provisioning file format
• Changes:
  • Updated org_id: 1 to orgId: 1 for both Prometheus and Loki datasources
  • Added explicit isDefault: false to Loki datasource for clarity

Result: Prometheus is now properly set as the default datasource, allowing dashboards with $datasource variable to work correctly. Both provisioning files now use consistent camelCase field naming.

---

4. Grafana Dashboard Provisioning Update

File: deployment/grafana/provisioning/dashboards/all.yml

Changes Made

4.1 Updated to New Dashboard Provisioning Format

• Reason: Grafana uses an updated dashboard provisioning configuration format with apiVersion and providers block
• Changes:
  • Added apiVersion: 1 at the top
  • Wrapped provider config in providers: block
  • Removed deprecated folder property (replaced with path)
  • Updated provider fields to match new format:
    • Changed org_id to orgId
    • Added disableDeletion: false and editable: true

Result: Eliminates deprecation warnings and ensures proper dashboard provisioning in Grafana 12.2.1

---

5. Grafana Version Upgrade

File: docker-compose-multirole.yaml

Changes Made

5.1 Updated Grafana Image Version

• Reason: Old version (7.5.16) could not parse dashboard queries designed for Grafana 8.0.4+, causing "Failed to upgrade legacy queries e.replace is not a function" errors
• Change: Updated image from grafana/grafana:7.5.16 to grafana/grafana:12.2.1

Result: Dashboards now load without compatibility errors and display data correctly

---

6. Grafana Volume Mounts Configuration

File: docker-compose-multirole.yaml

Changes Made

6.1 Added Missing Volume Mounts

• Reason: Dashboards and their provisioning configuration were not mounted to the Grafana container, so they were not loaded
• Changes: Added three volume mounts to the grafana service:

  volumes:
    - type: bind
      source: ./temporalio/deployment/grafana/provisioning/datasources
      target: /etc/grafana/provisioning/datasources
    - type: bind
      source: ./temporalio/deployment/grafana/provisioning/dashboards
      target: /etc/grafana/provisioning/dashboards
    - type: bind
      source: ./temporalio/deployment/grafana/dashboards
      target: /var/lib/grafana/dashboards

Result: All 4 dashboards are now properly provisioned and visible in Grafana:

• Temporal Server Metrics
• Temporal SDK Metrics
• PostgreSQL Database
• Docker Engine Metrics

[CLAassistant]

All committers have signed the CLA.