dev-health-ops

Open source analytics for developer health and team operating modes.

Dev Health Ops ingests engineering activity from Git providers, work trackers, deployments, incidents, and local repositories; normalizes it into persisted evidence; computes inspectable metrics; and serves those metrics to the GraphQL/API layer used by dev-health-web.

Why this exists

Developer health tooling often drifts into expensive, opaque scorecards that are easy to misuse. This project is intentionally different:

• Accessibility over extraction: derive insight from data teams already own.
• Learning, not judgment: show operating signals, not individual rankings.
• Trends over absolutes: emphasize change over time and distributions.
• Inspectable by default: metrics trace back to schemas, queries, and evidence.

Non-goals:

• Individual leaderboards or performance scores
• HR/performance-management workflows
• Dashboards that hide definitions, provenance, or missing data

Current architecture

Dev Health Ops follows a strict pipeline boundary:

Providers → Processors → Sinks → Metrics → API / Visualization

• Providers fetch raw provider data from GitHub, GitLab, Jira, Linear, local Git, CI/CD, deployments, incidents, and synthetic/demo sources.
• Processors normalize provider records into internal models.
• Sinks persist computed outputs. Analytics persistence is ClickHouse-only.
• Metrics jobs compute daily rollups, DORA, complexity, risk, investment, AI workflow, and work graph outputs from persisted data.
• API/GraphQL serves persisted analytics to dev-health-web and other consumers.

Python remains the owner of the API, GraphQL schema, provider fetch and normalization, processors, and the job implementations the Go fleet runs. The repository also contains an additive Go worker-runtime foundation under cmd/ and internal/; adding those process shells does not move a job out of Python or change its routing.

The primary visualization surface is now dev-health-web. Grafana is optional, and this repository no longer ships the old sample dashboard gallery in this README.

Install

Use the package directly:

pip install dev-health-ops
dev-hops --help

For local development from this repository:

pip install -r requirements.txt

The installed command is dev-hops.

Database model

Dev Health Ops uses two databases with different responsibilities:

Layer	Backend	Environment variable	Purpose
Semantic	PostgreSQL	POSTGRES_URI	Users, organizations, settings, credentials
Analytics	ClickHouse	CLICKHOUSE_URI	Commits, PRs/MRs, work items, metrics, graph data

ClickHouse is required for analytics features. MongoDB, SQLite, and PostgreSQL analytics sinks have been removed or deprecated; SQLite remains only for narrow test/local fixture paths.

Start local services and run migrations:

docker compose up -d postgres clickhouse valkey

export POSTGRES_URI="postgresql+asyncpg://postgres:postgres@localhost:5555/postgres"
export CLICKHOUSE_URI="clickhouse://ch:ch@localhost:8123/default"

dev-hops migrate postgres
dev-hops migrate clickhouse

See docs/contribute/architecture/data-and-storage.md and docs/reference/cli/index.md for details.

Common workflows

Sync source data

# Local git repository
CLICKHOUSE_URI="clickhouse://ch:ch@localhost:8123/default" \
  dev-hops sync git --provider local --repo-path /path/to/repo

# GitHub repository
CLICKHOUSE_URI="clickhouse://ch:ch@localhost:8123/default" \
  dev-hops sync git --provider github \
  --auth "$GITHUB_TOKEN" \
  --owner <owner> \
  --repo <repo>

# Pull requests
CLICKHOUSE_URI="clickhouse://ch:ch@localhost:8123/default" \
  dev-hops sync prs --provider github \
  --auth "$GITHUB_TOKEN" \
  --owner <owner> \
  --repo <repo>

# Work items from Jira, GitHub, GitLab, Linear, synthetic data, or all providers
CLICKHOUSE_URI="clickhouse://ch:ch@localhost:8123/default" \
  dev-hops sync work-items --provider all --backfill 30

# Teams into the ClickHouse team catalog (ClickHouse is the system of record)
CLICKHOUSE_URI="clickhouse://ch:ch@localhost:8123/default" \
  dev-hops sync teams --provider config --path src/dev_health_ops/config/team_mapping.yaml --allow-empty

The bundled team_mapping.yaml is an empty onboarding sample. sync teams exits non-zero when no teams are persisted; pass --allow-empty only for intentional empty/no-op syncs such as validating the sample config.

Provider authentication can come from CLI flags or environment variables such as GITHUB_TOKEN, GITLAB_TOKEN, JIRA_*, ATLASSIAN_*, and LINEAR_API_KEY.

Compute metrics

# Daily analytics rollups
CLICKHOUSE_URI="clickhouse://ch:ch@localhost:8123/default" \
  dev-hops metrics daily --backfill 30

# Complexity and hotspot snapshots for a repository
CLICKHOUSE_URI="clickhouse://ch:ch@localhost:8123/default" \
  dev-hops metrics complexity --repo-path /path/to/repo --backfill 30

Generate demo data

dev-hops fixtures generate \
  --sink "clickhouse://ch:ch@localhost:8123/default" \
  --days 30 \
  --with-metrics \
  --with-work-graph

Run the API

POSTGRES_URI="postgresql+asyncpg://postgres:postgres@localhost:5555/postgres" \
CLICKHOUSE_URI="clickhouse://ch:ch@localhost:8123/default" \
  dev-hops api --reload

OpenAPI docs are available at http://localhost:8000/docs when the API is running. GraphQL is served by the API for the web app.

Run workers

Celery is retired (CHAOS-4026, 2026-08-21): zero Python celery services run in prod since the 2026-08-19 stop. Go owns every periodic maintenance cadence. dev-hops workers start-worker/start-scheduler were deleted along with it — see docs/operate/run/workers-and-jobs.md for how to run the Go worker/scheduler/reconciler/stream-runner processes.

Go River processes use POSTGRES_URI for domain state and the separate, least-privilege WORKER_DATABASE_URI for direct queue control. The latter is a Go runtime setting, not a replacement Python database alias. River schema is applied only by the one-shot migration path when MIGRATION_DATABASE_URI and the two runtime role names are supplied. See Workers for the coexistence boundary.

Test tiers

Canonical local test commands:

make test:unit
make test:integration
make test:e2e
make test:live-e2e
make test:ci

All tiers route through one entrypoint:

./ci/run_tests.sh <unit|integration|e2e|live-e2e|ci>

Notes:

• integration is token-aware and skips provider tests cleanly when credentials are unavailable.
• live-e2e starts a live backend harness, generates deterministic ClickHouse fixtures, waits for API readiness, and asserts /health, /api/v1/meta, and /api/v1/home.
• ci blocks on flake8 and coverage-gated unit tests. black, isort, and mypy are advisory by default; set STRICT_QUALITY_GATES=1 to make them blocking.
• JUnit XML paths are stable under test-results/junit/ and can be overridden with TEST_RESULTS_DIR / JUNIT_XML_* variables.

Container images

The repository builds two reusable images from docker/Dockerfile:

Image	Purpose
dev-hops-api	Runs dev-hops api on port 8000
dev-hops-runner	Uses dev-hops as the entrypoint for sync, fixtures, metrics, and maintenance jobs

Build both images:

IMAGE_REGISTRY=ghcr.io/myorg/dev-health-ops \
VERSION=$(git describe --tags --abbrev=0 2>/dev/null || echo latest) \
  ./scripts/build-images.sh

Run the API image:

docker run --rm -p 8000:8000 \
  -e POSTGRES_URI="postgresql+asyncpg://postgres:postgres@postgres:5432/postgres" \
  -e CLICKHOUSE_URI="clickhouse://ch:ch@clickhouse:8123/default" \
  dev-hops-api:latest

Run a CLI job through the runner image:

docker run --rm -it \
  --network dev-health_default \
  -v "$(pwd)":/app \
  -w /app \
  -e CLICKHOUSE_URI="clickhouse://ch:ch@clickhouse:8123/default" \
  dev-hops-runner:latest \
  metrics daily --backfill 14

Key docs

• docs/get-started/index.md: setup and demo data
• docs/reference/cli/index.md: full CLI reference
• docs/contribute/architecture/data-and-storage.md: PostgreSQL/ClickHouse split and provider → processor → sink boundaries
• docs/use/investment/index.md: canonical Investment View
• docs/use/reports/index.md: Report Center and scheduled reports

Guardrails

• WorkUnits are evidence containers, not categories.
• Investment categorization runs at compute time and persists distributions.
• Theme rollups are deterministic from canonical subcategories.
• UX-time LLM usage is explanation-only and must not recompute categories.
• Analytics persistence goes through ClickHouse sinks, not file exports or debug dumps.