README.org
90 lines · 2438 bytes
hutch-stats
what
srht-contrib. python service. polls sourcehut activity, normalizes it into one event model, stores it in sqlite, and serves a contribution-calendar json api an ios app renders directly.
v1 scope:
- fastapi json api
- sqlite persistence
- polling ingestion
- full todo.sr.ht path
- git.sr.ht commit ingestion with repo auto-discovery
- public read-only endpoints, api-key auth for mutating routes
- alembic migrations
does
collects activity from sr.ht graphql services, aggregates by day, and returns zero-filled ranges so the client never patches missing dates. polls incrementally, backfills the last 365 days for new actors, prunes older activity.
services
implemented: todo.sr.ht, git.sr.ht. event types: ticket_created, ticket_comment, ticket_closed, commit.
env
API_KEY=replace-me
ENABLE_SCHEDULER=false
SRHT_TOKEN=replace-me
TODO_SRHT_ENDPOINT=https://todo.sr.ht/query
GIT_SRHT_ENDPOINT=https://git.sr.ht/query
DATABASE_URL=sqlite:///./srht_contrib.db
DEFAULT_ACTOR=~your-user
POLL_INTERVAL_SECONDS=300
more knobs (discovery batch size, repoll and backoff intervals, actor aliases, tracked repositories) live in config.py.
run
uv venv
source .venv/bin/activate
uv pip install -e ".[dev]"
cp .env.example .env
alembic upgrade head
uvicorn srht_contrib.main:app --reload
poll
manual poll:
curl -X POST \
"http://127.0.0.1:8000/api/contributions/poll?actor=~your-user" \
-H "X-API-Key: replace-me"
scheduled polling runs only when ENABLE_SCHEDULER=true. it drains due actors in batches, indexes fast, then backfills a year in bounded passes.
bulk queue without immediate indexing:
srht-enqueue-actors srht_usernames.txt --stagger-seconds 60
endpoints
calendar by year, calendar by date range, stats, and repository crud. full request and response shapes are in API.txt.
weights
commit 1.0, ticket_created 1.0, ticket_comment 0.5, ticket_closed 0.75, build_started 0.25, build_passed 0.25.
tests
pytest
limits
- git polling assumes repos are discoverable via graphql
- the scheduler runs in-process, not distributed
- new actors index asynchronously; the first read may be empty
- rolling one-year window; older activity is not kept
- alias management is config-driven, no crud api yet
- trusted-operator v1, not a public multi-tenant service