Change default limit from 100 to None. When limit is omitted,
EDM returns all candles in the requested range without truncation.
Also increased max limit from 1000 to 10000 for large requests.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When requesting candles with start date but no end date (SINCE mode),
EDM was skipping gap processing entirely because gap_end was None.
This caused requests to return cached data clipped to the most recent
N candles, ignoring the requested start date.
Fix: Normalize gaps with end=None to use current time, so SINCE mode
requests properly fetch data from the requested start time to now.
Bug symptoms:
- Request start=2026-03-01, limit=100
- Got 100 candles starting from 2026-03-04 (most recent 100)
- Expected 100 candles starting from 2026-03-01
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add busy_timeout (5 seconds) to both async and sync database connections.
This prevents "database is locked" errors when multiple requests try to
access the database simultaneously.
Changes:
- AsyncSQLitePool: Add busy_timeout=5000 and synchronous=NORMAL pragmas
- DatabaseCache: Add WAL mode, busy_timeout, synchronous=NORMAL, and
connection timeout for the sync database module
The busy_timeout tells SQLite to wait instead of immediately failing
when the database is locked by another connection.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Update CLAUDE.md to reflect actual module structure (add websocket,
connection_pool, monitoring modules; remove outdated "Planned" note)
- Enable additional exchanges in config.yaml (kucoin, kraken, okx, etc.)
- Fix cache manager to refetch when memory has fewer candles than limit
- Fix benchmark tests asyncio event loop issues (use new_event_loop)
- Update unit tests to match new cache behavior and config
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Comprehensive tests that fetch actual data from Binance to verify:
- Three-tier caching (memory → database → exchange)
- Gap detection and filling (start, end, middle, multiple gaps)
- Data integrity verification against direct exchange fetch
- Freshness checking for stale data
- Different timeframe handling
- Edge cases (single candle, large ranges, overlapping requests)
Tests are marked with @real_exchange and can be skipped in CI with:
pytest -m "not real_exchange"
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add freshness check to _detect_gaps() in all cache layers (memory,
database, async_database). For limit-only requests (no start/end),
verify the most recent candle is within 2 intervals of current time.
If stale, flag as a gap so fresh data is fetched from exchange.
This fixes the issue where EDM served hours-old cached data instead
of fetching current candles when clients requested "latest N candles".
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- CacheManager.get_candles_with_source() supports per-request connector
overrides and reports data source (memory/database/exchange)
- AsyncDatabaseCache now receives pool_size/max_overflow from config
- CacheManager.close() properly shuts down async DB connection pool
- /candles endpoint accepts optional session_id for authenticated access
- /candles records metrics on success and errors (latency, cache source)
- Added tests for connector override, source reporting, and metrics
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
rob