Commit graph

3 commits

Author SHA1 Message Date
01e89fffca fix(storage): resolve persistence restart bug across all storage adapters
Critical bug fix that restores data persistence after application restart for all storage adapters (GCS, S3, OPFS, FileSystem).

**Root Cause:**
Storage adapters were loading entity counts from _system/counts.json but not returning totalCount in pagination methods, causing rebuildIndexesIfNeeded() to see undefined and incorrectly assume no data existed.

**Changes:**
- GCS Storage: Return totalCount in getNodesWithPagination() and getVerbsWithPagination()
- S3 Storage: Call initializeCounts() in init() and return totalCount in pagination methods
- OPFS Storage: Call initializeCounts() in init() to load pre-calculated counts
- BaseStorage: Add defensive validation to prevent future adapters from missing totalCount

**Impact:**
- Production deployments now work correctly (Cloud Run, AWS, Kubernetes)
- Data persists correctly across container restarts
- Serverless scale-to-zero deployments now functional
- All storage adapters use O(1) pre-calculated counts (scales to millions of entities)

**Testing:**
- Build passes with no TypeScript errors
- All existing tests pass
- Defensive validation ensures future storage adapters won't have this bug

Fixes critical production issue affecting all persistent storage deployments.
2025-10-09 15:07:18 -07:00
1966c39f24 fix: move metadata routing to base class, fix GCS/S3 system key crashes
Critical fix for GCS/S3 native storage adapters crashing on metadata index keys.

Problem:
- GCS and S3 adapters crashed with "Invalid UUID format" errors
- System keys like __metadata_field_index__status are NOT UUIDs
- Adapters incorrectly tried to shard all metadata keys as UUIDs

Solution:
- Move sharding/routing logic from adapters to BaseStorage class
- Add analyzeKey() method to detect system keys vs entity UUIDs
- System keys route to _system/ directory (no sharding)
- Entity UUIDs route to sharded directories (256 shards)
- All adapters now implement 4 primitive operations:
  * writeObjectToPath(path, data)
  * readObjectFromPath(path)
  * deleteObjectFromPath(path)
  * listObjectsUnderPath(prefix)

Benefits:
- Impossible for future adapters to repeat this mistake
- Zero breaking changes, full backward compatibility
- No data migration required
- Cleaner architecture with better separation of concerns

Updated adapters: GcsStorage, S3CompatibleStorage, OPFSStorage,
FileSystemStorage, MemoryStorage

Added: docs/architecture/data-storage-architecture.md
Updated: README.md with architecture docs link
2025-10-09 13:10:06 -07:00
e2aa8e3253 feat: add native Google Cloud Storage adapter with ADC support
Implement native @google-cloud/storage adapter for better performance
and easier authentication in Cloud Run/GCE environments.

Features:
- Application Default Credentials (ADC) for zero-config auth
- Service account authentication (keyFilename, credentials)
- HMAC fallback for backward compatibility
- Full UUID-based sharding preservation
- Write buffers for high-volume mode
- Multi-level caching and adaptive backpressure
- Complete feature parity with S3-compatible adapter

Benefits over S3-compatible GCS:
- No HMAC key management required
- Native SDK performance optimizations
- Automatic authentication in Cloud Run/GCE
- Simpler configuration

Configuration:
- type: 'gcs-native'
- gcsNativeStorage: { bucketName, keyFilename?, credentials? }
- Zero data migration required (same path structure)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
2025-10-08 14:08:43 -07:00