Chroma

Chroma insights

Open SourceVersions: [current]6 metrics
Chroma
Backup Strategy Absence Leading to Unrecoverable Data Losscritical

Chroma deployments without regular backups are vulnerable to unrecoverable data loss from corruption, accidental deletion, disk failure, or operational errors. Embedded mode stores all data in local persist directory — if directory is lost, data is lost. Client-server mode reduces risk but still requires backup strategy. Production systems must implement automated backups with tested restore procedures.

sqlite.org
28d ago▸
Chroma
Authentication Bypass in Default Deploymentcritical

Chroma server deployments without explicit authentication configuration are accessible without credentials. Default configuration does not require authentication, allowing any network-accessible client to read, write, or delete data. This is a critical security risk in production environments, multi-tenant systems, or any deployment exposed beyond localhost.

28d ago▸
Chroma
Collection Deletion Not Freeing Disk Spaceinfo

Deleting a collection in Chroma removes metadata and marks vectors as deleted but may not immediately reclaim disk space, especially if vacuum has not been run. SQLite DELETE operations mark space as free but don't shrink the database file. HNSW indices may leave orphaned files. Disk usage remains high after collection deletion until explicit cleanup (vacuum, manual file deletion).

sqlite.org
28d ago▸
Chroma
Query Timeout from Inefficient Distance Calculationswarning

Complex distance functions (L2, cosine similarity) on high-dimensional vectors are computationally expensive. Large n_results with high ef_search parameters force HNSW to compute distances for thousands of candidates, causing query timeouts (>30s). This is worse without SIMD optimizations, on under-provisioned CPU, or when querying multiple collections sequentially rather than in parallel.

28d ago▸
Chroma
Concurrent Write Contention on SQLite Metadata Storewarning

Chroma uses SQLite for metadata storage. SQLite has limited concurrent write capability due to database-level write locks. High concurrent upsert, update, or delete operations cause lock contention, resulting in 'database is locked' errors, timeout failures, and degraded write throughput. Read operations are not affected but may be delayed by long-running write transactions.

28d ago▸
Chroma
Vector Index Memory Exhaustion on Large Collectionscritical

HNSW indices load entirely into memory for query performance. Collections with millions of high-dimensional vectors can exhaust available RAM, causing OOM errors, system instability, or swapping that degrades query performance by orders of magnitude. Memory requirements scale with collection size, dimensionality, and HNSW parameters (M, ef_construction).

28d ago▸
Chroma
Stale Query Results from Eventual Consistencyinfo

Chroma may exhibit eventual consistency behavior where recently added/updated vectors are not immediately reflected in query results. This occurs due to buffering, async persistence, or HNSW index update delays. Applications expecting read-your-writes consistency may observe missing or outdated results immediately after write operations.

28d ago▸
Chroma
Collection Metadata Corruption After Unclean Shutdowncritical

Chroma's SQLite metadata store is vulnerable to corruption if the process terminates abnormally (crash, SIGKILL, power loss) during a write transaction. Corruption manifests as collection initialization failures, missing collections, inconsistent vector counts, or foreign key constraint violations. This is more likely in embedded deployments without WAL mode enabled.

28d ago▸
Chroma
Embedding Provider Rate Limiting and Quota Exhaustionwarning

Chroma relies on embedding functions (OpenAI, Cohere, HuggingFace, custom) to generate vectors. Rate limits, quota exhaustion, or API failures from embedding providers cause upsert/query operations to fail or experience severe latency. This is especially problematic during bulk ingestion or high query concurrency.

28d ago▸
Chroma
Query Performance Degradation on High-Cardinality Metadata Filterswarning

Chroma supports WHERE clause filtering on metadata during queries. High-cardinality metadata fields (e.g., document_id, user_id with millions of unique values) without proper indexing cause full table scans in SQLite, dramatically increasing query latency. This is especially problematic when combining vector similarity search with metadata filters.

28d ago▸
Chroma
Chroma Collection Initialization Failure in Dockerwarning

Chroma deployed in Docker Compose may experience inconsistent collection initialization, where CreateCollection succeeds but subsequent Get/Query operations fail with 400 errors. This often stems from network configuration mismatches, DNS resolution issues, or API endpoint inconsistencies between client and server configurations in containerized environments.

stackoverflow.comdocs.docker.com
1mo ago▸
Chroma
Database Size Bloat Without Vacuumingwarning

Chroma's SQLite database accumulates deleted/updated record space without automatic cleanup. Before v0.5.6, no automatic pruning existed. Post v0.5.6, automatic pruning is enabled after manual vacuum, but legacy databases and high-churn workloads still require periodic manual intervention to reclaim disk space.

docs.trychroma.comcookbook.chromadb.devsqlite.org
1y ago▸
Chroma
Cross-Architecture HNSW Performance Degradationinfo

Default Chroma HNSW builds lack CPU-specific optimizations (SSE, AVX, AVX2, AVX-512 SIMD instructions), resulting in suboptimal vector distance calculations. Rebuilding HNSW for the target architecture enables hardware-accelerated vector operations, improving query throughput by 2-5x and reducing latency by 40-60%.

cookbook.chromadb.devgithub.com
1y ago▸
Chroma
HNSW Index Fragmentation from Updateswarning

HNSW indices are optimized for insert-heavy workloads. Frequent update operations (excluding pure adds) fragment the graph structure over time, creating suboptimal routing paths, increasing query latency, reducing recall accuracy, and bloating disk usage. Updates don't modify the graph in place but mark old vectors deleted and add new versions, leaving dead graph nodes.

cookbook.chromadb.devarxiv.org
1y ago▸
Chroma
Embedding Dimensionality Storage Overheadinfo

High-dimensional embeddings (e.g., OpenAI text-embedding-3-large with 3072 dimensions) significantly increase storage, memory, and query costs compared to lower-dimensional alternatives. Many embedding models support dimensionality reduction without major accuracy loss. Reducing from 3072 to 1536 or 768 dimensions can halve storage and memory requirements while maintaining 95-98% retrieval accuracy.

cookbook.chromadb.dev
1y ago▸
Chroma
HNSW Index Insufficient Results Errorwarning

When the HNSW index cannot return the requested number of results, queries fail with a 2D array error. This happens when M, ef_construction, or ef_search parameters are too small for the dataset structure, limiting the index's ability to traverse and find sufficient neighbors.

docs.trychroma.comgithub.com
26y ago▸
Chroma
SQLite Version Incompatibilitycritical

Chroma requires SQLite ≥3.35 for critical features like ALTER TABLE support and improved JSON handling. Older Python versions (< 3.10) or Linux distributions (Ubuntu 20.04, Debian buster) bundle incompatible SQLite versions, causing initialization failures, schema migration errors, and runtime crashes.

docs.trychroma.comsqlite.org
26y ago▸