conductor-oss/conductor · error · RuntimeException

Embeddings must be of dimensions

Error message

Embeddings must be of dimensions : <embeddingDimensions>

What it means

Thrown as a plain RuntimeException by PostgresVectorDB.upsertEmbeddings when the provided embeddings list size does not equal the configured embedding dimensions (default 256, from config.getDimensions()). The pgvector column has a fixed dimension, so a mismatched vector cannot be inserted and would fail at the DB anyway; this guard fails early with a clear message.

Solutions

1. Align config.getDimensions() (conductor.vectordb.postgres.dimensions) with the embedding model's output dimension.
2. Regenerate embeddings with the model whose dimension matches the index, or recreate the index/table at the new dimension.
3. Verify the embedding generation step uses the same model/dimensions setting as the index.
4. Ensure no other producer writes a different-dimension vector to the same namespace.

Example fix

// before
# config: conductor.vectordb.postgres.dimensions=256
# but model outputs 1536
// after
conductor.vectordb.postgres.dimensions=1536
# (and recreate the pgvector table/column at the new dimension)

// Match the vector dimension to config before upsert
int dim = config.getDimensions() != null ? config.getDimensions() : 256;
if (embeddings.size() != dim) {
    throw new IllegalArgumentException("expected " + dim + " dims, got " + embeddings.size());
}

When it happens

Trigger: Upserting embeddings whose length differs from config.getDimensions() (or 256 if unset). Commonly a model/dimension mismatch: e.g. a 1536-dim OpenAI embedding written into a 256-dim index, or a 384-dim MiniLM into a 1536-dim index.

Common situations: Switched embedding model without updating conductor.vectordb.postgres.dimensions; mixed models writing to the same index; dimensions config left at default 256 while using a model with a different output size; chunking code corrupting the vector length.

Related errors

• Invalid index name
• Invalid namespace
• Connection pool failed to start within
• Interrupted while waiting for connection pool
• Invalid index name

AI-assisted analysis of conductor-oss/conductor@cf7c3e4a8a (2026-08-14). Data as JSON: /api/errors/d64aa4ca18b43f70. Report an issue: GitHub.

Appendix: source

Thrown at ai/src/main/java/org/conductoross/conductor/ai/vectordb/postgres/PostgresVectorDB.java:192

                    "Error encountered while updating embeddings as : {}",
                    exception.getMessage(),
                    exception);
            throw new RuntimeException(exception);
        }
    }

    private int upsertEmbeddings(
            String namespace,
            String id,
            String parentDocId,
            String doc,
            List<Float> embeddings,
            Map<String, Object> metadata,
            Connection conn) {
        final int embeddingDimensions =
                config.getDimensions() != null ? config.getDimensions() : 256;
        if (embeddingDimensions != embeddings.size()) {
            throw new RuntimeException("Embeddings must be of dimensions : " + embeddingDimensions);
        }
        String tableName =
                config.getTablePrefix() != null
                        ? config.getTablePrefix() + "_" + namespace
                        : namespace;
        String UPSERT_QUERY =
                "INSERT INTO "
                        + tableName
                        + " AS n (id, parent_doc_id, embedding, doc, metadata) "
                        + "VALUES (?, ?, ?, ?, ?) "
                        + "ON CONFLICT (id) DO UPDATE SET parent_doc_id = ?, embedding = ?, doc = ?, metadata = ? WHERE n.id = ?";
        log.debug("Executing upsert query: {}", UPSERT_QUERY);
        log.debug(
                "Upserting document with id: {}, parentDocId: {}, doc length: {}, embedding dimensions: {}",
                id,
                parentDocId,
                doc.length(),
                embeddings.size());

View on GitHub (pinned to cf7c3e4a8a)