jd-opensource/joyagent-jdgenie · error · RuntimeException

批量保存向量数据失败

Error message

批量保存向量数据失败

What it means

In syncVectorInfo, schema vectors are saved in batches via vectorService.saveVector(req); any exception in a batch is logged and rethrown as a plain RuntimeException '批量保存向量数据失败' (failed to batch-save vector data), aborting the whole synchronization and losing the error's original type/stack in the rethrown exception.

Solutions

1. Check the log line 批量保存向量数据失败 for the original cause (e.getMessage() is logged with stack trace) and fix the underlying vector service issue
2. Verify the vector store is reachable and the SCHEMA_COLLECTION_NAME collection exists with the expected dimension
3. Reduce the batch size if timeouts or payload limits are involved
4. Rethrow preserving the cause (new RuntimeException(msg, e)) so the real failure isn't hidden

Example fix

// before
} catch (Exception e) {
    log.error("批量保存向量数据失败{}", e.getMessage(), e);
    throw new RuntimeException("批量保存向量数据失败");
}
// after
} catch (Exception e) {
    log.error("批量保存向量数据失败{}", e.getMessage(), e);
    throw new RuntimeException("批量保存向量数据失败: " + e.getMessage(), e);
}

// pre-check before syncVectorInfo
if (!vectorService.collectionExists(DataAgentConstants.SCHEMA_COLLECTION_NAME)) {
    vectorService.createCollection(DataAgentConstants.SCHEMA_COLLECTION_NAME, expectedDimension);
}
if (batch == null || batch.isEmpty()) {
    throw new IllegalStateException("No vector data to save");
}

try {
    long total = chatModelInfoService.syncVectorInfo(...);
} catch (RuntimeException e) {
    if ("批量保存向量数据失败".equals(e.getMessage())) {
        log.error("Vector sync failed; check vector store availability and collection dimension", e);
        // retry with smaller batch or back off
    }
}

When it happens

Trigger: vectorService.saveVector throwing for a batch — vector store unavailable/refusing connection, collection DataAgentConstants.SCHEMA_COLLECTION_NAME missing or dimension mismatch between embeddings and collection schema, embedding service failure producing bad vectors, or an oversized batch hitting payload limits.

Common situations: Vector database (e.g. Milvus/ES) is down or misconfigured in the environment; collection was recreated with a different embedding dimension; network timeouts on large batches; first-time sync where the collection hasn't been created yet.

Related errors

• collectionName is empty
• vector is empty
• 集合名为空！
• 向量集合为空！
• 元数据集合为空！

AI-assisted analysis of jd-opensource/joyagent-jdgenie@2417e0b8b6 (2026-09-08). Data as JSON: /api/errors/ab75b6f0c926ed05. Report an issue: GitHub.

Appendix: source

Thrown at genie-backend/src/main/java/com/jd/genie/service/ChatModelInfoService.java:167

    }

    private int syncVectorInfo(List<ChatModelSchema> chatModelSchemas) {
        List<VectorSaveReq.VectorData> vectorDataList = convertToVectorData(chatModelSchemas);
        // 分批处理
        int batchSize = 20;
        int total = 0;
        for (int i = 0; i < vectorDataList.size(); i += batchSize) {
            int endIndex = Math.min(i + batchSize, vectorDataList.size());
            List<VectorSaveReq.VectorData> batch = vectorDataList.subList(i, endIndex);
            try {
                VectorSaveReq req = new VectorSaveReq();
                req.setCollectionName(DataAgentConstants.SCHEMA_COLLECTION_NAME);
                req.setDataList(batch);
                vectorService.saveVector(req);
                total += endIndex - i;
            } catch (Exception e) {
                log.error("批量保存向量数据失败{}", e.getMessage(), e);
                throw new RuntimeException("批量保存向量数据失败");
            }
        }
        return total;
    }

    private List<VectorSaveReq.VectorData> convertToVectorData(List<ChatModelSchema> schemaList) {
        List<VectorSaveReq.VectorData> allVectors = new ArrayList<>();
        for (ChatModelSchema schema : schemaList) {
            String[] uuids = schema.getVectorUuid().split(",");
            addVectorSaveData(allVectors, schema, schema.getColumnName(), uuids[0]);
            addVectorSaveData(allVectors, schema, schema.getSynonyms(), uuids[1]);
            addVectorSaveData(allVectors, schema, schema.getColumnComment(), uuids[2]);
            if (!StandardColumnType.DECIMAL.name().equalsIgnoreCase(schema.getDataType())) {
                //数值类型fewShot不参与向量化
                addVectorSaveData(allVectors, schema, schema.getFewShot(), uuids[3]);
            }
        }
        return allVectors;

View on GitHub (pinned to 2417e0b8b6)