Que n’aimez-vous pas à propos de YugabyteDB?
Current Challenges and Limitations
We list below the most critical issues and limitations currently impacting our YugabyteDB deployment for the Iris application:
1. DDL Atomicity and Concurrency
Concurrent DDL on different objects often fails or causes schema mismatch errors.
2. Truncate Behavior
Truncate operations retain old tablets, causing resource sprawl (CPU, disk).
3. Slow Aggregations / Analytical Queries
Aggregate functions (e.g., COUNT, SUM, GROUP BY) perform poorly on large tables.
4. Large Query Errors
Queries fail with RPC message size errors; workarounds require non-trivial gflag tuning.
5. Index Creation Challenges
Index creation on large tables is slow (can take hours) and unstable if DMLs are running.
Failure of concurrent DDLs can result in application downtime or stale views.
6. Intermittent Application Slowness
During high-ingestion windows (e.g., Spark + C# clients), CPU spikes to 80–85%.
7. Slow Queries Despite Indexing
Poor performance even with correctly designed indexes.
8. DR Limitations
DR requires symmetrical 3-node cluster and does not replicate DDL—this increases manual effort.
9. Node Crashes
Occasional crashes due to pg_client_use_shared_memory bug.
10. Resource Utilization
Max 1800 concurrent connections across 6 nodes (300/node).
High CPU usage (80%+) under 5500 OPS and 1500+ connections.
11. PITR Disk Usage
PITR with 2-day retention consumes 1–2 TB of disk.
Expected behavior, but storage overhead is significant.
12. Audit Logging
pgaudit causes crashes and lacks centralized log management.
Prefer audit logs to be stored as queryable tables.
13. Tablet Rebalancing
Rebalancing takes 2–3 hours post node failures.
14. Schema Name Change Not Reflected in UI
15. Query Performance Monitoring
No centralized query metrics dashboard across nodes.
pg_stat_statements is per-node; requires custom data aggregation.
16. Lack of ORM Support
Prisma ORM lacks native Yugabyte support.
Clear timeline for a smart driver integration is still needed.
17. Other Issues
Dead tuples causing transaction failures .
Clock skew-related tserver crashes .
Incorrect health checks leading to table drop incidents.
Backup to S3 failed due to endpoint misconfig .
Recommendations & Expectations
Top Priorities for Upcoming Releases:
Full concurrent DDL/DML support
Improved join and aggregation performance
Central query dashboard across universe
Audit log offloading and centralization
Smart tablet rebalancing and table-level recovery
Simplified backup/restore UX (especially for S3)
Documentation and Usability:
Better defaults for performance-related gflags.
Clear guidance on best practices for DDL coordination and high-throughput ingestion.
Support & Training:
More structured training on query optimization and resource tuning
Roadmap visibility for critical features (e.g., Prisma ORM support)
Final Thoughts
We appreciate Yugabyte’s continued partnership and responsiveness to issues. The platform shows strong promise for OLTP workloads and mission-critical deployments, but there are clear gaps—especially around operational tooling, analytical query support, and DDL concurrency—that we hope to see addressed in the near-term roadmap.
Our team remains committed to collaborating with Yugabyte to improve the product and looks forward to further performance and reliability enhancement Avis collecté par et hébergé sur G2.com.