Storm 2.6.0.2 -
Apache Storm is a distributed, fault-tolerant, open-source computation system. It is designed for processing streaming data in real-time, capable of handling massive amounts of data with low latency.
- Bug fixes: Resolved crashes related to task scheduling under high concurrency and fixed memory leak in the tuple serializer.
- Performance: Improved throughput for worker heartbeats and reduced latency in tuple anchoring paths.
- Reliability: Hardened recovery logic for failed supervisors to prevent task duplication on restart.
- Compatibility: Backwards-compatible with 2.6.x topologies; no topology changes required.
- Upgrade notes: Rolling upgrade supported — upgrade supervisors/workers one-at-a-time; monitor logs for WARN-level recovery messages during the window.
1. Critical Bug Fixes (Stability)
- STORM-3908: Nimbus Thrift Serialization Failure. In previous 2.6.x versions, heavy topology loads caused Nimbus (the master node) to fail serializing topology metadata, leading to leader election storms. This is resolved.
- STORM-3912: Worker Heartbeat Timeout under GC Pressure. Garbage Collection pauses on the JVM would sometimes exceed the heartbeat timeout, causing supervisors to kill healthy workers. The heartbeat logic now includes a grace period based on GC logs.
- STORM-3915: Windowed Bolt State Corruption. When using the new
WindowedBolt with checkpointing, state snapshots occasionally failed to flush to the state store. This patch repairs the checkpointing sequence.
For teams currently running on older 1.x or early 2.x versions, 2.6.0.2 offers several compelling advantages: storm 2.6.0.2
STORM-3945: Worker Crash During Topology Rebalance
A race condition in the backpressure.manager could cause a worker process to throw an unhandled NullPointerException when a topology rebalance occurred simultaneously with a partial network partition. This fix backports a thread-safe state machine for backpressure signals. Bug fixes: Resolved crashes related to task scheduling
