How RocketMQ 4.8.0 Supercharges DLedger with Massive Performance and Stability Gains

Apache RocketMQ 4.8.0 has been released, bringing a large set of optimizations and bug fixes to the DLedger mode, which is the Raft‑based commit‑log implementation introduced in version 4.5.0 to provide automatic failover for broker groups.

Performance upgrades

Asynchronous pipeline mode – The previous synchronous processing required the SendMessageProcessor thread to wait for majority replication before responding. The new version uses CompletableFuture to decouple acknowledgment from the processing thread, allowing the next request to be handled immediately while a separate thread finalizes the ack, dramatically increasing throughput.

Batch log replication – Setting isEnableBatchPush=true aggregates multiple entries into a single packet, reducing system‑call overhead and context switches. This improves throughput under high load without adding latency to individual messages.

Additional lock and cache optimizations further multiply DLedger performance.

Stability upgrades

To validate DLedger reliability, the community applied the OpenMessaging‑Chaos framework, which injects faults such as random network partitions, packet loss, process kills, and slow nodes. The tests run a 5‑node DLedger cluster with four concurrent clients, alternating 100 s of normal operation and 100 s of fault injection for a total of 2400 s.

random‑partition – simulates symmetric network partitions.

random‑loss – drops a configurable proportion of packets.

random‑kill (minor/major/fixed) – simulates process crashes.

random‑suspend (minor/major/fixed) – simulates slow nodes (e.g., GC pauses, OOM).

bridge and partition‑majorities‑ring – simulate asymmetric partitions.

The latency chart shows brief periods of unavailability during fault injection, but the cluster recovers within 30 s, typically around 22 s on average. No message loss was observed among the 110,000 messages sent, confirming the At‑Least‑Once delivery guarantee.

These tests uncovered several hidden issues, which have been fixed, further improving DLedger's tolerance to process crashes, slow nodes, symmetric/asymmetric partitions, and packet loss.

Feature upgrades

Preferred Leader – The new preferredLeaderId configuration allows explicit leader selection, enabling better data‑center alignment and mixed‑deployment strategies.

Batch message support – Starting with 4.8.0, DLedger can send batch messages, making its capabilities fully comparable to the traditional Master‑Slave deployment.

The release also includes 25 improvements, 11 bug fixes, and 16 documentation/code‑format enhancements contributed by nearly 40 community members.