ActiveMQ高优先级消息阻塞：StoreQueueCursor异常排查求助

Alright, let's unpack this tricky ActiveMQ issue you're facing, starting with what StoreQueueCursor does and why it's causing your high-priority messages to get stuck. Your debug observations are super helpful here—they point straight to how the cursor is handling message sorting and delivery.

Think of StoreQueueCursor as the broker's "delivery manager" for queue messages. Its core jobs are:

• Pulling messages from the queue's backing store (in your case, in-memory since persistence is disabled)
• Maintaining a sorted view of messages when prioritized delivery is enabled
• Feeding messages to consumers in the correct order, while managing memory usage by loading messages in batches (hence the pendingCount metric you're tracking, which is the number of messages currently loaded and ready for delivery)

For priority queues, it’s responsible for ensuring high-priority messages are always at the front of the delivery line. With non-persistent queues, it uses an in-memory implementation (MemoryQueueCursor) to manage this sorted batch of pending messages.

From your debug data, here's what's happening under the hood:

1. When you send 444 priority-3 messages, the cursor loads all 444 into its pending pool (pendingCount=444), but for some reason, they're not being picked up by consumers.
2. Adding 72 priority-1 messages pushes pendingCount to 516, but the high-priority messages still stay stuck—meaning the cursor's sorted view got corrupted, and it's not prioritizing the higher-priority messages anymore.
3. Calling removeMatchingMessages resets the cursor entirely: it clears the pending pool (pendingCount=0) and forces a fresh load of messages from the queue's in-memory store. This fresh load triggers the sorting logic correctly, so the priority-3 messages are finally delivered.

This is almost certainly a bug in ActiveMQ 5.15.x's in-memory priority queue cursor. Specifically, when batches of mixed-priority messages are added to a queue with a large backlog, the cursor fails to re-sort its pending pool correctly. The reset triggered by removeMatchingMessages is the only thing that gets the sorting logic back on track.

Your attempt to disable caching (useCache=false) didn't help because that setting controls consumer-side caching, not the broker's cursor-level message caching. The prefetchPolicy.all=0 is a good call (it ensures consumers don't hoard messages), but it doesn't fix the cursor's sorting bug.

Given you're on ActiveMQ 5.15.12, here are actionable steps to resolve this:

1. Upgrade ActiveMQ (best long-term fix)
   This specific priority cursor bug was addressed in later versions (5.16.x and above). Upgrading will eliminate the root cause entirely if your stack allows it.

2. Tweak cursor memory limits
   Limit how many messages the cursor loads into its pending pool at once, which reduces the chance of sorting corruption. Add these settings to your PolicyEntry:

   policyEntry.setCursorMemoryHighWaterMark(1024 * 1024); // 1MB threshold
   policyEntry.setCursorMemoryLowWaterMark(512 * 1024); // 512KB threshold
   

   This forces the cursor to load messages in smaller, more manageable batches.

3. Disable optimized message storage
   Turn off the in-memory queue's storage optimization, which makes the cursor re-sort messages every time it prepares to deliver one. Add this to your PolicyEntry:

   policyEntry.setOptimizeMessageStorage(false);
   

   This adds a small performance overhead but ensures priority sorting works consistently.

4. Workaround: Use dedicated queues for high priority
   If upgrading isn't an option, split your traffic into two queues: one for high-priority messages, one for low. Configure your consumers to poll the high-priority queue first (using a message selector or separate consumer instances) to bypass the priority cursor bug entirely.

内容的提问来源于stack exchange，提问作者Michel Jung