Metrics and Dashboards

Three panels display inflight, buffered, and stashed messages. These panels are key tools for monitoring integration traffic and can help identify potential flow misconfigurations.

Connect persists all incoming messages. A limited number of messages are processed concurrently, while additional messages are buffered in memory. When the buffer reaches its limit, persisted messages are no longer held in memory. These are referred to as stashed messages. Stashed messages must be read back from persistence once buffer space becomes available before they can be processed.

A consistently high inflight message count is not always a problem, but it may indicate inefficient configuration. For large messages in particular, high inflight counts can consume significant memory and should be monitored closely.

If the buffer remains full for extended periods, it could indicate a configuration issue. Buffered messages are kept in memory to handle short bursts of load without reading from persistence. For extended periods of high loads, the buffer can fill up, causing messages to move to the stash. In this situation, the buffer continues to consume memory, but you will have to refill the buffer periodically from stash. This effectively removes the performance gain you get by keeping messages in memory.

If you observe consistently high numbers of both buffered and stashed messages, review your buffer configuration. In some cases, reducing the buffer size may improve overall performance.

The panels displaying processing and persistence times are useful for identifying performance issues. When Connect is under heavy load, the persistence provider can sometimes become the main bottleneck. Monitoring average persistence time can help determine whether this is the case.

If processing time per message is high, it may indicate problematic processors, for example, a suboptimal map implementation.

This section provides basic JVM metrics along with side-channel metrics. You can use these metrics to identify unhealthy memory usage patterns, and gain insight into the messages that are side-channeled, as well as the performance of those side-channels.

Mailbox metrics are primarily intended for debugging flow-server internals and are mainly used when troubleshooting with the Connect engineering team.