I claim that the reproducer is producing correct results.

As I understand it, you are creating a queue with an upward threshold of 100 bytes and a downward threshold of 11 bytes.  You then produce 50 bytes, then drain all 50 bytes.

Since the queue size never crossed the upward threshold, it is correct that the downward threshold alert was _not_ generated.

This feature implements hysteresis in threshold alert generation.  It prevents floods of threshold alerts being sent.  Imagine if the queue was flowing steadily with 20 byte messages being produced and consumed continuously.  You certainly would not want a crossing-downward alert sent for every message.

This behavior seems to be correct, but it should be more detailed in the documentation. I will change this bug to documentation one.

Please add to documentation that: 

In above light current documentation in MICG is not complete:

Threshold crossed increasing
The increasing event is raised when the queue depth goes from (upper-threshold - 1) to upper-threshold.

Threshold crossed decreasing
    The decreasing event is raised when the queue depth goes from (lower-threshold + 1) to lower-threshold. 

Should be:
Threshold crossed increasing
    The increasing event is raised when the queue depth goes from (upper-threshold - 1) to upper-threshold and queue depth crosses lower threshold before.

Threshold crossed decreasing
    The decreasing event is raised when the queue depth goes from (lower-threshold + 1) to lower-threshold and queue depth crosses upper threshold before.

Note:
Both threshold should be set to receive alert events correctly.

http://deathstar1.usersys.redhat.com:3000/builds/18173-Messaging_Installation_and_Configuration_Guide/#Queue_Threshold_Alerts_Edge-triggered


New doc text:

Note: The upper and lower thresholds should have a gap between them to limit event rates, and both must be set to generate event notifications.

Events

There are two different events:

Threshold crossed increasing

The increasing event is raised when the queue depth goes from (upper-threshold - 1) to upper-threshold. This event is not retriggered by fluctuation around the upper-threshold, until a decreasing event is triggered. This prevents multiple continuous triggering of the increasing event.

Threshold crossed decreasing

The decreasing event is raised when the queue depth goes from (lower-threshold + 1) to lower-threshold. After a decreasing event, a further rise to the upper-threshold will trigger the increasing event.

The new text is good, but could you possibly fill in a note that the decreasing event is raised only if the increasing one was raised before it (similarly like the increasing event)?

current version:
The decreasing event is raised when the queue depth goes from (lower-threshold + 1) to lower-threshold. After a decreasing event, a further rise to the upper-threshold will trigger the increasing event.

Should be:
The decreasing event is raised when the queue depth goes from (lower-threshold + 1) to lower-threshold. The decreasing event is triggered only if the increasing event was raised before it. After a decreasing event, a further rise to the upper-threshold will trigger the increasing event.

I've introduced the idea of the "increasing event flag", which makes it easier to describe the behaviour. It's probably internally implemented using something like this, but as a conceptual entity it makes it clearer:

http://deathstar1.usersys.redhat.com:3000/builds/18173-Messaging_Installation_and_Configuration_Guide/#Queue_Threshold_Alerts_Edge-triggered

Threshold crossed increasing

The increasing event is raised when the queue depth goes from (upper-threshold - 1) to upper-threshold and the increasing event flag is not already set. When the an increasing event occurs the increasing event flag is set. The increasing event flag must be cleared (by a decreasing event) before further increasing events will be raised. This prevents multiple retriggering of this event by fluctuation of queue depth around the upper-threshold.

Threshold crossed decreasing

The decreasing event is raised when the increasing event flag is set and the queue depth goes from (lower-threshold + 1) to lower-threshold. The decreasing event clears the increasing event flag, allowing further increasing events to be triggered and preventing multiple retriggering of this event by fluctuation of queue depth around the lower-threshold.

Thanks for the additional information.

Updated: http://deathstar1.usersys.redhat.com:3000/builds/18173-Messaging_Installation_and_Configuration_Guide/#Queue_Threshold_Alerts_Edge-triggered

Documentation of queue threshold alerts is current and reflects the current state.

--> VERIFIED