Feat[MQBSTAT]: Add queue utilization metrics #678
Conversation
Signed-off-by: Aleksandr Ivanov <[email protected]>
src/plugins/bmqprometheus/bmqprometheus_prometheusstatconsumer.cpp
Outdated
Show resolved
Hide resolved
| double avg = 0; | ||
| if (latestSnapshot > oldestSnapshot) { | ||
| avg = bmqst::StatUtil::average( | ||
| context.value(bmqst::StatContext::e_DIRECT_VALUE, currentValue), | ||
| latestSnapshot, | ||
| oldestSnapshot); | ||
| } | ||
| else { | ||
| avg = static_cast<double>(bmqst::StatUtil::value( | ||
| context.value(bmqst::StatContext::e_DIRECT_VALUE, currentValue), | ||
| latestSnapshot)); | ||
| } |
There was a problem hiding this comment.
What is the exact semantic of the calculated value? Is it "current utilization", "max utilization", etc?
There was a problem hiding this comment.
I consider that utilization is calculated as average value of current messages/bytes within publish period divided by limit value. Snapshot interval is usually 1 sec, publish interval is usually 10 sec. So, the average value of messages/bytes for the last 10 snapshots is calculated. Added comments into the code describing this logic.
There was a problem hiding this comment.
I think we should keep the semantics aligned with the existing history metrics, which are either current or max. Given the choice between the two, I think max is the more useful option. Domains already provide the ability to alarm (internally in the broker) if capacity crosses a threshold (e.g. 80% full), so exposing that max would be the the closest stat to the stat triggering that alarm.
There was a problem hiding this comment.
Agree, refactored to use max value for calculation of queue utilization.
Signed-off-by: Aleksandr Ivanov <[email protected]>
Signed-off-by: Aleksandr Ivanov <[email protected]>
| @@ -110,8 +110,10 @@ class QueueStatsDomain { | |||
| e_NB_CONSUMER, | |||
| e_MESSAGES_CURRENT, | |||
| e_MESSAGES_MAX, | |||
| e_MESSAGES_UTILIZATION, | |||
There was a problem hiding this comment.
What do you think about ending with "_MAX" to make it clear it's peak utilization? Otherwise, lgtm.
There was a problem hiding this comment.
Agree, added suffix _MAX to the new metrics.
Add two new queue metrics representing queue utilization (in precents):
queue.content_msgs / queue.cfg_msgs * 100%;queue.content_bytes/ queue.cfg_bytes * 100%;