Agent should collect and report CPU and memory usage of monitoring components

[ycombinator]

Currently, Agent is only collecting and, therefore, reporting CPU and memory usage for itself and it's non-monitoring Beats components. It is not collecting and reporting CPU and memory usage for the monitoring Beats, leading to undercounting.

So, for a default Fleet policy with the system integration and monitoring (logs and metrics both) enabled, Agent runs:

• 1 metricbeat instance for the system integration: system/metrics-default,
• 1 filebeat instance for the system integration: log-default,
• 2 metricbeat instances for metrics monitoring: http/metrics-monitoring and beat/metrics-monitoring, and
• 1 filebeat instance for logs monitoring: filestream-monitoring.

In the above scenario, Agent is currently only collecting and reporting CPU and memory usage for the metricbeat and filebeat instances for the system integration: system/metrics-default and log-default. It is not collecting and reporting CPU and memory usage for the metricbeat and filebeat instances for monitoring: http/metrics-monitoring, beat/metrics-monitoring, and filestream-monitoring.

[elasticmachine]

Pinging @elastic/elastic-agent (Team:Elastic-Agent)

[fearful-symmetry]

@ycombinator when you say "collect and report" what are you imagining? We make sure metrics for these end up in the last 30s log lines? Something else?

[ycombinator]

@ycombinator when you say "collect and report" what are you imagining? We make sure metrics for these end up in the last 30s log lines? Something else?

I meant collect and report monitoring metrics of the monitoring beats' resource usage the same way we do today for any non-monitoring beats. The issue description has an example that should help clarify things further, but let me know if it doesn't.

[fearful-symmetry]

@ycombinator I think we're on different pages here. So, the last 30s metrics for the monitoring beats are exposed via the agent logs:

grep -rh "last 30s" . | jq .component.id
"filestream-monitoring"
"system/metrics-default"
"beat/metrics-monitoring"
"http/metrics-monitoring"
"filestream-monitoring"
"system/metrics-default"
"beat/metrics-monitoring"
"http/metrics-monitoring"
"filestream-monitoring"
"system/metrics-default"
"beat/metrics-monitoring"
"http/metrics-monitoring"
"filestream-monitoring"
"system/metrics-default"
"beat/metrics-monitoring"
"http/metrics-monitoring"
"filestream-monitoring"

What specific monitoring interface or UX are you thinking of here? The Kibana/fleet agent resource usage GUI? Something else?

[ycombinator]

What specific monitoring interface or UX are you thinking of here? The Kibana/fleet agent resource usage GUI? Something else?

Currently, the circled metrics do not include resource usage for the monitoring beats themselves. This results in undercounting the resources being used by Agent and all it's component processes. See #4005 (comment) for background discussion that led to this (and a few other) issues.

[cmacknz]

Re-opening, this was reverted because the test was failing #4451

[pierrehilbert]

@fearful-symmetry I think we can close this one now that #4462 is merged right?