Leverage one-shot container metrics in Docker metricsets

[jsoriano]

Describe the enhancement:

Docker metricsets get stats per container using the Docker ContainerStats API. Till Docker 20.10.0 (released on Dec 2020), this endpoint takes more than 1 second to reply, this is because it takes two measures to help on calculation of CPU usage percentages.

In 20.10.0, after moby/moby#40478 (see also moby/moby#28881), a new parameter is added, one-shot, that can be set to true to avoid this second measure, making these requests to reply much faster.

Take advantage of this new argument when possible.

This implies:

• Update Docker client to use a version including the ContainersStatOneShot method.
• Start using ContainerStatsOneShot instead of ContainerStats. In principle this change can be directly done on any metricset that is not using PreCPUStats.
• In the cpu metricset, refactor the code to keep supporting current metrics based on PreCPUStats.
• Keep supporting older versions of Docker.

Describe a specific use case for the enhancement or feature:

This would improve the use of resources when monitoring Docker daemons with many running containers.

[elasticmachine]

Pinging @elastic/integrations (Team:Integrations)

[elasticmachine]

Pinging @elastic/integrations (Team:Platforms)

[botelastic]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.