Add support for podman metrics in docker module #41889
Conversation
botelastic
bot
added
the
needs_team
|
This pull request does not have a backport label.
To fixup this pull request, you need to add the backport labels for the needed
|
|
|
mergify
bot
added
the
backport-8.x
|
The relevant docker integration should also be updated with the new variable podman, right? |
bturquet
added
the
Team:Elastic-Agent-Data-Plane
|
Pinging @elastic/elastic-agent-data-plane (Team:Elastic-Agent-Data-Plane) |
botelastic
bot
removed
the
needs_team
|
@elastic/elastic-agent-data-plane could I get a review here as you are the codeowners? |
metricbeat/module/docker/docker.go
Outdated
| if !stream { | ||
| if err := decoder.Decode(&event.Stats); err != nil { | ||
| return event | ||
| } | ||
| } else { |
There was a problem hiding this comment.
Looks like in either case (error or not) we return the content of the event variable here. Does it even make sense to check for the error in this case?
There was a problem hiding this comment.
You are right, I was just receiving a linter error. I decided to add an debug log message.
MichaelKatsoulis
added
backport-8.16
* Add support for podman metrics (cherry picked from commit 1fefdbb)
* Add support for podman metrics (cherry picked from commit 1fefdbb)
* Add support for podman metrics (cherry picked from commit 1fefdbb)
* Add support for podman metrics (cherry picked from commit 1fefdbb) Co-authored-by: Michael Katsoulis <[email protected]>
* Add support for podman metrics (cherry picked from commit 1fefdbb) Co-authored-by: Michael Katsoulis <[email protected]>
#41967) * Add support for podman metrics in docker module (#41889) --------- Co-authored-by: Michael Katsoulis <[email protected]>
* Add support for podman metrics
Proposed commit message
Details
Podman offers a docker compatible API for metrics collection. So using docker module to collect podman metrics should be working out of the box.
In reality the memory metrics were not published at all and cpu usage percentage calculation was incorrect.
The reason is that the podman api returns zero values for
precpu_stats. These stats refer to the latest's read cpu statistics, needed for cpu percentage calculation. Also due to these stats being zero, the memory metrics were not populated because of a sanity check in the code.beats/metricbeat/module/docker/memory/helper.go
Line 55 in 3f51793
For docker, the
precpu_statsare returned every time.The solution in the Podman's case is to stream the api response. By default we had set the stream option to false as there was no need in case of docker.
As part of this PR we introduce a new configuration parameter named
podman(by default false). If set to true, indicating that podman is used, then the stream parameter is set to true, only for the collection of cpu and memory stats.The reason is that for cpu and memory stats the
precpu_statsare needed.From the streamed response, we get the second response as it was noticed during testing that the
precpu_statsof the first response were incorrect and the cpu was miscalculated.Checklist
CHANGELOG.next.asciidocorCHANGELOG-developer.next.asciidoc.Disruptive User Impact
There is no disruptive user impact.
How to test this PR locally
podman run -d -p 8080:80 --name mynginx nginxa.
podman exec -ti mynginx bashb.
while true; do :; done &podmanparameter set to truepodman statscommand results.Related issues
Use cases
Screenshots
Podman Stats command (check for mynginx container)
