-
Notifications
You must be signed in to change notification settings - Fork 19
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks great, thanks! Minor nits.
62ecc78
to
8d230e3
Compare
/hold |
/hold cancel |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Very nice :) I just have one suggestion.
8d230e3
to
09b3391
Compare
@timuthy and I reworked |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
/lgtm
09b3391
to
c1b5ac0
Compare
Rebased to resolve conflicts. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Had another rough sketch, but I'm anyways certain that you guys have done it very thoroughly.
/lgtm
What this PR does / why we need it:
This PR adds a cache to the client for the target cluster. That way,
grm
's health controller will not retrieve all objects of the manged resource from the API server (by default every minute), but rather use the cache instead. Additionally it only retrieves the resource, if it is of aKind
, that will actually be checked for its health.Also,
grm
now does not set.status.resources[].namespace
todefault
for cluster scoped resources (e.g.CustomResourceDefinitions
). The previous behaviour seemed to cause problems, when retrieving such resources from the cached client, so I changed it along the way.Which issue(s) this PR fixes:
Part of gardener/gardener#1953
I did some rudimentary before/after comparison via prometheus in my local setup with the following results:
Received bytes for
grm
ingarden
namespace down by~65%
(~15.5kB/s
->~5.5kB/s
)grm
in each shoot namespace down by~30%
(~16.9kB/s
->~11.8kB/s
) (should not really contribute much to external traffic, as most of the managed resources here target the Shoot's API server (in-cluster))Sent bytes for
grm
ingarden
namespace down by~41%
(~2.81kB/s
->~1.65kB/s
)grm
in each shoot namespace down by~14%
(~7.31kB/s
->~6.31kB/s
) (again, should not really contribute much to external traffic, as most of the managed resources here target the Shoot's API server (in-cluster))Special notes for your reviewer:
Release note: