CAPD load balancer crashes on startup

[killianmuldoon]

CAPD's load balancer is known to crash by causing a resource leak on some systems - notably Fedora. This issue is caused by the initial configuration of the haproxy container used by CAPD.

This issue has been partially fixed in #8246 which updates the config to one that will not crash when used with CAPD - i.e. setting maxconn. This change has also been made in Kind.

The load balancer will still cause a resource leak with its initial configuration, however. The fix for this is either to update to the latest kindest/haproxy once one an image with the correct config is published, or to an alternative haproxy image.

Note: The upstream haproxytect/haproxy-alpine has sufficient config to avoid the resource leakage (thanks @richardcase!)

See also: #7344 (comment)

/assign
/area provider/docker
/kind bug

[fabriziopandini]

/triage accepted

[sbueringer]

Given how long we lived with this issue, I would prefer waiting for the next kindest/haproxy image.

I"m not sure when it will be released, but we could ask.

[killianmuldoon]

Given how long we lived with this issue, I would prefer waiting for the next kindest/haproxy image.

Agreed - there's no rush on this. Just wanted to capture the info from the closed PRs and track the image bump somewhere.

[sbueringer]

Yup. Just saw the two options in the issue description, which is why I wrote which one I would prefer 😃

[richardcase]

@sbueringer @killianmuldoon - i'm curious why there is a preference to use the kindest image and not an upstream image?

[killianmuldoon]

From my PoV - CAPD is tightly coupled to kind, so there's no real separation from changing to an upstream image.

Otherwise this is a part of the code that's worked well for a long time without changes. I don't see any real upside to making the change to a new image, so I default to not changing it. But really I don't have very strong opinions here.

[richardcase]

Thats good to know, thanks for explaining @killianmuldoon 👍