Signed-off-by: Taylor Silva <dev@taydev.net>

Signed-off-by: Taylor Silva <dev@taydev.net>

on systems where cgroupv1 is no longer enabled (Ubuntu 24.04, Fedora's
last few releases) the worker fails to start containers or fails to set
container limits on containers (if set). This happens because
/sys/fs/cgroup/cgroup.subtree_control is empty in the parent cgroup.

Ensuring the worker is in the hosts cgroup _usually_ ensures that it's
in a cgroup that has its cgroup.subtree_control populated. When
containerd makes cgroups from this cgroup it's then able to pass the
controllers in /sys/fs/cgroup/cgroup.controllers to its child cgroups.

If the cgroup that containerd starts in does not have these controllers
then this can result in containerd failing to create containers or
failing to set cpu/memory limits on containers. I've ran into both
errors on Fedora 39, 40, and Ubuntu 24.04.

Signed-off-by: Taylor Silva <dev@taydev.net>

when cgroupsv1 is disabled all the subfolders in /sys/fs/cgroup/ are
gone and we have to use the cgroupv2 interface files instead

Signed-off-by: Taylor Silva <dev@taydev.net>

Signed-off-by: Taylor Silva <dev@taydev.net>

Signed-off-by: Taylor Silva <dev@taydev.net>

some of our external repos into this one

Signed-off-by: Taylor Silva <dev@taydev.net>

on hosts running cgroup v2 only, which is all of the OSS workers, we
need to start in the parent cgroup so we have access to all the
controllers. Otherwise docker starts us in a cgroup with no controllers
and we can only use controllers that are our parent cgroup passed down
to us.

Signed-off-by: Taylor Silva <dev@taydev.net>

and make LKE the default engine. We won't be running tests on GKE or PKS

Signed-off-by: Taylor Silva <dev@taydev.net>

Signed-off-by: Taylor Silva <dev@taydev.net>

Signed-off-by: Taylor Silva <dev@taydev.net>

should keep waiting

Signed-off-by: Taylor Silva <dev@taydev.net>

ran:

go get -u ./...
go mod tidy

Signed-off-by: Taylor Silva <dev@taydev.net>

ran in the root and web/wats/

yarn upgrade --latest --caret

Signed-off-by: Taylor Silva <dev@taydev.net>

the library says it will just use the existing reader timeout value if
this timeout is set to zero

Signed-off-by: Taylor Silva <dev@taydev.net>

Signed-off-by: Taylor Silva <dev@taydev.net>

Signed-off-by: Taylor Silva <dev@taydev.net>

Signed-off-by: Taylor Silva <dev@taydev.net>

running on Broadcom's GCP workers now so we can run these tests again
(for now)
In the future we'll probably skip or remove them unless gdn starts
working with cgroups v2

Signed-off-by: Taylor Silva <dev@taydev.net>