CRI runtimes do not supply cpu nano core usage as it is not part of CRI
stats. However, there are upstream components that still rely on such
stats to function. The previous fix was faulty because the multiple
callers could compete and update the stats, causing
inconsistent/incoherent metrics. This change, instead, creates a
separate call for updating the usage, and rely on eviction manager,
which runs periodically, to trigger the updates. The caveat is that if
eviction manager is completley turned off, no one would compute the
usage.
assert.Contains() checks if its second argument (which is supposed to be
a single element) is contained in its first argument (which is supposed
to be a slice/map etc.) The third and following arguments are supposed
to be message and args for the output in case of failure.
Because of this bad form, a failure was hidden, the system container is
named "misc", not "system".
Automatic merge from submit-queue
Add memory available to summary stats provider
To support out of resource killing when low on memory, we want to let operators specify eviction thresholds based on available memory instead of memory usage for ease of use when working with heterogeneous nodes.
So for example, a valid eviction threshold would be the following:
* If node.memory.available < 200Mi for 30s, then evict pod(s)
For the node, `memory.availableBytes` is always known since the `memory.limit_in_bytes` is always known for root cgroup. For individual containers in pods, we only populate the `availableBytes` if the container was launched with a memory limit specified. When no memory limit is specified, the cgroupfs sets a value of 1 << 63 in the `memory.limit_in_bytes` so we look for a similar max value to handle unbounded limits, and ignore setting `memory.availableBytes`.
FYI @vishh @timstclair - as discussed on Slack.
/cc @kubernetes/sig-node @kubernetes/rh-cluster-infra
`--kubelet-cgroups` and `--system-cgroups` respectively.
Updated `--runtime-container` to `--runtime-cgroups`.
Cleaned up most of the kubelet code that consumes these flags to match
the flag name changes.
Signed-off-by: Vishnu kannan <vishnuk@google.com>
* Metrics will not be expose until they are hooked up to a handler
* Metrics are not cached and expose a dos vector, this must be fixed before release or the stats should not be exposed through an api endpoint
