Previously, the verious Merge() policies of the TopologyManager all
eturned their own lifecycle.PodAdmitResult result. However, for
consistency in any failed admits, this is better handled in the
top-level Topology manager, with each policy only returning a boolean
about whether or not they would like to admit the pod or not. This
commit changes the semantics to match this logic.
Previously, we unconditionally removed *all* topology hints from a pod
whenever just one container was being removed. This commit makes it so
we only remove the hints for the single container being removed, and
then conditionally remove the pod from the podTopologyHints[podUID] when
no containers left in it.
On bionic, we don't have eth0 hard coded. example below, so we use `ip
route` to figure out the default ethernet interface
```
dims@kubernetes-master:~$ ip link
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN mode DEFAULT group default qlen 1000
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
2: ens4: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1460 qdisc fq_codel state UP mode DEFAULT group default qlen 1000
link/ether 42:01:0a:80:00:23 brd ff:ff:ff:ff:ff:ff
3: docker0: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 qdisc noqueue state DOWN mode DEFAULT group default
link/ether 02:42:b2:4e:dd:86 brd ff:ff:ff:ff:ff:ff
```
Also, bionic uses systemd-resolver by default and adds entries in
/etc/resolv.conf that CoreDNS does not link. So follow the
recommendation in the documentation to specify resolv.conf explicitly
In addition to getting overall performance measurements from golang benchmark,
collect metrics that provides information about insides of the scheduler itself.
This is a first step towards improving what we collect about the scheduler.
Metrics in question:
- scheduler_scheduling_algorithm_predicate_evaluation_seconds
- scheduler_scheduling_algorithm_priority_evaluation_seconds
- scheduler_binding_duration_seconds
- scheduler_e2e_scheduling_duration_seconds
Scheduling throughput is computed on the fly inside perfScheduling.
In 24d105995d, a fix was made in bazel
based builds to ensure that we add `selinux` tag when we build all
binaries especially the `kubelet`. We need to do the same for in our
hack scripts so things like `make release` will work properly as well.
Some scripts use `GOFLAGS=-tags=providerless` for example, So we should
support the tags to be specified in GOFLAGS as well. We parse out the
tags from there and ensure selinux is added to the list of tags we used
for building the binaries. Note that we add our own `-tags` with the
full set of tags and since we specify our parameter at the end, ours
full list takes precendence
Organizes the behaviors directory based on SIG, and provides a few
example behavior descriptions for Pods and Services to start. Note
that these are very incomplete lists of behaviors at this point.
- Makes Windows Server 2019 the default version for Windows clusters on
GCP, since 1809 will be EOL in a few months.
- Adds Windows Server version 1909 as a Windows node choice.
- Use Windows images with updates from January 2020.
- Cleans up the code that sets the node image.
because metric initializations are too early. This causes actual metric
instance become no-op.
modification made in thie commit to make sure actual metric instance won't be no-op metrics:
- re-initialize scheduler/metrics.PodSchedule{Successes, Failure, Errors} after metric creation
- scheduler/metrics.Register() should be called before initializing SchedulingQueue,
