* Build Kubelet container images internally and publish
to Quay and Dockerhub (new) as an alternative in case of
registry outage or breach
* Use our infra to provide single and multi-arch (default)
Kublet images for possible future use
* Docs: Show how to use alternative Kubelet images via
snippets and a systemd dropin (builds on #737)
Changes:
* Update docs with changes to Kubelet image building
* If you prefer to trust images built by Quay/Dockerhub,
automated image builds are still available with unique
tags (albeit with some limitations):
* Quay automated builds are tagged `build-{short_sha}`
(limit: only amd64)
* Dockerhub automated builts are tagged `build-{tag}`
and `build-master` (limit: only amd64, no shas)
Links:
* Kubelet: https://github.com/poseidon/kubelet
* Docs: https://typhoon.psdn.io/topics/security/#container-images
* Registries:
* quay.io/poseidon/kubelet
* docker.io/psdn/kubelet
* Write the systemd kubelet.service to use `KUBELET_IMAGE`
as the Kubelet. This provides a nice way to use systemd
dropins to temporarily override the image (e.g. during a
registry outage)
Note: Only Typhoon Kubelet images and registries are supported.
* With Fedora CoreOS image stream support (#727), the latest
resolved image will change over the lifecycle of a cluster.
* Fix issue where an image diff proposed replacing a Fedora
CoreOS controller on GCP, introduced in #727 (unreleased)
* Also ignore image diffs to the GCP managed instance group
of workers. This aligns with worker AMI diffs being ignored
on AWS and similar on Azure, since workers update themselves.
Background:
* Controller nodes should strictly not be recreated by Terraform,
they are stateful (etcd) and should not be replaced
* Across cloud platforms, OS image diffs are ignored since both
Flatcar Linux and Fedora CoreOS nodes update themselves. For
workers, user-data or disk size diffs (where relevant) are allowed
to recreate workers templates/configs since these are considered
to be user-initiated declarations that a reprovision should be done
* Set a consistent MCS level/range for Calico install-cni
* Note: Rebooting a node was a workaround, because Kubelet
relabels /etc/kubernetes(/cni/net.d)
Background:
* On SELinux enforcing systems, the Calico CNI install-cni
container ran with default SELinux context and a random MCS
pair. install-cni places CNI configs by first creating a
temporary file and then moving them into place, which means
the file MCS categories depend on the containers SELinux
context.
* calico-node Pod restarts creates a new install-cni container
with a different MCS pair that cannot access the earlier
written file (it places configs every time), causing the
init container to error and calico-node to crash loop
* https://github.com/projectcalico/cni-plugin/issues/874
```
mv: inter-device move failed: '/calico.conf.tmp' to
'/host/etc/cni/net.d/10-calico.conflist'; unable to remove target:
Permission denied
Failed to mv files. This may be caused by selinux configuration on
the
host, or something else.
```
Note, this isn't a host SELinux configuration issue.
Related:
* https://github.com/poseidon/terraform-render-bootstrap/pull/186
* Enable bootstrap token authentication on kube-apiserver
* Generate the bootstrap.kubernetes.io/token Secret that
may be used as a bootstrap token
* Generate a bootstrap kubeconfig (with a bootstrap token)
to be securely distributed to nodes. Each Kubelet will use
the bootstrap kubeconfig to authenticate to kube-apiserver
as `system:bootstrappers` and send a node-unique CSR for
kube-controller-manager to automatically approve to issue
a Kubelet certificate and kubeconfig (expires in 72 hours)
* Add ClusterRoleBinding for bootstrap token subjects
(`system:bootstrappers`) to have the `system:node-bootstrapper`
ClusterRole
* Add ClusterRoleBinding for bootstrap token subjects
(`system:bootstrappers`) to have the csr nodeclient ClusterRole
* Add ClusterRoleBinding for bootstrap token subjects
(`system:bootstrappers`) to have the csr selfnodeclient ClusterRole
* Enable NodeRestriction admission controller to limit the
scope of Node or Pod objects a Kubelet can modify to those of
the node itself
* Ability for a Kubelet to delete its Node object is retained
as preemptible nodes or those in auto-scaling instance groups
need to be able to remove themselves on shutdown. This need
continues to have precedence over any risk of a node deleting
itself maliciously
Security notes:
1. Issued Kubelet certificates authenticate as user `system:node:NAME`
and group `system:nodes` and are limited in their authorization
to perform API operations by Node authorization and NodeRestriction
admission. Previously, a Kubelet's authorization was broader. This
is the primary security motivation.
2. The bootstrap kubeconfig credential has the same sensitivity
as the previous generated TLS client-certificate kubeconfig.
It must be distributed securely to nodes. Its compromise still
allows an attacker to obtain a Kubelet kubeconfig
3. Bootstrapping Kubelet kubeconfig's with a limited lifetime offers
a slight security improvement.
* An attacker who obtains the kubeconfig can likely obtain the
bootstrap kubeconfig as well, to obtain the ability to renew
their access
* A compromised bootstrap kubeconfig could plausibly be handled
by replacing the bootstrap token Secret, distributing the token
to new nodes, and expiration. Whereas a compromised TLS-client
certificate kubeconfig can't be revoked (no CRL). However,
replacing a bootstrap token can be impractical in real cluster
environments, so the limited lifetime is mostly a theoretical
benefit.
* Cluster CSR objects are visible via kubectl which is nice
4. Bootstrapping node-unique Kubelet kubeconfigs means Kubelet
clients have more identity information, which can improve the
utility of audits and future features
Rel: https://kubernetes.io/docs/reference/command-line-tools-reference/kubelet-tls-bootstrapping/
Rel: https://github.com/poseidon/terraform-render-bootstrap/pull/185
* Race: During initial bootstrap, static control plane pods
could hang with Permission denied to bootstrap secrets. A
manual fix involved restarting Kubelet, which relabeled mounts
The race had no effect on subsequent reboots.
* bootstrap.service runs podman with a private unshared mount
of /etc/kubernetes/bootstrap-secrets which uses an SELinux MCS
label with a category pair. However, bootstrap-secrets should
be shared as its mounted by Docker pods kube-apiserver,
kube-scheduler, and kube-controller-manager. Restarting Kubelet
was a manual fix because Kubelet relabels all /etc/kubernetes
* Fix bootstrap Pod to use the shared volume label, which leaves
bootstrap-secrets files with SELinux level s0 without MCS
* Also allow failed bootstrap.service to be re-applied. This was
missing on bare-metal and AWS
* In v1.18.0, kubectl apply would fail to apply manifests if any
single manifest was unable to validate. For example, if a CRD and
CR were defined in the same directory, apply would fail since the
CR would be invalid as the CRD wouldn't exist
* Typhoon temporary workaround was to separate CNI CRD manifests
and explicitly apply them first. No longer needed in v1.18.1+
* Kubernetes v1.18.1 restored the prior behavior where kubectl apply
applies as many valid manifests as it can. In the example above, the
CRD would be applied and the CR could be applied if the kubectl apply
was re-run (allowing for apply loops).
* Upstream fix: https://github.com/kubernetes/kubernetes/pull/89864
* No change to Fedora CoreOS modules
* For Container Linx AWS and Azure, change the `os_image` default
from coreos-stable to flatcar-stable
* For Container Linux GCP and DigitalOcean, change `os_image` to
be required since users should upload a Flatcar Linux image and
set the variable
* For Container Linux bare-metal, recommend users change the
`os_channel` to Flatcar Linux. No actual module change.
* Fix bootstrap error for missing `manifests-networking/crd*yaml`
when `networking = "flannel"`
* Cleanup manifest-networking directory left during bootstrap
* Regressed in v1.18.0 changes for Calico https://github.com/poseidon/typhoon/pull/675
* Change kube-proxy, flannel, and calico-node DaemonSet
tolerations to tolerate `node.kubernetes.io/not-ready`
and `node-role.kubernetes.io/master` (i.e. controllers)
explicitly, rather than tolerating all taints
* kube-system DaemonSets will no longer tolerate custom
node taints by default. Instead, custom node taints must
be enumerated to opt-in to scheduling/executing the
kube-system DaemonSets
* Consider setting the daemonset_tolerations variable
of terraform-render-bootstrap at a later date
Background: Tolerating all taints ruled out use-cases
where certain nodes might legitimately need to keep
kube-proxy or CNI networking disabled
Related: https://github.com/poseidon/terraform-render-bootstrap/pull/179
* Problem: Fedora CoreOS images are manually uploaded to GCP. When a
cluster is created with a stale image, Zincati immediately checks
for the latest stable image, fetches, and reboots. In practice,
this can unfortunately occur exactly during the initial cluster
bootstrap phase.
* Recommended: Upload the latest Fedora CoreOS image regularly
* Mitigation: Allow a failed bootstrap.service run (which won't touch
the done ConditionalPathExists) to be re-run by running `terraforma apply`
again. Add a known issue to CHANGES
* Update docs to show the current Fedora CoreOS stable version to
reduce likelihood users see this issue
Longer term ideas:
* Ideal: Fedora CoreOS publishes a stable channel. Instances will always
boot with the latest image in a channel. The problem disappears since
it works the same way AWS does
* Timer: Consider some timer-based approach to have zincati delay any
system reboots for the first ~30 min of a machine's life. Possibly just
configured on the controller node https://github.com/coreos/zincati/pull/251
* External coordination: For Container Linux, locksmith filled a similar
role and was disabled to allow CLUO to coordinate reboots. By running
atop Kubernetes, it was not possible for the reboot to occur before
cluster bootstrap
* Rely on https://github.com/coreos/zincati/issues/115 to delay the
reboot since bootstrap involves an SSH session
* Use path-based activation of zincati on controllers and set that
path at the end of the bootstrap process
Rel: https://github.com/coreos/fedora-coreos-tracker/issues/239
* Fix delete-node service that runs on worker (cloud-only)
shutdown to delete a Kubernetes node. Regressed in #669
(unreleased)
* Use rkt `--exec` to invoke kubectl binary in the kubelet
image
* Use podman `--entrypoint` to invoke the kubectl binary in
the kubelet image
* Before Kubernetes v1.18.0, Kubelet only supported kubectl
`--limit-bytes` with the Docker `json-file` log driver so
the Fedora CoreOS default was overridden for conformance.
See https://github.com/poseidon/typhoon/pull/642
* Kubelet v1.18+ implemented support for other docker log
drivers, so the Fedora CoreOS default `journald` can be
used again
Rel: https://github.com/kubernetes/kubernetes/issues/86367
* Kubernetes plans to stop releasing the hyperkube container image
* Upstream will continue to publish `kube-apiserver`, `kube-controller-manager`,
`kube-scheduler`, and `kube-proxy` container images to `k8s.gcr.io`
* Upstream will publish Kubelet only as a binary for distros to package,
either as a DEB/RPM on traditional distros or a container image on
container-optimized operating systems
* Typhoon will package the upstream Kubelet (checksummed) and its
dependencies as a container image for use on CoreOS Container Linux,
Flatcar Linux, and Fedora CoreOS
* Update the Typhoon container image security policy to list
`quay.io/poseidon/kubelet`as an official distributed artifact
Hyperkube: https://github.com/kubernetes/kubernetes/pull/88676
Kubelet Container Image: https://github.com/poseidon/kubelet
Kubelet Quay Repo: https://quay.io/repository/poseidon/kubelet
* Quay has historically generated ACI signatures for images to
facilitate rkt's notions of verification (it allowed authors to
actually sign images, though `--trust-keys-from-https` is in use
since etcd and most authors don't sign images). OCI standardization
didn't adopt verification ideas and checking signatures has fallen
out of favor.
* Fix an issue where Quay no longer seems to be generating ACI
signatures for new images (e.g. quay.io/coreos/etcd:v.3.4.4)
* Don't be alarmed by rkt `--insecure-options=image`. It refers
to disabling image signature checking (i.e. docker pull doesn't
check signatures either)
* System containers for Kubelet and bootstrap have transitioned
to the docker:// transport, so there is precedent and this brings
all the system containers on Container Linux controllers into
alignment
* On clouds where workers can scale down or be preempted
(AWS, GCP, Azure), shutdown runs delete-node.service to
remove a node a prevent NotReady nodes from lingering
* Add the delete-node.service that wasn't carried over
from Container Linux and port it to use podman
* Add Terraform strip markers to consume beginning and
trailing whitespace in templated Kubelet arguments for
podman (Fedora CoreOS only)
* Fix initial `worker_node_labels` being quietly ignored
on Fedora CoreOS cloud platforms that offer the feature
* Close https://github.com/poseidon/typhoon/issues/650
* Fix the last minor issue for Fedora CoreOS clusters to pass CNCF's
Kubernetes conformance tests
* Kubelet supports a seldom used feature `kubectl logs --limit-bytes=N`
to trim a log stream to a desired length. Kubelet handles this in the
CRI driver. The Kubelet docker shim only supports the limit bytes
feature when Docker is configured with the default `json-file` logging
driver
* CNCF conformance tests started requiring limit-bytes be supported,
indirectly forcing the log driver choice until either the Kubelet or
the conformance tests are fixed
* Fedora CoreOS defaults Docker to use `journald` (desired). For now,
as a workaround to offer conformant clusters, the log driver can
be set back to `json-file`. RHEL CoreOS likely won't have noticed the
non-conformance since its using crio runtime
* https://github.com/kubernetes/kubernetes/issues/86367
Note: When upstream has a fix, the aim is to drop the docker config
override and use the journald default
* Add Typhoon Fedora CoreOS on Google Cloud as alpha
* Add docs on uploading the Fedora CoreOS GCP gzipped tarball to
Google Cloud storage to create a boot disk image
* Typhoon Google Cloud is compatible with `terraform-provider-google`
v3.x releases
* No v3.x specific features are used, so v2.19+ provider versions are
still allowed, to ease migrations
* Configure kube-proxy --metrics-bind-address=0.0.0.0 (default
127.0.0.1) to serve metrics on 0.0.0.0:10249
* Add firewall rules to allow Prometheus (resides on a worker) to
scrape kube-proxy service endpoints on controllers or workers
* Add a clusterIP: None service for kube-proxy endpoint discovery
* Change kubelet.service on Container Linux nodes to ExecStart Kubelet
inline to replace the use of the host OS kubelet-wrapper script
* Express rkt run flags and volume mounts in a clear, uniform way to
make the Kubelet service easier to audit, manage, and understand
* Eliminate reliance on a Container Linux kubelet-wrapper script
* Typhoon for Fedora CoreOS developed a kubelet.service that similarly
uses an inline ExecStart (except with podman instead of rkt) and a
more minimal set of volume mounts. Adopt the volume improvements:
* Change Kubelet /etc/kubernetes volume to read-only
* Change Kubelet /etc/resolv.conf volume to read-only
* Remove unneeded /var/lib/cni volume mount
Background:
* kubelet-wrapper was added in CoreOS around the time of Kubernetes v1.0
to simplify running a CoreOS-built hyperkube ACI image via rkt-fly. The
script defaults are no longer ideal (e.g. rkt's notion of trust dates
back to quay.io ACI image serving and signing, which informed the OCI
standard images we use today, though they still lack rkt's signing ideas).
* Shipping kubelet-wrapper was regretted at CoreOS, but remains in the
distro for compatibility. The script is not updated to track hyperkube
changes, but it is stable and kubelet.env overrides bridge most gaps
* Typhoon Container Linux nodes have used kubelet-wrapper to rkt/rkt-fly
run the Kubelet via the official k8s.gcr.io hyperkube image using overrides
(new image registry, new image format, restart handling, new mounts, new
entrypoint in v1.17).
* Observation: Most of what it takes to run a Kubelet container is defined
in Typhoon, not in kubelet-wrapper. The wrapper's value is now undermined
by having to workaround its dated defaults. Typhoon may be better served
defining Kubelet.service explicitly
* Typhoon for Fedora CoreOS developed a kubelet.service without the use
of a host OS kubelet-wrapper which is both clearer and eliminated some
volume mounts
* Rename Container Linux Config (CLC) files to *.yaml to align
with Fedora CoreOS Config (FCC) files and for syntax highlighting
* Replace common uses of Terraform `element` (which wraps around)
with `list[index]` syntax to surface index errors
* Original tutorials favored including the platform (e.g.
google-cloud) in modules (e.g. google-cloud-yavin). Prefer
naming conventions where each module / cluster has a simple
name (e.g. yavin) since the platform is usually redundant
* Retain the example cluster naming themes per platform
* Allow generated assets (TLS materials, manifests) to be
securely distributed to controller node(s) via file provisioner
(i.e. ssh-agent) as an assets bundle file, rather than relying
on assets being locally rendered to disk in an asset_dir and
then securely distributed
* Change `asset_dir` from required to optional. Left unset,
asset_dir defaults to "" and no assets will be written to
files on the machine that runs terraform apply
* Enhancement: Managed cluster assets are kept only in Terraform
state, which supports different backends (GCS, S3, etcd, etc) and
optional encryption. terraform apply accesses state, runs in-memory,
and distributes sensitive materials to controllers without making
use of local disk (simplifies use in CI systems)
* Enhancement: Improve asset unpack and layout process to position
etcd certificates and control plane certificates more cleanly,
without unneeded secret materials
Details:
* Terraform file provisioner support for distributing directories of
contents (with unknown structure) has been limited to reading from a
local directory, meaning local writes to asset_dir were required.
https://github.com/poseidon/typhoon/issues/585 discusses the problem
and newer or upcoming Terraform features that might help.
* Observation: Terraform provisioner support for single files works
well, but iteration isn't viable. We're also constrained to Terraform
language features on the apply side (no extra plugins, no shelling out)
and CoreOS / Fedora tools on the receive side.
* Take a map representation of the contents that would have been splayed
out in asset_dir and pack/encode them into a single file format devised
for easy unpacking. Use an awk one-liner on the receive side to unpack.
In pratice, this has worked well and its rather nice that a single
assets file is transferred by file provisioner (all or none)
Rel: https://github.com/poseidon/terraform-render-bootstrap/pull/162
* terraform-provider-google v2.19.0 deprecates `instance_template`
within `google_compute_region_instance_group_manager` in order to
support a scheme with multiple version blocks. Adapt our single
version to the new format to resolve deprecation warnings.
* Fixes: Warning: "instance_template": [DEPRECATED] This field
will be replaced by `version.instance_template` in 3.0.0
* Require terraform-provider-google v2.19.0+ (action required)
* Set small CPU requests on static pods kube-apiserver,
kube-controller-manager, and kube-scheduler to align with
upstream tooling and for edge cases
* Effectively, a practical case for these requests hasn't been
observed. However, a small static pod CPU request may offer
a slight benefit if a controller became overloaded and the
below mechanisms were insufficient
Existing safeguards:
* Control plane nodes are tainted to isolate them from
ordinary workloads. Even dense workloads can only compress
CPU resources on worker nodes.
* Control plane static pods use the highest priority class, so
contention favors control plane pods (over say node-exporter)
and CPU is compressible too.
See: https://github.com/poseidon/terraform-render-bootstrap/pull/161
* Update terraform-render-bootstrap module to adopt the
Terrform v0.12 templatefile function feature to replace
the use of terraform-provider-template's `template_dir`
* Require Terraform v0.12.6+ which adds `for_each`
Background:
* `template_dir` was added to `terraform-provider-template`
to add support for template directory rendering in CoreOS
Tectonic Kubernetes distribution (~2017)
* Terraform v0.12 introduced a native `templatefile` function
and v0.12.6 introduced native `for_each` support (July 2019)
that makes it possible to replace `template_dir` usage
* Document worker pools `node_labels` variable to set the
initial node labels for a homogeneous set of workers
* Document `worker_node_labels` convenience variable to
set the initial node labels for default worker nodes
* Fix issue (present since bootkube->bootstrap switch) where
controller asset copy could fail if /etc/kubernetes/manifests
wasn't created in time on platforms using path activation for
the Kubelet (observed on DigitalOcean, also possible on
bare-metal)
* Drop `node-role.kubernetes.io/master` and
`node-role.kubernetes.io/node` node labels
* Kubelet (v1.16) now rejects the node labels used
in the kubectl get nodes ROLES output
* https://github.com/kubernetes/kubernetes/issues/75457
* Rename render module from bootkube to bootstrap. Avoid
confusion with the kubernetes-incubator/bootkube tool since
it is no longer used
* Use the poseidon/terraform-render-bootstrap Terraform module
(formerly poseidon/terraform-render-bootkube)
* https://github.com/poseidon/terraform-render-bootkube/pull/149
* Run a kube-apiserver, kube-scheduler, and kube-controller-manager
static pod on each controller node. Previously, kube-apiserver was
self-hosted as a DaemonSet across controllers and kube-scheduler
and kube-controller-manager were a Deployment (with 2 or
controller_count many replicas).
* Remove bootkube bootstrap and pivot to self-hosted
* Remove pod-checkpointer manifests (no longer needed)
* Intended as part of #504 improvement
* Single controller clusters only require one controller
instance group (previously created zone-many)
* Multi-controller clusters must "wrap" controllers over
zonal heterogeneous instance groups. For example, 5
controllers over 3 zones (no change)
* Allow updating terraform-provider-ct to any release
beyond v0.3.2, but below v1.0. This relaxes the prior
constraint that allowed only v0.3.y provider versions
* Run kube-apiserver as a non-root user (nobody). User
no longer needs to bind low number ports.
* On most platforms, the kube-apiserver load balancer listens
on 6443 and fronts controllers with kube-apiserver pods using
port 6443. Google Cloud TCP proxy load balancers cannot listen
on 6443. However, GCP's load balancer can be made to listen on
443, while kube-apiserver uses 6443 across all platforms.
* Fix a GCP errata item https://github.com/poseidon/typhoon/wiki/Errata
* Removal of a Google Cloud cluster often required 2 runs of
`terraform apply` because network resource deletes timeout
after 4m. Raise the network deletion timeout to 6m to
ensure apply only needs to be run once to remove a cluster
* google_compute_backend_services use nested blocks to define
backends (instance groups heterogeneous controllers)
* Use Terraform v0.12.x dynamic blocks so the apiserver backend
service can refer to (up to zone-many) controller instance groups
* Previously, with Terraform v0.11.x, the apiserver backend service
had to list a fixed set of backends to span controller nodes across
zones in multi-controller setups. 3 backends were used because each
GCP region offered at least 3 zones. Single-controller clusters had
the cosmetic ugliness of unused instance groups
* Allow controllers to span more than 3 zones if avilable in a
region (e.g. currently only us-central1, with 4 zones)
Related:
* https://www.terraform.io/docs/providers/google/r/compute_backend_service.html
* https://www.terraform.io/docs/configuration/expressions.html#dynamic-blocks
* Replace v0.11 bracket type hints with Terraform v0.12 list expressions
* Use expression syntax instead of interpolated strings, where suggested
* Update Google Cloud tutorial and worker pools documentation
* Define Terraform and plugin version requirements in versions.tf
* Require google ~> 2.5 to support Terraform v0.12
* Require ct ~> 0.3.2 to support Terraform v0.12
* This change affects users who use worker pools on AWS, GCP, or
Azure with a Container Linux derivative
* Rename worker pool modules' `count` variable to `worker_count`,
because `count` will be a reserved variable name in Terraform v0.12
* Fix to remove a trailing slash that was erroneously introduced
in the scripting that updated from v1.14.1 to v1.14.2
* Workaround before this fix was to re-run `terraform init`
* Change flannel port from the kernel default 8472 to the
IANA assigned VXLAN port 4789
* Update firewall rules or security groups for VXLAN
* Why now? Calico now offers its own VXLAN backend so
standardizing on the IANA port will simplify config
* https://github.com/coreos/flannel/blob/master/Documentation/backends.md#vxlan
If no region is set at the Google provider level, Terraform fails to
create the google_compute_target_pool.workers resource and complains
with "Cannot determine region: set in this resource, or set provider-level 'region' or 'zone'."
This commit fixes the issue by explicitly setting the region for the
google_compute_target_pool.workers resource.
* Add an `enable_aggregation` variable to enable the kube-apiserver
aggregation layer for adding extension apiservers to clusters
* Aggregation is **disabled** by default. Typhoon recommends you not
enable aggregation. Consider whether less invasive ways to achieve your
goals are possible and whether those goals are well-founded
* Enabling aggregation and extension apiservers increases the attack
surface of a cluster and makes extensions a part of the control plane.
Admins must scrutinize and trust any extension apiserver used.
* Passing a v1.14 CNCF conformance test requires aggregation be enabled.
Having an option for aggregation keeps compliance, but retains the
stricter security posture on default clusters
* Background: A managed instance group of workers is used in backend
services for global load balancing (HTTP/HTTPS Ingress) and output
for custom global load balancing use cases
* Add worker instances to a target pool load balancing TCP/UDP
applications (NodePort or proxied). Output as `worker_target_pool`
* Health check for workers with a healthy Ingress controller. Forward
rules (regional) to target pools don't support different external and
internal ports so choosing nodes with Ingress allows proxying as a
workaround
* A target pool is a logical grouping only. It doesn't add costs to
clusters or worker pools
* Add calico-ipam CRDs and RBAC permissions
* Switch IPAM from host-local to calico-ipam
* `calico-ipam` subnets `ippools` (defaults to pod CIDR) into
`ipamblocks` (defaults to /26, but set to /24 in Typhoon)
* `host-local` subnets the pod CIDR based on the node PodCIDR
field (set via kube-controller-manager as /24's)
* Create a custom default IPv4 IPPool to ensure the block size
is kept at /24 to allow 110 pods per node (Kubernetes default)
* Retaining host-local was slightly preferred, but Calico v3.6
is migrating all usage to calico-ipam. The codepath that skipped
calico-ipam for KDD was removed
* https://docs.projectcalico.org/v3.6/release-notes/
* Resolve in-addr.arpa and ip6.arpa DNS PTR requests for Kubernetes
service IPs and pod IPs
* Previously, CoreDNS was configured to resolve in-addr.arpa PTR
records for service IPs (but not pod IPs)
* Support terraform-provider-google v1.19.0, v1.19.1, v1.20.0
and v2.0+ (and allow for future 2.x.y releases)
* Require terraform-provider-google v1.19.0 or newer. v1.19.0
introduced `network_interface` fields `network_ip` and `nat_ip`
to deprecate `address` and `assigned_nat_ip`. Those deprecated
fields are removed in terraform-provider-google v2.0
* https://github.com/terraform-providers/terraform-provider-google/releases/tag/v2.0.0
* Assign pod priorityClassNames to critical cluster and node
components (higher is higher priority) to inform node out-of-resource
eviction order and scheduler preemption and scheduling order
* Priority Admission Controller has been enabled since Typhoon
v1.11.1
* Intel Haswell or better is available in every zone around the world
* Neither Kubernetes nor Typhoon have a particular minimum processor
family. However, a few Google Cloud zones still default to Sandy/Ivy
bridge (scheduled to shift April 2019). Price is only based on machine
type so it is beneficial to opt for the next processor family
* Intel Haswell is a suitable minimum since it still allows plenty of
liberty in choosing any region or machine type
* Likely a slight increase to preemption probability in a few zones,
but any lower probability on Sandy/Ivy bridge is due to lower
desirability as they're phased out
* https://cloud.google.com/compute/docs/regions-zones/
* Fix a regression caused by lowering the Kubelet TLS client
certificate to system:nodes group (#100) since dropping
cluster-admin dropped the Kubelet's ability to delete nodes.
* On clouds where workers can scale down (manual terraform apply,
AWS spot termination, Azure low priority deletion), worker shutdown
runs the delete-node.service to remove a node to prevent NotReady
nodes from accumulating
* Allow Kubelets to delete cluster nodes via system:nodes group. Kubelets
acting with system:node and kubelet-delete ClusterRoles is still an
improvement over acting as cluster-admin
* System components that require certificates signed by the cluster
CA can submit a CSR to the apiserver, have an administrator inspect
and approve it, and be issued a certificate
* Configure kube-controller-manager to sign Approved CSR's using the
cluster CA private key
* Admins are responsible for approving or denying CSRs, otherwise,
no certificate is issued. Read the Kubernetes docs carefully and
verify the entity making the request and the authorization level
* https://kubernetes.io/docs/tasks/tls/managing-tls-in-a-cluster
* Use a single admin kubeconfig for initial bootkube bootstrap
and for use by a human admin. Previously, an admin kubeconfig
without a named context was used for bootstrap and direct usage
with KUBECONFIG=path, while one with a named context was used
for `kubectl config use-context` style usage. Confusing.
* Provide the admin kubeconfig via `assets/auth/kubeconfig`,
`assets/auth/CLUSTER-config`, or output `kubeconfig-admin`
* terraform-render-bootkube module deprecated kube_dns_service_ip
output in favor of cluster_dns_service_ip
* Rename k8s_dns_service_ip to cluster_dns_service_ip for
consistency too
* Kubelets can use a lower-privilege TLS client certificate with
Org system:nodes and a binding to the system:node ClusterRole
* Admin kubeconfig's continue to belong to Org system:masters to
provide cluster-admin (available in assets/auth/kubeconfig or as
a Terraform output kubeconfig-admin)
* Remove bare-metal output variable kubeconfig
* Add ServiceAccounts and ClusterRoleBindings for kube-apiserver
and kube-scheduler
* Remove the ClusterRoleBinding for the kube-system default ServiceAccount
* Rename the CA certificate CommonName for consistency with upstream
* On GCP, kubectl port-forward connections to pods are closed
after a timeout (unlike AWS NLB's or Azure load balancers)
* Increase the GCP apiserver backend service timeout from 1 minute
to 5 minutes to be more similar to AWS/Azure LB behavior
* Add kube-router for pod networking and NetworkPolicy
as an experiment
* Experiments are not documented or supported in any way,
and may be removed without notice. They have known issues
and aren't enabled without special options.
* Remove bullet about isolating workloads on workers, its
now common practice and new users will assume it
* List advanced features available in each module
* Fix erroneous Kubernetes version listing for Google Cloud
Fedora Atomic
* Calico Felix has been reporting anonymous usage data about the
version and cluster size, which violates Typhoon's privacy policy
where analytics should be opt-in only
* Add a variable enable_reporting (default: false) to allow opting
in to reporting usage data to Calico (or future components)
* loop sends an initial query to detect infinite forwarding
loops in configured upstream DNS servers and fast exit with
an error (its a fatal misconfiguration on the network that
will otherwise cause resolvers to consume memory/CPU until
crashing, masking the problem)
* https://github.com/coredns/coredns/tree/master/plugin/loop
* loadbalance randomizes the ordering of A, AAAA, and MX records
in responses to provide round-robin load balancing (as usual,
clients may still cache responses though)
* https://github.com/coredns/coredns/tree/master/plugin/loadbalance
* Prefer InternalIP and ExternalIP over the node's hostname,
to match upstream behavior and kubeadm
* Previously, hostname-override was used to set node names
to internal IP's to work around some cloud providers not
resolving hostnames for instances (e.g. DO droplets)
* Updating the `terraform-provider-ct` plugin is known to produce
a `user_data` diff in all pre-existing clusters. Applying the
diff to pre-existing cluster destroys controller nodes
* Ignore changes to controller `user_data`. Once all managed
clusters use a release containing this change, it is possible
to update the `terraform-provider-ct` plugin (worker `user_data`
will still be modified)
* Changing the module `ref` for an existing cluster and
re-applying is still NOT supported (although this PR
would protect controllers from being destroyed)
* Allowing serving IPv6 applications via Kubernetes Ingress
on Typhoon Google Cloud clusters
* Add `ingress_static_ipv6` output variable for use in AAAA
DNS records
* Run at least two replicas of CoreDNS to better support
rolling updates (previously, kube-dns had a pod nanny)
* On multi-master clusters, set the CoreDNS replica count
to match the number of masters (e.g. a 3-master cluster
previously used replicas:1, now replicas:3)
* Add AntiAffinity preferred rule to favor distributing
CoreDNS pods across controller nodes nodes
* Continue to ensure scheduler and controller-manager run
at least two replicas to support performing kubectl edits
on single-master clusters (no change)
* For multi-master clusters, set scheduler / controller-manager
replica count to the number of masters (e.g. a 3-master cluster
previously used replicas:2, now replicas:3)
* Add new bird and felix readiness checks
* Read MTU from ConfigMap veth_mtu
* Add RBAC read for serviceaccounts
* Remove invalid description from CRDs
* Broaden internal-etcd firewall rule to allow etcd client
traffic (2379) from other controller nodes
* Previously, kube-apiservers were only able to connect to their
node's local etcd peer. While master node outages were tolerated,
reaching a healthy peer took longer than neccessary in some cases
* Reduce time needed to bootstrap a cluster
* Release v1.11.1 erroneously left Fedora Atomic clusters using
the v1.11.0 Kubelet. The rest of the control plane ran v1.11.1
as expected
* Update Kubelet from v1.11.0 to v1.11.1 so Fedora Atomic matches
Container Linux
* Container Linux modules were not affected
* Switch Ingress from regional network load balancers to global
HTTP/TCP Proxy load balancing
* Reduce cost by ~$19/month per cluster. Google bills the first 5
global and regional forwarding rules separately. Typhoon clusters now
use 3 global and 0 regional forwarding rules.
* Worker pools no longer include an extraneous load balancer. Remove
worker module's `ingress_static_ip` output.
* Add `ingress_static_ipv4` output variable
* Add `worker_instance_group` output to allow custom global load
balancing
* Deprecate `controllers_ipv4_public` module output
* Deprecate `ingress_static_ip` module output. Use `ingress_static_ipv4`
* Adjust firewall rules, security groups, cloud load balancers,
and generated kubeconfig's
* Facilitates some future simplifications and cost reductions
* Bare-Metal users who exposed kube-apiserver on a WAN via their
router or load balancer will need to adjust its configuration.
This is uncommon, most apiserver are on LAN and/or behind VPN
so no routing infrastructure is configured with the port number
* Use Kubelet bearer token authn/authz to scrape metrics
* Drop RBAC permission from nodes/proxy to nodes/metrics
* Stop proxying kubelet scrapes through the apiserver, since
this required higher privilege (nodes/proxy) and can add
load to the apiserver on large clusters
* Raise minimum Terraform version to v0.11.0
* Terraform v0.11.x has been supported since Typhoon v1.9.2
and Terraform v0.10.x was last released in Nov 2017. I'd like
to stop worrying about v0.10.x and remove migration docs as
a later followup
* Migration docs docs/topics/maintenance.md#terraform-v011x
* Observed frequent kube-scheduler and controller-manager
restarts with Calico as the CNI provider. Root cause was
unclear since control plane was functional and tests of
pod to pod network connectivity passed
* Root cause: Calico sets up cali* and tunl* network interfaces
for containers on hosts. NetworkManager tries to manage these
interfaces. It periodically disconnected veth pairs. Logs did
not surface this issue since its not an error per-se, just Calico
and NetworkManager dueling for control. Kubernetes correctly
restarted pods failing health checks and ensured 2 replicas were
running so the control plane functioned mostly normally. Pod to
pod connecitivity was only affected occassionally. Pain to debug.
* Solution: Configure NetworkManager to ignore the Calico ifaces
per Calico's recommendation. Cloud-init writes files after
NetworkManager starts, so a restart is required on first boot. On
subsequent boots, the file is present so no restart is needed
* (containerized) kube-proxy warns that it is unable to
load the ip_vs kernel module despite having the correct
mounts. Atomic uses an xz compressed module and modprobe
in the container was not compiled with compression support
* Workaround issue for now by always loading ip_vs on-host
* https://github.com/kubernetes/kubernetes/issues/60
* Use the upstream bootkube image packaged with the
required metadata to be usable as a system container
under systemd
* Run bootkube with runc so no host level components
use Docker any more. Docker is still the runtime
* Remove bootkube script and old systemd unit
* Change kubelet system image to use --cgroups-per-qos=true
(default) instead of false
* Change kubelet system image to use --enforce-node-allocatable=pods
instead of an empty string
* Fix kubelet port-forward on Google Cloud / Fedora Atomic
* Mount the host's /etc/hosts in kubelet system containers
* Problem: kubelet runc system containers on Atomic were not
mounting the host's /etc/hosts, like rkt-fly does on Container
Linux. `kubectl port-forward` calls socat with localhost. DNS
servers on AWS, DO, and in many bare-metal environments resolve
localhost to the caller as a convenience. Google Cloud notably
does not nor is it required to do so and this surfaced the
missing /etc/hosts in runc kubelet namespaces.
* Allow multi-controller clusters on Google Cloud
* GCP regional network load balancers have a long open
bug in which requests originating from a backend instance
are routed to the instance itself, regardless of whether
the health check passes or not. As a result, only the 0th
controller node registers. We've recommended just using
single master GCP clusters for a while
* https://issuetracker.google.com/issues/67366622
* Workaround issue by switching to a GCP TCP Proxy load
balancer. TCP proxy lb routes traffic to a backend service
(global) of instance group backends. In our case, spread
controllers across 3 zones (all regions have 3+ zones) and
organize them in 3 zonal unmanaged instance groups that
serve as backends. Allows multi-controller cluster creation
* GCP network load balancers only allowed legacy HTTP health
checks so kubelet 10255 was checked as an approximation of
controller health. Replace with TCP apiserver health checks
to detect unhealth or unresponsive apiservers.
* Drawbacks: GCP provision time increases, tailed logs now
timeout (similar tradeoff in AWS), controllers only span 3
zones instead of the exact number in the region
* Workaround in Typhoon has been known and posted for 5 months,
but there still appears to be no better alternative. Its
probably time to support multi-master and accept the downsides
* Expose etcd metrics to workers so Prometheus can
run on a worker, rather than a controller
* Drop temporary firewall rules allowing Prometheus
to run on a controller and scrape targes
* Related to https://github.com/poseidon/typhoon/pull/175
* Use etcd v3.3 --listen-metrics-urls to expose only metrics
data via http://0.0.0.0:2381 on controllers
* Add Prometheus discovery for etcd peers on controller nodes
* Temporarily drop two noisy Prometheus alerts
* AWS and Google Cloud make use of auto-scaling groups
and managed instance groups, respectively. As such, the
kubeconfig is already held in cloud user-data
* Controller instances are provisioned with a kubeconfig
from user-data. Its redundant to use a Terraform remote
file copy step for the kubeconfig.
Dalton Hubble
Suraj Deshmukh
Valer Cara
Jordan Pittier
JordanP
Lucas Servén Marín
Ben Drucker
William Zhang