From 4d03016c4ea8ad561d875f0dd3e031dbf38d1522 Mon Sep 17 00:00:00 2001 From: Michael Taufen Date: Fri, 25 May 2018 16:30:11 -0700 Subject: [PATCH] WIP update dynamic kubelet config docs for v1.11 --- .../administer-cluster/reconfigure-kubelet.md | 387 +++++++----------- 1 file changed, 155 insertions(+), 232 deletions(-) diff --git a/content/en/docs/tasks/administer-cluster/reconfigure-kubelet.md b/content/en/docs/tasks/administer-cluster/reconfigure-kubelet.md index d6af91637372b..343ef37e8cb31 100644 --- a/content/en/docs/tasks/administer-cluster/reconfigure-kubelet.md +++ b/content/en/docs/tasks/administer-cluster/reconfigure-kubelet.md @@ -7,28 +7,28 @@ content_template: templates/task --- {{% capture overview %}} -{{< feature-state state="alpha" >}} -As of Kubernetes 1.8, the new -[Dynamic Kubelet Configuration](https://github.com/kubernetes/features/issues/281) -feature is available in alpha. This allows you to change the configuration of -Kubelets in a live Kubernetes cluster via first-class Kubernetes concepts. -Specifically, this feature allows you to configure individual Nodes' Kubelets -via ConfigMaps. +{{< feature-state for_k8s_version="v1.11" state="beta" >}} + +The [Dynamic Kubelet Configuration](https://github.com/kubernetes/features/issues/281) +feature allows you to change the configuration of each Kubelet in a live Kubernetes +cluster by deploying a ConfigMap and configuring each Node to use it. **Warning:** All Kubelet configuration parameters may be changed dynamically, but not all parameters are safe to change dynamically. This feature is intended for system experts who have a strong understanding of how configuration changes -will affect behavior. No documentation currently exists which plainly lists +will affect behavior. In general, you should always carefully test config changes +on a small set of nodes before rolling them out to your entire cluster. +No documentation currently exists which plainly lists "safe to change" fields, but we plan to add it before this feature graduates -from alpha. +from alpha. // TODO(mtaufen): this will be made available very very soon (in v1.11) on the type spec, in https://github.com/kubernetes/kubernetes/pull/64354 {{% /capture %}} {{% capture prerequisites %}} -- A live Kubernetes cluster with both Master and Node at v1.8 or higher must be -running, with the `DynamicKubeletConfig` feature gate enabled and the Kubelet's -`--dynamic-config-dir` flag set to a writable directory on the Node. +- A live Kubernetes cluster with both Master and Node at v1.11 or higher must +be running and the Kubelet's `--dynamic-config-dir` flag must be set to a +writable directory on the Node. This flag must be set to enable Dynamic Kubelet Configuration. -- The kubectl command-line tool must be also v1.8 or higher, and must be +- The kubectl command-line tool must be v1.11 or higher, and must be configured to communicate with the cluster. {{% /capture %}} @@ -57,11 +57,10 @@ and is overridden by command-line flags. Unspecified values in the new configura will receive default values appropriate to the configuration version (e.g. `kubelet.config.k8s.io/v1beta1`), unless overridden by flags. -The status of the Node's Kubelet configuration is reported via the `KubeletConfigOK` -condition in the Node status. Once you have updated a Node to use the new -ConfigMap, you can observe this condition to confirm that the Node is using the -intended configuration. A table describing the possible conditions can be found -at the end of this article. +The status of the Node's Kubelet configuration is reported via +`Node.Spec.Status.Config`. Once you have updated a Node to use the new +ConfigMap, you can observe this status to confirm that the Node is using the +intended configuration. This document describes editing Nodes using `kubectl edit`. There are other ways to modify a Node's spec, including `kubectl patch`, for @@ -70,16 +69,17 @@ example, which facilitate scripted workflows. This document only describes a single Node consuming each ConfigMap. Keep in mind that it is also valid for multiple Nodes to consume the same ConfigMap. -### Node Authorizer Workarounds +**Warning:** Note that while it is *possible* to change the configuration by +updating the ConfigMap in-place, this will cause all Kubelets configured with +that ConfigMap to update simultaneously. It is much safer to treat ConfigMaps +as immutable by convention, aided by `kubectl`'s `--append-hash` option, +and incrementally roll out updates to `Node.Spec.ConfigSource`. -The Node Authorizer does not yet pay attention to which ConfigMaps are assigned -to which Nodes. If you currently use the Node authorizer, your Kubelets will not -be automatically granted permission to download their respective ConfigMaps. +### Note Regarding the Node Authorizer -The temporary workaround used in this document is to manually create the RBAC -Roles and RoleBindings for each ConfigMap. The Node Authorizer will be extended -before the Dynamic Kubelet Configuration feature graduates from alpha, so doing -this in production should never be necessary. +Old versions of this document required users to manually create RBAC rules +for Nodes to access their assigned ConfigMaps. The Node Authorizer now +automatically configures these rules, so this step is no longer necessary. ### Generating a file that contains the current configuration @@ -90,12 +90,13 @@ and debug issues. The compromise, however, is that you must start with knowledge of the existing configuration to ensure that you only change the fields you intend to change. -In the future, the Kubelet will be bootstrapped from a file on disk +In the future, the Kubelet will be bootstrapped from just a file on disk (see [Set Kubelet parameters via a config file](/docs/tasks/administer-cluster/kubelet-config-file)), and you will simply edit a copy of this file (which, as a best practice, should live in version control) while creating the first Kubelet ConfigMap. Today, -however, the Kubelet is still bootstrapped with command-line flags. Fortunately, -there is a dirty trick you can use to generate a config file containing a Node's +however, the Kubelet is bootstrapped with a combination of this file and command-line flags +that can override the configuration in the file. +Fortunately, there is a dirty trick you can use to generate a config file containing a Node's current configuration. The trick involves accessing the Kubelet server's `configz` endpoint via the kubectl proxy. This endpoint, in its current implementation, is intended to be used only as a debugging aid, which is part of why this is a @@ -152,39 +153,14 @@ metadata: uid: 946d785e-998a-11e7-a8dd-42010a800006 ``` -Note that the configuration data must appear under the ConfigMap's -`kubelet` key. - We create the ConfigMap in the `kube-system` namespace, which is appropriate because this ConfigMap configures a Kubernetes system component - the Kubelet. The `--append-hash` option appends a short checksum of the ConfigMap contents to the name. This is convenient for an edit->push workflow, as it will -automatically, yet deterministically, generate new names for new ConfigMaps. - -We use the `-o yaml` output format so that the name, namespace, and uid are all -reported following creation. We will need these in the next step. We will refer -to the name as CONFIG_MAP_NAME and the uid as CONFIG_MAP_UID. - -### Authorize your Node to read the new ConfigMap - -Now that you've created a new ConfigMap, you need to authorize your node to -read it. First, create a Role for your new ConfigMap with the -following commands: - -``` -$ export CONFIG_MAP_NAME=name-from-previous-output -$ kubectl -n kube-system create role ${CONFIG_MAP_NAME}-reader --verb=get --resource=configmap --resource-name=${CONFIG_MAP_NAME} -``` - -Next, create a RoleBinding to associate your Node with the new Role: - -``` -$ kubectl -n kube-system create rolebinding ${CONFIG_MAP_NAME}-reader --role=${CONFIG_MAP_NAME}-reader --user=system:node:${NODE_NAME} -``` - -Once the Node Authorizer is updated to do this automatically, you will -be able to skip this step. +automatically, yet deterministically, generate new names for new ConfigMaps. +We will refer to the name that includes this generated hash as +`CONFIG_MAP_NAME` below. ### Set the Node to use the new configuration @@ -199,46 +175,79 @@ Once in your editor, add the following YAML under `spec`: ``` configSource: - configMapRef: + configMap: name: CONFIG_MAP_NAME namespace: kube-system - uid: CONFIG_MAP_UID + kubeletConfigKey: kubelet ``` -Be sure to specify all three of `name`, `namespace`, and `uid`. +Be sure to specify all three of `name`, `namespace`, and `kubeletConfigKey`. +The last parameter tells the Kubelet which key of the ConfigMap it can find +its config in. ### Observe that the Node begins using the new configuration -Retrieve the Node with `kubectl get node ${NODE_NAME} -o yaml`, and look for the -`KubeletConfigOK` condition in `status.conditions`. You should see the message -`Using current (UID: CONFIG_MAP_UID)` when the Kubelet starts using the new -configuration. +Retrieve the Node with `kubectl get node ${NODE_NAME} -o yaml`, and inspect +`Node.Status.Config`. You should see the config sources corresponding to the `active`, +`assigned`, and `lastKnownGood` configurations reported in the status. The `active` +configuration is the version the Kubelet is currently running with, the `assigned` +configuration is the latest version the Kubelet has resolved based on +`Node.Spec.ConfigSource`, and the `lastKnownGood` configuration is the version the +Kubelet will fall back to if an invalid config is assigned in `Node.Spec.ConfigSource`. + +You might not see `lastKnownGood` appear in the status if it is set to its default value, +the local config deployed with the node. The status will update `lastKnownGood` to +match a valid `assigned` config after the Kubelet becomes comfortable with the config. +The details of how the Kubelet determines a config should become the `lastKnownGood` are +not guaranteed by the API, though it may be useful, for debugging purposes, to know that +this is presently implemented as a 10-minute grace period. For convenience, you can use the following command (using `jq`) to filter down -to the `KubeletConfigOK` condition: - -``` -$ kubectl get no ${NODE_NAME} -o json | jq '.status.conditions|map(select(.type=="KubeletConfigOK"))' -[ - { - "lastHeartbeatTime": "2017-09-20T18:08:29Z", - "lastTransitionTime": "2017-09-20T18:08:17Z", - "message": "using current: /api/v1/namespaces/kube-system/configmaps/my-node-config-gkt4c2m4b2", - "reason": "passing all checks", - "status": "True", - "type": "KubeletConfigOK" +to the config status: + +``` +$ kubectl get no ${NODE_NAME} -o json | jq '.status.config' +{ + "active": { + "configMap": { + "kubeletConfigKey": "kubelet", + "name": "my-node-config-9mbkccg2cc", + "namespace": "kube-system", + "resourceVersion": "1326", + "uid": "705ab4f5-6393-11e8-b7cc-42010a800002" + } + }, + "assigned": { + "configMap": { + "kubeletConfigKey": "kubelet", + "name": "my-node-config-9mbkccg2cc", + "namespace": "kube-system", + "resourceVersion": "1326", + "uid": "705ab4f5-6393-11e8-b7cc-42010a800002" + } + }, + "lastKnownGood": { + "configMap": { + "kubeletConfigKey": "kubelet", + "name": "my-node-config-9mbkccg2cc", + "namespace": "kube-system", + "resourceVersion": "1326", + "uid": "705ab4f5-6393-11e8-b7cc-42010a800002" + } } -] +} + ``` -If something goes wrong, you may see one of several different error conditions, -detailed in the table of KubeletConfigOK conditions, below. When this happens, you -should check the Kubelet's log for more details. +If something goes wrong, the Kubelet will report any configuration related errors +in `Node.Status.Config.Error`. You may see one of several possible errors, which +are detailed in a table at the end of this article. If you see any of these errors, +you can search for the same error message in the Kubelet's log for additional details. ### Edit the configuration file again To change the configuration again, we simply repeat the above workflow. -Try editing the `kubelet` file, changing the previously changed parameter to a +Try editing the `kubelet_configz_${NODE_NAME}` file, changing the previously changed parameter to a new value. ### Push the newly edited configuration to the control plane @@ -247,209 +256,123 @@ Push the new configuration to the control plane in a new ConfigMap with the following command: ``` -$ kubectl create configmap my-node-config --namespace=kube-system --from-file=kubelet=kubelet_configz_${NODE_NAME} --append-hash -o yaml +$ kubectl -n kube-system create configmap my-node-config --from-file=kubelet=kubelet_configz_${NODE_NAME} --append-hash -o yaml ``` This new ConfigMap will get a new name, as we have changed the contents. -We will refer to the new name as NEW_CONFIG_MAP_NAME and the new uid -as NEW_CONFIG_MAP_UID. - -### Authorize your Node to read the new ConfigMap - -Now that you've created a new ConfigMap, you need to authorize your node to -read it. First, create a Role for your new ConfigMap with the -following commands: - -``` -$ export NEW_CONFIG_MAP_NAME=name-from-previous-output -$ kubectl -n kube-system create role ${NEW_CONFIG_MAP_NAME}-reader --verb=get --resource=configmap --resource-name=${NEW_CONFIG_MAP_NAME} -``` - -Next, create a RoleBinding to associate your Node with the new Role: - -``` -$ kubectl -n kube-system create rolebinding ${NEW_CONFIG_MAP_NAME}-reader --role=${NEW_CONFIG_MAP_NAME}-reader --user=system:node:${NODE_NAME} -``` - -Once the Node Authorizer is updated to do this automatically, you will -be able to skip this step. +We will refer to the new name as `NEW_CONFIG_MAP_NAME`. ### Configure the Node to use the new configuration -Once more, edit the Node's `spec.configSource` with -`kubectl edit node ${NODE_NAME}`. Your new `spec.configSource` should look like -the following, with `name` and `uid` substituted as necessary: +Once more, edit `Node.Spec.ConfigSource` via `kubectl edit node ${NODE_NAME}`. +Your new `Node.Spec.ConfigSource` should look like the following, +with `${NEW_CONFIG_MAP_NAME}` substituted as necessary: ``` configSource: - configMapRef: + configMap: name: ${NEW_CONFIG_MAP_NAME} namespace: kube-system - uid: ${NEW_CONFIG_MAP_UID} + kubeletConfigKey: kubelet ``` ### Observe that the Kubelet is using the new configuration Once more, retrieve the Node with `kubectl get node ${NODE_NAME} -o yaml`, and -look for the `KubeletConfigOK` condition in `status.conditions`. You should see the message -`using current: /api/v1/namespaces/kube-system/configmaps/${NEW_CONFIG_MAP_NAME}` when the Kubelet starts using the -new configuration. - -### Deauthorize your Node fom reading the old ConfigMap - -Once you know your Node is using the new configuration and are confident that -the new configuration has not caused any problems, it is a good idea to -deauthorize the node from reading the old ConfigMap. Run the following -commands to remove the RoleBinding and Role: - -``` -$ kubectl -n kube-system delete rolebinding ${CONFIG_MAP_NAME}-reader -$ kubectl -n kube-system delete role ${CONFIG_MAP_NAME}-reader -``` - -Note that this does not necessarily prevent the Node from reverting to the old -configuration, as it may locally cache the old ConfigMap for an indefinite -period of time. - -You may optionally also choose to remove the old ConfigMap: - -``` -$ kubectl -n kube-system delete configmap ${CONFIG_MAP_NAME} -``` - -Once the Node Authorizer is updated to do this automatically, you will -be able to skip this step. +look for a `Node.Status.Config` that reports the new configuration as `assigned` +and `active`, with no errors. ### Reset the Node to use its local default configuration Finally, if you wish to reset the Node to use the configuration it was provisioned with, simply edit the Node with `kubectl edit node ${NODE_NAME}` and -remove the `spec.configSource` subfield. +remove the `Node.Spec.ConfigSource` field. ### Observe that the Node is using its local default configuration -After removing this subfield, you should eventually observe that the KubeletConfigOK -condition's message reverts to `using current: local`. +After removing this subfield, you should eventually observe that `Node.Status.Config` +has become empty, as all config sources have been reset to `nil` (indicating the local +default config is `assigned`, `active`, and `lastKnownGood`), and no error is reported. -### Deauthorize your Node fom reading the old ConfigMap +{{% /capture %}} -Once you know your Node is using the default configuration again, it is a good -idea to deauthorize the node from reading the old ConfigMap. Run the following -commands to remove the RoleBinding and Role: +{{% capture discussion %}} +## Kubectl Patch Example +As mentioned above, there are many ways to change a Node's configSource. +Here is an example command that uses `kubectl patch`: ``` -$ kubectl -n kube-system delete rolebinding ${NEW_CONFIG_MAP_NAME}-reader -$ kubectl -n kube-system delete role ${NEW_CONFIG_MAP_NAME}-reader +kubectl patch node ${NODE_NAME} -p "{\"spec\":{\"configSource\":{\"configMap\":{\"name\":\"${CONFIG_MAP_NAME}\",\"namespace\":\"kube-system\",\"kubeletConfigKey\":\"kubelet\"}}}}" ``` -Note that this does not necessarily prevent the Node from reverting to the old -ConfigMap, as it may locally cache the old ConfigMap for an indefinite -period of time. +## Understanding how the Kubelet checkpoints config -You may optionally also choose to remove the old ConfigMap: +When a new config is assigned to the Node, the Kubelet downloads and unpacks the +config payload as a set of files on local disk. The Kubelet also records metadata +that locally tracks the assigned and last-known-good config sources, so that the +Kubelet knows which config to use across restarts, even if the API server becomes +unavailable. After checkpointing a config and the relevant metadata, the Kubelet +will exit if the assigned config has changed. When the Kubelet is restarted by the +babysitter process, it will read the new metadata, and use the new config. -``` -$ kubectl -n kube-system delete configmap ${NEW_CONFIG_MAP_NAME} -``` +The recorded metadata is fully resolved, meaning that it contains all necessary +information to choose a specific config version - typically a `UID` and `ResourceVersion`. +This is in contrast to `Node.Spec.ConfigSource`, where the intended config is declared +via the idempotent `namespace/name` that identifies the target ConfigMap; the Kubelet +tries to use the latest version of this ConfigMap. -Once the Node Authorizer is updated to do this automatically, you will -be able to skip this step. - -{{% /capture %}} - -{{% capture discussion %}} -## Kubectl Patch Example -As mentioned above, there are many ways to change a Node's configSource. -Here is an example command that uses `kubectl patch`: +It can sometimes be useful to inspect the Kubelet's config metadata and checkpoints +when debugging a Node. The structure of the Kubelet's checkpointing directory is as follows: ``` -kubectl patch node ${NODE_NAME} -p "{\"spec\":{\"configSource\":{\"configMapRef\":{\"name\":\"${CONFIG_MAP_NAME}\",\"namespace\":\"kube-system\",\"uid\":\"${CONFIG_MAP_UID}\"}}}}" +- --dynamic-config-dir (root for managing dynamic config) +| - meta + | - assigned (encoded kubeletconfig/v1beta1.SerializedNodeConfigSource object, indicating the assigned config) + | - last-known-good (encoded kubeletconfig/v1beta1.SerializedNodeConfigSource object, indicating the last-known-good config) +| - checkpoints + | - uid1 (dir for versions of object identified by uid1) + | - resourceVersion1 (dir for unpacked files from resourceVersion1 of object with uid1) + | - ... + | - ... ``` -## Understanding KubeletConfigOK Conditions +## Understanding Node.Status.Config.Error messages -The following table describes several of the `KubeletConfigOK` Node conditions you -might encounter in a cluster that has Dynamic Kubelet Config enabled. If you -observe a condition with `status=False`, you should check the Kubelet log for -more error details by searching for the message or reason text. +The following table describes the error messages you might encounter +when using Dynamic Kubelet Config. You can search for the same text +as the error message in the Kubelet log for additional details +on the error. - -
- - - + + + + + + - - - + + - - - + + - - - + + - - - + + - - - + +
Possible MessagesPossible ReasonsStatusError MessagePossible Causes

failed to load config, see Kubelet log for details

The Kubelet likely could not parse the downloaded config payload, or encountered a filesystem error attempting to load the payload from disk.

using current: local

when the config source is nil, the Kubelet uses its local config

True

failed to validate config, see Kubelet log for details

The configuration in the payload, combined with any command-line flag overrides, and the sum of feature gates from flags, the config file, and the remote payload, was determined to be invalid by the Kubelet.

using current: /api/v1/namespaces/${CURRENT_CONFIG_MAP_NAMESPACE}/configmaps/${CURRENT_CONFIG_MAP_NAME}

passing all checks

True

invalid NodeConfigSource, exactly one subfield must be non-nil, but all were nil

Since Node.Spec.ConfigSource is validated by the API server to contain at least one non-nil subfield, this likely means that the Kubelet is older than the API server and does not recognize a newer source type.

using last-known-good: local

 -
    -
  • failed to load current: /api/v1/namespaces/${CURRENT_CONFIG_MAP_NAMESPACE}/configmaps/${CURRENT_CONFIG_MAP_NAME}
    • -
  • failed to parse current: /api/v1/namespaces/${CURRENT_CONFIG_MAP_NAMESPACE}/configmaps/${CURRENT_CONFIG_MAP_NAME}
    • -
  • failed to validate current: /api/v1/namespaces/${CURRENT_CONFIG_MAP_NAMESPACE}/configmaps/${CURRENT_CONFIG_MAP_NAME}
    • -
-

False

failed to sync: failed to download config, see Kubelet log for details

The Kubelet could not download the config. It is possible that Node.Spec.ConfigSource could not be resolved to a concrete API object, or that network errors disrupted the download attempt. The Kubelet will retry the download when in this error state.

using last-known-good: /api/v1/namespaces/${LAST_KNOWN_GOOD_CONFIG_MAP_NAMESPACE}/configmaps/${LAST_KNOWN_GOOD_CONFIG_MAP_NAME}

 -
    -
  • failed to load current: /api/v1/namespaces/${CURRENT_CONFIG_MAP_NAMESPACE}/configmaps/${CURRENT_CONFIG_MAP_NAME}
    • -
  • failed to parse current: /api/v1/namespaces/${CURRENT_CONFIG_MAP_NAMESPACE}/configmaps/${CURRENT_CONFIG_MAP_NAME}
    • -
  • failed to validate current: /api/v1/namespaces/${CURRENT_CONFIG_MAP_NAMESPACE}/configmaps/${CURRENT_CONFIG_MAP_NAME}
    • -
-

False

failed to sync: internal failure, see Kubelet log for details

The Kubelet encountered some internal problem and failed to update its config as a result. Examples include filesystem errors and reading objects from the internal informer cache.
-

- The reasons in the next column could potentially appear for any of - the above messages. -

-

- This condition indicates that the Kubelet is having trouble - reconciling `spec.configSource`, and thus no change to the in-use - configuration has occurred. -

-

- The "failed to sync" reasons are specific to the failure that - occurred, and the next column does not necessarily contain all - possible failure reasons. -

- 		-

failed to sync, reason:

-
    -
  • failed to read Node from informer object cache
    • -
  • failed to reset to local config
    • -
  • invalid NodeConfigSource, exactly one subfield must be non-nil, but all were nil
    • -
  • invalid ObjectReference, all of UID, Name, and Namespace must be specified
    • -
  • invalid ConfigSource.ConfigMapRef.UID: ${UID} does not match ${API_PATH}.UID: ${UID_OF_CONFIG_MAP_AT_API_PATH}
    • -
  • failed to determine whether object ${API_PATH} with UID ${UID} was already checkpointed
    • -
  • failed to download ConfigMap with name ${NAME} from namespace ${NAMESPACE}
    • -
  • failed to save config checkpoint for object ${API_PATH} with UID ${UID}
    • -
  • failed to set current config checkpoint to local config
    • -
  • failed to set current config checkpoint to object ${API_PATH} with UID ${UID}
    • -
-

False

internal failure, see Kubelet log for details

The Kubelet encountered some internal problem while manipulating config, outside of the configuration sync loop.
-{{% /capture %}} - - +{{% /capture %}} \ No newline at end of file