csi/server.statefulset: custom security context
This adds flexibility to have custom pod template and container
`securityContext` and preserves current default values and behavior.
Fixes https://github.com/hashicorp/vault-helm/issues/663.
This also is a way to address https://github.com/hashicorp/vault-helm/pull/599
so that people can specify, for example, the CSI to run in a privileged
container for OpenShift.
This is a follow-up to https://github.com/hashicorp/vault-helm/pull/750
and builds on the same principles.
Side note: I am not able to run `helm schema-gen` since it is
unmaintained and does not work with M1 Macs.
Allow the injector's pod- and container-level securityContext to be
fully specified by the user, via new options
`injector.securityContext.pod` and
`injector.securityContext.container` with more complete
defaults. Deprecates `injector.uid` and `injector.gid`.
If `injector.uid` or `injector.gid` are set by the user, the old pod
securityContext settings will be used. Otherwise the new defaults and
settings are used.
Co-authored-by: Theron Voran <tvoran@users.noreply.github.com>
Set default object selector for webhooks to exclude injector itself
If `injector.failurePolicy` is set to `Fail`, there is a race condition
where if the mutating webhook config is setup before the injector, then
the injector can fail to start because it tries to inject itself.
We can work around this by ignoring the injector pod in in the webhook
by default.
Thanks to @joeyslalom for the object selector to exclude the pod.
Fixes https://github.com/hashicorp/vault-k8s/issues/258
VAULT-571 Matching documented behavior and consul
Consul's helm template defaults most of the enabled to the special value
`"-"`, which means to inherit from global. This is what is implied
should happen in Vault as well according to the documentation for the
helm chart:
> [global.enabled] The master enabled/disabled configuration. If this is
> true, most components will be installed by default. If this is false,
> no components will be installed by default and manually opting-in is
> required, such as by setting server.enabled to true.
(https://www.vaultproject.io/docs/platform/k8s/helm/configuration#enabled)
We also simplified the chart logic using a few template helpers.
Co-authored-by: Theron Voran <tvoran@users.noreply.github.com>
* Prepare default values for MutatingWebhookConfiguration #691
* Add values.yaml values to injector-mutating-webhook.yaml #691
* Duplicate and deprecate top-level webhook settings and put them in a webhook object
* Made the new values default with the fallback to the old values.yaml
* Fix _helpers.tpl to support both old and new webhook annotations
* Add new tests and deprecate old ones for injector webhook configuration
* Old tests now work with old values.yaml
* Add all new fields showing that they have priority over old ones
* Add deprecation note to injector.failurePolicy #691
Sets up a vault-enterprise license for autoloading on vault
startup. Mounts an existing secret to /vault/license and sets
VAULT_LICENSE_PATH appropriately.
* fix serviceaccount and clusterrole name reference (full name)
* add server.enabled option, align with documentation
* add unit tests
* update server.enabled behaviour to explicit true and update tests
* don't set VAULT_DEV_ROOT_TOKEN_ID by default in dev mode
* don't template environment variables that no longer exist
* fix tests after removing VAULT_DEV_ROOT_TOKEN_ID env variable
* removed a typo
* allow overriding VAULT_DEV_ROOT_TOKEN_ID in dev mode
* correct ambiguous description
* don't set default values in templates for visibility, update tests and set uncomment devRootToken in values.yaml
* Update devRootToken description
* Make serviceAccount name a configuration option
Follow Helm Best Practices when defining serviceAccount names
https://helm.sh/docs/chart_best_practices/#using-rbac-resources
* Use enabled instead of create for consistency
* Add unit tests for user-defined service account name
* ServiceAccount under server
Co-authored-by: David Holsgrove <david@apnic.net>
* Update ServiceAccount in RoleBindings
to address https://github.com/hashicorp/vault-helm/pull/56#pullrequestreview-297856433
Co-authored-by: David Holsgrove <david@apnic.net>
* Update tests for helm template arg --show-only
Co-authored-by: David Holsgrove <david@apnic.net>
* Fix server-serviceaccount tests
* serviceAccount: rename enabled to create
* statefulSet: add tests for serviceAccount
Co-authored-by: Nick Satterly <nick@diabol.se>
Co-authored-by: David Holsgrove <david@apnic.net>
* Initial commit
* Added openshift flag
* added self signed certificate for service annotation
* added OpenShift flag
* Added OpenShift flag
* cleanup
* Cleanup
* Further cleanup
* Further cleanup
* reverted security context on injector
* Extra corrections
* cleanup
* Removed Raft config for OpenShift, removed generated certs for ha and standby services
* Add openshift flag to global block, route disabled by default, condition for injector in network policy
* Added Unit tests for OpenShift
* Fixed unit test for HA statefulset for OpenShift
* Removed debug log level from stateful set
* Added port 8201 to networkpolicy
* Updated injector image
* Add openshift beta support
* Add openshift beta support
* Remove comments from configs
* Remove vault-k8s note from values
* Change route to use active service when HA
Co-authored-by: Radu Domnu <radu.domnu@sixdx.com>
Co-authored-by: Radu Domnu <radu.domnu@gmail.com>
Changed/added helper functions to detect if the annotations value
is a string or yaml, and apply `tpl` or `toYaml`
accordingly. Defaults are left as `{}` since yaml is more likely
to be used with helm on the command line. This means a warning
will be shown when setting an annotation to a multi-line
string (which has been the existing behavior).
* Change config specification
As it is right now, the specification of the config is done through an
string. When using storage backends like PostgreSQL, the password for the
database has to be included in the config variable of the values file.
This change allows to specify the configuration through a map, making
the chart GitOps friendly. Now, sensitive values can be stored in a
different values file or passed on deployment time with --set.
To have a very generic specification:
- I've assumed that the combination stanza (eg. storage) name (eg. file)
is unique.
- Quoted values for all stanza parameters. I tested a generated
configuration in a vault docker image and it seems to work just fine.
* Change config format to json
* Add conditional formatting
* Add config for raft mode
Annotations for various objects were either multi-line strings or yaml
maps strings, so this is making them all multi-line strings for
consistency. Also updated the doc comment for namespaceSelector, since
it's being read as a yaml map (toYaml).
Adds affinity, tolerations, and nodeSelector options for the
injector deployment that are separate from those options on the vault
server statefulset.
Co-authored-by: Sergei Shishov <sergei.shishov@dubizzle.com>
Uses Values.injector.externalVaultAddr to control the vault address
env variable and server yaml rendering.
If injector.externalVaultAddr is empty, both the injector and vault
are deployed, with the injector using the local vault. If
injector.externalVaultAddr is not empty, only the injector is
deployed, and it uses the vault at the address specified in
injector.externalVaultAddr.
