- The flag
--watch-namespace
and--operator-namespace
was added tooperator-sdk run --local
,operator-sdk test --local
andoperator-sdk cleanup
commands in order to replace the flag--namespace
which was deprecated.(#2617) - The methods
ctx.GetOperatorNamespace()
andctx.GetWatchNamespace()
was addedpkg/test
in order to replacectx.GetNamespace()
which is deprecated. (#2617) - The
--crd-version
flag was added to thenew
,add api
,add crd
, andgenerate crds
commands so that users can opt-in tov1
CRDs. (#2684) - The printout for the compatible Kubernetes Version #2446
- The scorecard when creating a Custom Resource, will produce a message to the user if that CR already exists. (#2683)
- Breaking Change: The
--namespace
flag fromoperator-sdk run --local
,operator-sdk test --local
andoperator-sdk cleanup
command was deprecated and will be removed in the future versions. Use--watch-namespace
and--operator-namespace
instead of. (#2617) - Breaking Change: The method
ctx.GetNamespace()
from thepkg/test
is deprecated and will be removed in the future versions. Usectx.GetOperatorNamespace()
andctx.GetWatchNamespace()
instead of. (#2617)
- Breaking Change: remove
pkg/restmapper
which was deprecated inv0.14.0
. Projects that use this package must switch to theDynamicRESTMapper
implementation in controller-runtime. (#2544) - Breaking Change: remove deprecated
operator-sdk generate openapi
subcommand. (#2740)
- The Ansible Operator proxy server now properly supports the Pod
exec
API (#2716) - Resources that use '-' in the APIGroup name can now be directly accessed by Ansible. (#2712)
- Fixed issue in CSV generation that caused an incorrect path to be generated for descriptors on types that are fields in array elements. (#2721)
- Add a new option to set the minimum log level that triggers stack trace generation in logs (
--zap-stacktrace-level
) (#2319) - Added
pkg/status
with several new types and interfaces that can be used inStatus
structs to simplify handling of status conditions. (#1143) - Added support for relative Ansible roles and playbooks paths in the Ansible-based operator watches files. (#2273)
- Added watches file support for roles that were installed as Ansible collections. (#2587)
- Add Prometheus metrics support to Ansible-based operators. (#2179)
- On
generate csv
, populate a CSV manifest’sspec.icon
,spec.keywords
, andspec.mantainers
fields with empty values to better inform users how to add data. (#2521) - Scaffold code in
cmd/manager/main.go
for Go operators and add logic to Ansible/Helm operators to handle multinamespace caching ifWATCH_NAMESPACE
contains multiple namespaces. (#2522) - Add a new flag option (
--skip-cleanup-error
) to the test framework to allow skip the function which will remove all artefacts when an error be faced to perform this operation. (#2512) - Add event stats output to the operator logs for Ansible based-operators. (2580)
- Improve Ansible logs by allowing output the full Ansible result for Ansible based-operators configurable by environment variable. (2589)
- Add the --max-workers flag to the commands operator-sdk exec-entrypoint and operator-sdk run --local for Helm based-operators with the purpose of controling the number of concurrent reconcile workers. (2607)
- Add the --proxy-port flag to the operator-sdk scorecard command allowing users to override the default proxy port value (8889). (2634)
- Add support for Metrics with MultiNamespace scenario. (#2603)
- Add Prometheus metrics support to Helm-based operators. (#2603)
- Add a new flag option
--olm-namespace
tooperator-sdk run --olm
,operator-sdk cleanup --olm
andoperator-sdk olm status
command, which allows specifying the namespace in which OLM is installed. (#2613)
- The base image now includes version 0.10.3 of the OpenShift Python client. This should fix hanging in Python3
- The Kubernetes modules have migrated to the Kubernetes Ansible collection. All scaffolded code now references modules from this collection instead of Ansible Core. No immediate action is required for existing users of the modules from core, though it is recommended they switch to using the collection to continue to get non-critical bugfixes and features. To install the collection, users will need to add the install step to their
build/Dockerfile
. New projects will have arequirements.yml
scaffolded that includes thecommunity.kubernetes
collection, as well as the corresponding install step in thebuild/Dockerfile
. (#2646) - Breaking change
The operator_sdk.util
collection is no longer installed by default in the base image. Existing projects will need to install it in thebuild/Dockerfile
. New projects will have arequirements.yml
scaffolded that includes theoperator_sdk.util
collection, as well as the corresponding install step in thebuild/Dockerfile
. (#2652) - Ansible scaffolding has been rewritten to be simpler and make use of newer features of Ansible and Molecule. (#2425)
- No longer generates the build/test-framework directory or molecule/test-cluster scenario
- Adds new
cluster
scenario that can be used to test against an existing cluster - There is no longer any Ansible templating done in the
deploy/
directory, any templates used for testing will be located inmolecule/templates/
instead. - The scaffolded molecule.yml files now use the Ansible verifier. All asserts.yml files were renamed to verify.yml to reflect this.
- The prepare/converge/verify tasks now make use of the new
k8s
wait
option to simplify the deployment logic.
- Operator user setup and entrypoint scripts no longer insert dynamic runtime user entries into
/etc/passwd
. To use dynamic runtime users, use a container runtime that supports it (e.g. CRI-O). (#2469) - Changed the scorecard basic test,
Writing into CRs has an effect
, to include the http.MethodPatch as part of its test criteria alongside http.MethodPut and http.MethodPost. (#2509) - Changed the scorecard to use the init-timeout configuration setting as a wait time when performing cleanup instead of a hard-coded time. (#2597)
- Upgrade the Helm dependency version from
v3.0.1
tov3.0.2
. (#2621) - Changed the scaffolded
serveCRMetrics
to use the namespaces informed in the environment variableWATCH_NAMESPACE
in the MultiNamespace scenario. (#2603) - Improve skip metrics logs when running the operator locally in order to make clear the information for Helm based operators. (#2603)
-
The type name
TestCtx
inpkg/test
has been deprecated and renamed toContext
. It now exists only as a type alias to maintain backwards compatibility. Users of the e2e framework should migrate to use the new name,Context
. TheTestCtx
alias will be removed in a future version. (2549) -
The additional of the dependency
inotify-tools
on Ansible based-operator images. (#2586) -
Breaking Change: The scorecard feature now only supports YAML config files. So, any config file with other extension is deprecated and should be changed for the YAML format. For further information see
scorecard config file
(#2591)
- Breaking Change: The additional Ansible sidecar container. (#2586)
- Fixed issue with Go dependencies caused by removed tag in
openshift/api
repository (#2466) - Fixed a regression in the
operator-sdk run
command that caused--local
flags to be ignored (#2478) - Fix command
operator-sdk run --local
which was not working on Windows. (#2481) - Fix
ServiceMonitor
creation when the operator is cluster-scoped and the environment variableWATCH_NAMESPACE
has a different value than the namespace where the operator is deployed. (#2601) - Fix error faced when the
ansible.operator-sdk/verbosity
annotation for Ansible based-operators is 0 or less. (#2651) - Fix missing error status when the error faced in the Ansible do not return an event status. (#2661)
- Operator user setup and entrypoint scripts no longer insert dynamic runtime user entries into
/etc/passwd
. To use dynamic runtime users, use a container runtime that supports it (e.g. CRI-O). (#2469)
- Fixed a regression in the
operator-sdk run
command that caused--local
flags to be ignored (#2478)
- Fixed issue with Go dependencies caused by removed tag in
openshift/api
repository (#2466)
- Added the
cleanup
subcommand andrun --olm
to manage deployment/deletion of operators. These commands currently interact with OLM via an in-cluster registry-server created using an operator's on-disk manifests and managed byoperator-sdk
. (#2402, #2441) - Added
bundle create
which builds, and optionally generates metadata for, operator bundle images. (#2076, #2438) - Added
bundle validate
which validates operator bundle images. (#2411) - Added
blacklist
field to thewatches.yaml
for Ansible based operators. Blacklisted secondary resources will not be watched or cached.(#2374)
- Changed error wrapping according to Go version 1.13+ error handling. (#2355)
- Added retry logic to the cleanup function from the e2e test framework in order to allow it to be achieved in the scenarios where temporary network issues are faced. (#2277)
- Breaking Change: Moved
olm-catalog gen-csv
to thegenerate csv
subcommand. (#2439) - Breaking Change:
run ansible/helm
are now the hidden commandsexec-entrypoint ansible/helm
. All functionality of each subcommand is the same. (#2441) - Breaking Change:
up local
is nowrun --local
. All functionality of this command is the same. (#2441) - Breaking Change: Moved the
olm
subcommand fromalpha
to its own subcommand. All functionality of this command is the same. (#2447)
- Fixed a regression in the helm-operator that caused all releases to be deployed in the same namespace that the operator was deployed in, regardless of which namespace the CR was created in. Now release resources are created in the same namespace as the CR. (#2414)
- Fix issue when the test-framework would attempt to create a namespace exceeding 63 characters.
pkg/test/NewCtx()
now creates a unique id instead of using the test name.TestCtx.GetNamespace()
uses this unique id to create a namespace that avoids this scenario. (#2335)
- Fixed a regression in the helm-operator that caused all releases to be deployed in the same namespace that the operator was deployed in, regardless of which namespace the CR was created in. Now release resources are created in the same namespace as the CR. (#2414)
- Added new
--bundle
flag to theoperator-sdk scorecard
command to support bundle validation testing using the validation API (https://github.com/operator-framework/api). (#1916 - Added new
log
field to theoperator-sdk scorecard
v1alpha2 output to support tests that produce logging. (#1916 - Added new
bundle validation
test to theoperator-sdk scorecard
OLM tests. (#1916 - Added scorecard test short names to each scorecard test to allow users to run a specific scorecard test using the selector flag. (#1916
- Improve Ansible logs in the Operator container for Ansible-based Operators. (#2321)
- Added support for override values with environment variable expansion in the
watches.yaml
file for Helm-based operators. (#2325)
- Replace usage of
github.com/operator-framework/operator-sdk/pkg/restmapper.DynamicRESTMapper
withsigs.k8s.io/controller-runtime/pkg/client/apiutil.DynamicRESTMapper
. (#2309) - Upgraded Helm operator packages and base image from Helm v2 to Helm v3. Cluster state for pre-existing CRs using Helm v2-based operators will be automatically migrated to Helm v3's new release storage format, and existing releases may be upgraded due to changes in Helm v3's label injection. (#2080)
- Fail
operator-sdk olm-catalog gen-csv
if it is not run from a project's root, which the command already assumes is the case. (#2322) - Breaking Change: Extract custom Ansible module
k8s_status
, which is now provided by theoperator_sdk.util
Ansible collection. See developer_guide for new usage. (#2310) - Upgrade minimal Ansible version in the init projects from
2.6
to2.9
for collections support. (#2310) - Improve skip metrics logs when running the operator locally in order to make clear the information. (#2190)
- Upgrade
controller-tools
version fromv0.2.2
tov0.2.4
. (#2368)
- Deprecated
github.com/operator-framework/operator-sdk/pkg/restmapper
in favor of theDynamicRESTMapper
implementation in controller-runtime. (#2309)
- Fix
operator-sdk build
's--image-build-args
to support spaces within quotes like--label some.name="First Last"
. (#2312) - Fix misleading Helm operator "release not found" errors during CR deletion. (#2359)
- Fix Ansible based image in order to re-trigger reconcile when playbooks are runner with error. (#2375)
- Support for vars in top level ansible watches. (#2147)
- Support for
"ansible.operator-sdk/verbosity"
annotation on Custom Resources watched by Ansible based operators to override verbosity on an individual resource. (#2102) - Support for relative helm chart paths in the Helm operator's watches.yaml file. (#2287)
- New
operator-sdk generate crds
subcommand, which generates CRDs from Go types. (#2276) - Go API code can now be annotated to populate a CSV's
spec.customresourcedefinitions.owned
field on invokingolm-catalog gen-csv
. (#1162)
- Upgrade minimal Ansible version in the init projects from
2.4
to2.6
. (#2107) - Upgrade Kubernetes version from
kubernetes-1.15.4
tokubernetes-1.16.2
. (#2145) - Upgrade Helm version from
v2.15.0
tov2.16.1
. (#2145) - Upgrade
controller-runtime
version fromv0.3.0
tov0.4.0
. (#2145) - Updated
pkg/test/e2eutil.WaitForDeployment()
andpkg/test/e2eutil.WaitForOperatorDeployment()
to successfully complete waiting when the available replica count is at least (rather than exactly) the minimum replica count required. (#2248) - Replace in the Ansible based operators module tests
k8s_info
fork8s_facts
which is deprecated. (#2168) - Upgrade the Ansible version from
2.8
to2.9
on the Ansible based operators image. (#2168) - Updated CRD generation for non-Go operators to use valid structural schema. (#2275)
- Replace Role verb
"*"
with list of verb strings in generated files so the Role is compatible with OpenShift and Kubernetes. (#2175) - Breaking change: An existing CSV's
spec.customresourcedefinitions.owned
is now always overwritten except for eachname
,version
, andkind
on invokingolm-catalog gen-csv
when Go API code annotations are present. (#1162) - Ansible and Helm operator reconcilers use a cached client for reads instead of the default unstructured client. (#1047)
- Deprecated the
operator-sdk generate openapi
command. CRD generation is still supported withoperator-sdk generate crds
. It is now recommended to use openapi-gen directly for OpenAPI code generation. Thegenerate openapi
subcommand will be removed in a future release. (#2276)
- Fixed log formatting issue that occurred while loading the configuration for Ansible-based operators. (#2246)
- Fix issue faced in the Ansible based operators when
jmespath
queries are used because it was not installed. (#2252) - Updates
operator-sdk build
for go operators to compile the operator binary based on Go's built-in GOARCH detection. This fixes an issue that caused anamd64
binary to be built into non-amd64
base images when using operator-sdk on non-amd64
architectures. (#2268) - Fix scorecard behavior such that a CSV file is read correctly when
olm-deployed
is set totrue
. (#2274) - A CSV config's
operator-name
field will be used if--operator-name
is not set. (#2297) - Populates a CSV's
spec.install
strategy if either name or strategy body are missing with a deployment-type strategy. (#2298) - When the current leader pod has been hard evicted but not deleted, another pod is able to delete the evicted pod, triggering garbage collection and allowing leader election to continue. (#2210)
- Added
Operator Version: X.Y.Z
information in the operator logs.(#1953) - Make Ansible verbosity configurable via the
ansible-verbosity
flag. (#2087) - Autogenerate CLI documentation via
make cli-doc
(#2099)
- Breaking change: Changed required Go version from
1.12
to1.13
. This change applies to the SDK project itself and Go projects scaffolded by the SDK. Projects that import this version of the SDK require Go 1.13 to compile. (#1949) - Upgrade Kubernetes version from
kubernetes-1.14.1
tokubernetes-1.15.4
. (#2083) - Upgrade Helm version from
v2.14.1
tov2.15.0
. (#2083) - Upgrade
controller-runtime
version fromv0.2.0
tov0.3.0
. (#2083) - Upgrade
controller-tools
version fromv0.2.1+git
tov0.2.2
. (#2083)
- Removed
--dep-manager
flag and support fordep
-based projects. Projects will be scaffolded to use Go modules. (#1949)
- OLM internal manager is not returning errors in the initialization. (#1976)
- Added missing default role permission for
deployments
, which is required to create the metrics service for the operator. (#2090) - Handle invalid maxArtifacts annotation on CRs for Ansible based operators. (2093)
- When validating package manifests, only return an error if default channel is not set and more than one channel is available. (#2116)
- Added support for event filtering for ansible operator. (#1968)
- Added new
--skip-generation
flag to theoperator-sdk add api
command to support skipping generation of deepcopy and OpenAPI code and OpenAPI CRD specs. (#1890) - The
operator-sdk olm-catalog gen-csv
command now produces indented JSON for thealm-examples
annotation. (#1793) - Added flag
--dep-manager
to commandoperator-sdk print-deps
to specify the type of dependency manager file to print. The choice of dependency manager is inferred from top-level dependency manager files present if--dep-manager
is not set. (#1819) - Ansible based operators now gather and serve metrics about each custom resource on port 8686 of the metrics service. (#1723)
- Added the Go version, OS, and architecture to the output of
operator-sdk version
(#1863) - Added support for
ppc64le-linux
for theoperator-sdk
binary and the Helm operator base image. (#1533) - Added new
--version
flag to theoperator-sdk scorecard
command to support a new output format for the scorecard. (#1916 - Added new
--selector
flag to theoperator-sdk scorecard
command to support filtering scorecard tests based on labels added to each test. (#1916 - Added new
--list
flag to theoperator-sdk scorecard
command to support listing scorecard tests that would be executed based on selector filters. (#1916 - For scorecard version v1alpha2 only, return code logic was added to return 1 if any of the selected scorecard tests fail. A return code of 0 is returned if all selected tests pass. (#1916
- The Helm operator now uses the CR name for the release name for newly created CRs. Existing CRs will continue to use their existing UID-based release name. When a release name collision occurs (when CRs of different types share the same name), the second CR will fail to install with an error about a duplicate name. (#1818)
- Commands
olm uninstall
andolm status
no longer use a--version
flag to specify OLM version. This information is now retrieved from the running cluster. (#1634) - The Helm operator no longer prints manifest diffs in the operator log at verbosity levels lower than INFO (#1857)
- CRD manifest
spec.version
is still supported, but users will see a warning message ifspec.versions
is not present and an error ifspec.versions
is populated but the version inspec.version
is not inspec.versions
. (#1876) - Upgrade base image for Go, Helm, and scorecard proxy from
registry.access.redhat.com/ubi7/ubi-minimal:latest
toregistry.access.redhat.com/ubi8/ubi-minimal:latest
. (#1952) - Upgrade base image for Ansible from
registry.access.redhat.com/ubi7/ubi:latest
toregistry.access.redhat.com/ubi8/ubi:latest
. (#1990 and #2004) - Updated kube-state-metrics dependency from
v1.6.0
tov1.7.2
. (#1943)
- Upgrade Kubernetes version from
kubernetes-1.13.4
tokubernetes-1.14.1
(#1876) - Upgrade
github.com/operator-framework/operator-lifecycle-manager
version fromb8a4faf68e36feb6d99a6aec623b405e587b17b1
to0.10.1
(#1876) - Upgrade
controller-runtime
version fromv0.1.12
tov0.2.0
(#1876)- The package
sigs.k8s.io/controller-runtime/pkg/runtime/scheme
is deprecated, and contains no code. Replace this import withsigs.k8s.io/controller-runtime/pkg/scheme
where relevant. - The package
sigs.k8s.io/controller-runtime/pkg/runtime/log
is deprecated. Replace this import withsigs.k8s.io/controller-runtime/pkg/log
where relevant. - The package
sigs.k8s.io/controller-runtime/pkg/runtime/signals
is deprecated. Replace this import withsigs.k8s.io/controller-runtime/pkg/manager/signals
where relevant. - All methods on
sigs.k8s.io/controller-runtime/pkg/client.Client
(except forGet()
) have been updated. Instead of each using astruct
-typed or variadic functional option parameter, or having no option parameter, each now uses a variadic interface option parameter typed for each method. SeeList()
below for an example. sigs.k8s.io/controller-runtime/pkg/client.Client
'sList()
method signature has been updated:List(ctx context.Context, opts *client.ListOptions, list runtime.Object) error
is nowList(ctx context.Context, list runtime.Object, opts ...client.ListOption) error
. To migrate:import ( "context" "sigs.k8s.io/controller-runtime/pkg/client" ) ... // Old listOpts := &client.ListOptions{} listOpts.InNamespace("namespace") err = r.client.List(context.TODO(), listOps, podList) // New listOpts := []client.ListOption{ client.InNamespace("namespace"), } err = r.client.List(context.TODO(), podList, listOpts...)
- The package
pkg/test.FrameworkClient
methodsList()
andDelete()
have new signatures corresponding to the homonymous methods ofsigs.k8s.io/controller-runtime/pkg/client.Client
. (#1876)- CRD file names were previously of the form
<group>_<version>_<kind>_crd.yaml
. Now that CRD manifestspec.version
is deprecated in favor ofspec.versions
, i.e. multiple versions can be specified in one CRD, CRD file names have the form<full group>_<resource>_crd.yaml
.<full group>
is the full group name of your CRD while<group>
is the last subdomain of<full group>
, ex.foo.bar.com
vsfoo
.<resource>
is the plural lower-case CRD Kind found atspec.names.plural
. (#1876) - Upgrade Python version from
2.7
to3.6
, Ansible version from2.8.0
to~=2.8
and ansible-runner from1.2
to1.3.4
in the Ansible based images. (#1947) - Made the default scorecard version
v1alpha2
which is new for this release and could break users that were parsing the older scorecard output (v1alpha1
). Users can still specify versionv1alpha1
on the scorecard configuration to use the older style for some period of time untilv1alpha1
is removed. - Replaced
pkg/kube-metrics.NewCollectors()
withpkg/kube-metrics.NewMetricsStores()
and changed exported function signature forpkg/kube-metrics.ServeMetrics()
due to a breaking change in kube-state-metrics. (#1943)
- Removed flag
--as-file
from commandoperator-sdk print-deps
, which now only prints packages and versions in dependency manager file format. The choice of dependency manager type is set by--dep-manager
or inferred from top-level dependency manager files present if--dep-manager
is not set. (#1819)
- Configure the repo path correctly in
operator-sdk add crd
and prevent the command from running outside of an operator project. (#1660) - In the Helm operator, skip owner reference injection for cluster-scoped resources in release manifests. The Helm operator only supports namespace-scoped CRs, and namespaced resources cannot own cluster-scoped resources. (#1817)
- Package manifests generated with
gen-csv
respect the--operator-name
flag, channel names are checked for duplicates before (re-)generation. (#1693) - Generated inventory for Ansible-based Operators now sets the localhost's
ansible_python_interpreter
to{{ ansible_playbook_python }}
, to properly match the implicit localhost. (#1952) - Fixed an issue in
operator-sdk olm-catalog gen-csv
where the generated CSV is missing the expected set of owned CRDs. (#2017) - The command
operator-sdk olm-catalog gen-csv --csv-version=<version> --update-crds
would fail to copy over CRD manifests intodeploy/olm-catalog
for manifests whose name didn't end with a_crd.yaml
suffix. This has been fixed sogen-csv
now copies all CRD manifests specified bydeploy/olm-catalog/csv_config.yaml
by checking the type of the manifest rather than the filename suffix. (#2015) - Added missing
jmespath
dependency to Ansible-based Operator .travis.yml file template. (#2027) - Fixed invalid usage of
logr.Logger.Info()
in the Ansible-based operator implementation, which caused unnecessary operator panics. (#2031)
- Document new compile-time dependency
mercurial
in user-facing documentation. (#1683) - Adds new flag
--zap-time-encoding
to the flagset provided bypkg/log/zap
. This flag configures the timestamp format produced by the zap logger. See the logging doc for more information. (#1529)
- Breaking Change: New configuration format for the
operator-sdk scorecard
using config files. Seedoc/test-framework/scorecard
for more info (#1641) - Breaking change: CSV config field
role-path
is nowrole-paths
and takes a list of strings. Users can now specify multipleRole
andClusterRole
manifests usingrole-paths
. (#1704) - Make
ready
package idempotent. Now, a user can callSet()
orUnset()
to set the operator's readiness without knowing the current state. (#1761)
- Check if
metadata.annotations['alm-examples']
is non-empty before creating contained CR manifests in the scorecard. (#1789)
- Adds support for building OCI images with podman, e.g.
operator-sdk build --image-builder=podman
. (#1488) - New option for
operator-sdk up local --enable-delve
, which can be used to start the operator in remote debug mode with the delve debugger listening on port 2345. (#1422) - Enables controller-runtime metrics in Helm operator projects. (#1482)
- New flags
--vendor
and--skip-validation
foroperator-sdk new
that direct the SDK to initialize a new project with avendor/
directory, and without validating project dependencies.vendor/
is not written by default. (#1519) - Generating and serving info metrics about each custom resource. By default these metrics are exposed on port 8686. (#1277)
- Scaffold a
pkg/apis/<group>/group.go
package file to avoidgo/build
errors when running Kubernetes code generators. (#1401) - Adds a new extra variable containing the unmodified CR spec for ansible based operators. #1563
- New flag
--repo
for subcommandsnew
andmigrate
specifies the repository path to be used in Go source files generated by the SDK. This flag can only be used with Go modules. (#1475) - Adds
--go-build-args
flag tooperator-sdk build
for providing additional Go build arguments. (#1582) - New flags
--csv-channel
and--default-channel
for subcommandgen-csv
that add channels to and update the package manifest indeploy/olm-catalog/<operator-name>
when generating a new CSV or updating an existing one. (#1364) - Adds
go.mod
andgo.sum
to switch fromdep
to Go modules to manage dependencies for the SDK project itself. (#1566) - New flag
--operator-name
foroperator-sdk olm-catalog gen-csv
to specify the operator name, ex.memcached-operator
, to use in CSV generation. The project's name is used (old behavior) if--operator-name
is not set. (#1571) - New flag
--local-operator-flags
foroperator-sdk test local --up-local
to specify flags to run a local operator with during a test. (#1509)
- Upgrade the version of the dependency controller-runtime from
v0.1.10
tov0.1.12
. (#1612) - Remove TypeMeta declaration from the implementation of the objects (#1462)
- Relaxed API version format check when parsing
pkg/apis
in code generators. API dir structures can now be of the formatpkg/apis/<group>/<anything>
, where<anything>
was previously required to be in the Kubernetes version format, ex.v1alpha1
. (#1525) - The SDK and operator projects will work outside of
$GOPATH/src
when using Go modules. (#1475) CreateMetricsService()
function from the metrics package accepts a REST config (*rest.Config) and an array of ServicePort objects ([]v1.ServicePort) as input to create Service metrics.CRPortName
constant is added to describe the string of custom resource port name. (#1560 and #1626)- Changed the flag
--skip-git-init
to--git-init
. This changes the default behavior ofoperator-sdk new
to not initialize the new project directory as a git repository withgit init
. This behavior is now opt-in with--git-init
. (#1588) operator-sdk new
will no longer create the initial commit for a new project, even with--git-init=true
. (#1588)- When errors occur setting up the Kubernetes client for RBAC role generation,
operator-sdk new --type=helm
now falls back to a default RBAC role instead of failing. (#1627)
- The SDK no longer depends on a
vendor/
directory to manage dependencies only if using Go modules. The SDK and operator projects will only use vendoring if usingdep
, or modules and avendor/
dir is present. (#1519) - Breaking change:
ExposeMetricsPort
is removed and replaced withCreateMetricsService()
function.PrometheusPortName
constant is replaced withOperatorPortName
. (#1560) - Removes
Gopkg.toml
andGopkg.lock
to drop the use ofdep
in favor of Go modules to manage dependencies for the SDK project itself. (#1566)
- Fixes header file content validation when the content contains empty lines or centered text. (#1544)
- Generated CSV's that include a deployment install strategy will be checked for a reference to
metadata.annotations['olm.targetNamespaces']
, and if one is not found a reference will be added to theWATCH_NAMESPACE
env var for all containers in the deployment. This is a bug because any other value that references the CSV's namespace is incorrect. (#1396) - Build
-trimpath
was not being respected.$GOPATH
was not expanding becauseexec.Cmd{}
is not executed in a shell environment. (#1535) - Running the scorecard with
--olm-deployed
will now only use the first CR set in either thecr-manifest
config option or the CSV'smetadata.annotations['alm-examples']
as was intended, and access manifests correctly from the config. (#1565) - Use the correct domain names when generating CRD's instead that of the first CRD to be parsed. (#1636)
- Fixes a regression that causes Helm RBAC generation to contain an empty custom ruleset when the chart's default manifest contains only namespaced resources. (#1456)
- Fixes an issue that causes Helm RBAC generation to fail when creating new operators with a Kubernetes context configured to connect to an OpenShift cluster. (#1461)
- New option for
operator-sdk build --image-builder
, which can be used to specify which image builder to use. Adds support for buildah. (#1311) - Manager is now configured with a new
DynamicRESTMapper
, which accounts for the fact that the defaultRESTMapper
, which only checks resource types at startup, can't handle the case of first creating a CRD and then an instance of that CRD. (#1329) - Unify CLI debug logging under a global
--verbose
flag (#1361) - Go module support by default for new Go operators and during Ansible and Helm operator migration. The dependency manager used for a new operator can be explicitly specified for new operators through the
--dep-manager
flag, available inoperator-sdk new
andoperator-sdk migrate
.dep
is still available through--dep-manager=dep
. (#1001) - New optional flag
--custom-api-import
foroperator-sdk add controller
to specify that the new controller reconciles a built-in or external Kubernetes API, and what import path and identifier it should have. (#1344) - Operator Scorecard plugin support (#1379). Documentation for scorecard plugins can be found in the main scorecard doc: doc/test-framework/scorecard.md
- When Helm operator projects are created, the SDK now generates RBAC rules in
deploy/role.yaml
based on the chart's default manifest. (#1188) - When debug level is 3 or higher, we will set the klog verbosity to that level. (#1322)
- Relaxed requirements for groups in new project API's. Groups passed to
operator-sdk add api
's--api-version
flag can now have no subdomains, excore/v1
. See (#1191) for discussion. (#1313) - Renamed
--docker-build-args
option to--image-build-args
option forbuild
subcommand, because this option can now be shared with other image build tools than docker when--image-builder
option is specified. (#1311) - Reduces Helm release information in CR status to only the release name and manifest and moves it from
status.conditions
to a new top-leveldeployedRelease
field. (#1309)- WARNING: Users with active CRs and releases who are upgrading their helm-based operator should upgrade to one based on v0.7.0 before upgrading further. Helm operators based on v0.8.0+ will not seamlessly transition release state to the persistent backend, and will instead uninstall and reinstall all managed releases.
- Go operator CRDs are overwritten when being regenerated by
operator-sdk generate openapi
. Users can now rely on+kubebuilder
annotations in their API code, which provide access to most OpenAPIv3 validation properties (the full set will be supported in the near future, see this PR) and other CRD fields. (#1278) - Use
registry.access.redhat.com/ubi7/ubi-minimal:latest
base image for the Go and Helm operators and scorecard proxy (#1376) - Allow "Owned CRDs Have Resources Listed" scorecard test to pass if the resources section exists
- The SDK will no longer run
defaulter-gen
on runningoperator-sdk generate k8s
. Defaulting for CRDs should be handled with mutating admission webhooks. (#1288) - The
--version
flag was removed. Users should use theoperator-sdk version
command. (#1444) - Breaking Change: The
test cluster
subcommand and the corresponding--enable-tests
flag for thebuild
subcommand have been removed (#1414) - Breaking Change: The
--cluster-scoped
flag foroperator-sdk new
has been removed so it won't scaffold a cluster-scoped operator. Read the operator scope documentation on the changes needed to run a cluster-scoped operator. (#1434)
operator-sdk generate openapi
no longer overwrites CRD values derived from+kubebuilder
annotations in Go API code. See issues (#1212) and (#1323) for discussion. (#1278)- Running
operator-sdk gen-csv
on operators that do not have a CRDs directory, ex.deploy/crds
, or do not have any owned CRDs, will not generate a "deploy/crds not found" error.
- Pin dependency versions in Ansible build and test framework Dockerfiles to fix broken build and test framework images. (#1348)
- In Helm-based operators, when a custom resource with a failing release is reverted back to a working state, the
ReleaseFailed
condition is now correctly removed. (#1321)
- New optional flag
--header-file
for commandsoperator-sdk generate k8s
andoperator-sdk add api
to supply a boilerplate header file for generated code. (#1239) - JSON output support for
operator-sdk scorecard
subcommand (#1228)
- Updated the helm-operator to store release state in kubernetes secrets in the same namespace of the custom resource that defines the release. (#1102)
- WARNING: Users with active CRs and releases who are upgrading their helm-based operator should not skip this version. Future versions will not seamlessly transition release state to the persistent backend, and will instead uninstall and reinstall all managed releases.
- Change
namespace-manifest
flag in scorecard subcommand tonamespaced-manifest
to match other subcommands - Subcommands of
operator-sdk generate
are now verbose by default. (#1271) operator-sdk olm-catalog gen-csv
parses Custom Resource manifests fromdeploy/crds
or a custom path specified incsv-config.yaml
, encodes them in a JSON array, and sets the CSV'smetadata.annotations.alm-examples
field to that JSON. (#1116)
- Fixed an issue that caused
operator-sdk new --type=helm
to fail for charts that have template files in nested template directories. (#1235) - Fix bug in the YAML scanner used by
operator-sdk test
andoperator-sdk scorecard
that could result in a panic if a manifest file started with---
(#1258)
- New flags for
operator-sdk new --type=helm
, which can be used to populate the project with an existing chart. (#949) - Command
operator-sdk olm-catalog
flag--update-crds
optionally copies CRD's fromdeploy/crds
when creating a new CSV or updating an existing CSV, and--from-version
uses another versioned CSV manifest as a base for a new CSV version. (#1016) - New flag
--olm-deployed
to direct thescorecard
command to only use the CSV at--csv-path
for manifest data, except for those provided to--cr-manifest
. (#1044) - Command
version
prints the version of operator-sdk. (#1171)
- Changed the Go, Helm, and Scorecard base images to
registry.access.redhat.com/ubi7-dev-preview/ubi-minimal:7.6
(#1142) - CSV manifest are now versioned according to the
operator-registry
manifest format. See issue #900 for more details. (#1016) - Unexported
CleanupNoT
function frompkg/test
, as it is only intended to be used internally (#1167)
- Fix issue where running
operator-sdk test local --up-local
would sometimes leave a running process in the background after exit (#1089)
- Updated the Kubernetes dependencies to
1.13.1
(#1020) - Updated the controller-runtime version to
v0.1.10
. See the controller-runtimev0.1.10
release notes for new features and bug fixes. (#1020) - By default the controller-runtime metrics are exposed on port 8383. This is done as part of the scaffold in the main.go file, the port can be adjusted by modifying the
metricsPort
variable. #786 - A new command
operator-sdk olm-catalog
to be used as a parent for SDK subcommands generating code related to Operator Lifecycle Manager (OLM) Catalog integration, and subcommandoperator-sdk olm-catalog gen-csv
which generates a Cluster Service Version for an operator so the OLM can deploy the operator in a cluster. (#673) - Helm-based operators have leader election turned on by default. When upgrading, add environment variable
POD_NAME
to your operator's Deployment using the Kubernetes downward API. To see an example, runoperator-sdk new --type=helm ...
and see filedeploy/operator.yaml
. #1000 - A new command
operator-sdk generate openapi
which generates OpenAPIv3 validation specs in Go and in CRD manifests as YAML. (#869) - The
operator-sdk add api
command now generates OpenAPIv3 validation specs in Go for that API, and in all CRD manifests as YAML.
- In new Helm operator projects, the scaffolded CR
spec
field now contains the default values.yaml from the generated chart. (#967)
- Make
up local
subcommand respectKUBECONFIG
env var (#996) - Make
up local
subcommand use default namespace set in kubeconfig instead of hardcodeddefault
and also add ability to watch all namespaces for ansible and helm type operators (#996) - Added k8s_status modules back to generation (#972)
- Update checks for gvk registration to cover all cases for ansible (#973 & #1019)
- Update reconciler for ansible and helm to use the cache rather than the API client. (#1022 & #1048 & #1054)
- Update reconciler to will update the status everytime for ansible (#1066)
- Update ansible proxy to recover dependent watches when pod is killed (#1067)
- Update ansible proxy to handle watching cluster scoped dependent watches (#1031)
- A new command
operator-sdk migrate
which adds a main.go source file and any associated source files for an operator that is not of the "go" type. (#887 and #897) - New commands
operator-sdk run ansible
andoperator-sdk run helm
which run the SDK as ansible and helm operator processes, respectively. These are intended to be used when running in a Pod inside a cluster. Developers wanting to run their operator locally should continue to useup local
. (#887 and #897) - Ansible operator proxy added the cache handler which allows the get requests to use the operators cache. #760
- Ansible operator proxy added ability to dynamically watch dependent resource that were created by ansible operator. #857
- Ansible-based operators have leader election turned on by default. When upgrading, add environment variable
POD_NAME
to your operator's Deployment using the Kubernetes downward API. To see an example, runoperator-sdk new --type=ansible ...
and see filedeploy/operator.yaml
. - A new command
operator-sdk scorecard
which runs a series of generic tests on operators to ensure that an operator follows best practices. For more information, see theScorecard Documentation
- The official images for the Ansible and Helm operators have moved! Travis now builds, tags, and pushes operator base images during CI (#832).
- Fixes deadlocks during operator deployment rollouts, which were caused by operator pods requiring a leader election lock to become ready (#932)
- Helm type operator generation support (#776)
- The SDK's Kubernetes Golang dependency versions/revisions have been updated from
v1.11.2
tov1.12.3
. (#807) - The controller-runtime version has been updated from
v0.1.4
tov0.1.8
. See thev0.1.8
release notes for details. - The SDK now generates the CRD with the status subresource enabled by default. See the client doc on how to update the status subresource. (#787)
- Pin controller-runtime version to v0.1.4 to fix dependency issues and pin ansible idna package to version 2.7 (#831)
- The SDK now uses logr as the default logger to unify the logging output with the controller-runtime logs. Users can still use a logger of their own choice. See the logging doc on how the SDK initializes and uses logr.
- Ansible Operator CR status better aligns with conventions. (#639)
- A new command
operator-sdk print-deps
which prints Golang packages and versions expected by the current Operator SDK version. Supplying--as-file
prints packages and versions in Gopkg.toml format. (#772) - Add
cluster-scoped
flag tooperator-sdk new
command (#747) - Add
up-local
flag totest local
subcommand (#781) - Add
no-setup
flag totest local
subcommand (#770) - Add
image
flag totest local
subcommand (#768) - Ansible Operator log output includes much more information for troubleshooting ansible errors. (#713)
- Ansible Operator periodic reconciliation can be disabled (#739)
- Make operator-sdk command work with composed GOPATH (#676)
- Ansible Operator "--kubeconfig" command line option fixed (#705)
- Fix hardcoded CRD version in crd scaffold (#690)
- Use controller runtime library for controller and client APIs
- See migration guide to migrate your project to
v0.1.0
- Service account generation (#454)
- Leader election (#530)
- Incluster test support for test framework (#469)
- Ansible type operator generation support (#486, #559)
- Moved the rendering of
deploy/operator.yaml
to theoperator-sdk new
command instead ofoperator-sdk build
- Added
operator-sdk up
command to help deploy an operator. Currently supports running an operator locally against an existing cluster e.goperator-sdk up local --kubeconfig=<path-to-kubeconfig> --namespace=<operator-namespace>
. Seeoperator-sdk up -h
for help. #219 #274 - Added initial default metrics to be captured and exposed by Prometheus. #323 exposes the metrics port and #349 adds the initial default metrics.
- Added initial test framework for operators #377, #392, #393
- All the modules in
pkg/sdk
have been combined into a single package.action
,handler
,informer
types
andquery
pkgs have been consolidated intopkg/sdk
. #242 - The SDK exposes the Kubernetes clientset via
k8sclient.GetKubeClient()
#295 - The SDK now vendors the k8s code-generators for an operator instead of using the prebuilt image
gcr.io/coreos-k8s-scale-testing/codegen:1.9.3
#319 - The SDK exposes the Kubernetes rest config via
k8sclient.GetKubeConfig()
#338 - Use
time.Duration
instead ofint
forsdk.Watch
#427
- The cache of available clients is being reset every minute for discovery of newely added resources to a cluster. #280