fix: Ensure that insecureSkipTLSVerify is false before setting caBundle on APIService #160

JorTurFer · 2023-11-21T11:50:52Z

If the APIService has already registered the value insecureSkipTLSVerify: true, cert-controller fails patching the APIService because caBundle and insecureSkipTLSVerify: true aren't compatible.
This PR updates the behaviour to ensure that insecureSkipTLSVerify is false before setting the caBundle

…dle on APIService Signed-off-by: Jorge Turrado <jorge.turrado@scrm.lidl>

acpana · 2023-11-30T20:04:22Z

pkg/rotator/rotator.go

+	if err := unstructured.SetNestedField(apiService.Object, false, "spec", "insecureSkipTLSVerify"); err != nil {
+		return err
+	}


I'm curious to get some of the maintainers thoughts on whether this actually has to happen here, in cert-controller. If cert-controller is going to be used with APIService, presumably we are going to inject a caBundle so why not set this field outside of cert-controller to begin with ?

The problem we have faced with is that when you set insecureSkipTLSVerify: true , you have to set explicitly insecureSkipTLSVerify: false to remove the value.
As insecureSkipTLSVerify: false is the default value, k8s api removes it automatically and CD tools like ArgoCD/Flux enter on permanent out-of-sync status.
kedacore/charts#550

In this scenario, we have some options:

Adding a step to remove the field manually (breaking the helm upgrade process)

Adding explicitly the value in helm chart, producing out-of-sync on CD tools

Updating insecureSkipTLSVerify before setting caBundle

Maybe I've missed something and there is any other option

I'm a bit confused here. You're saying that having insecureSkipTLSVerify: false in the Helm chart causes CI/CD pipelines to detect drift because the API server strips default values, correct?

I'm curious what makes cert-controller different in this regard, since it's merely another thing essentially running kubectl apply. Would cert-controller also be subject to the same default-stripping issue, leaving you right back where you began?

Or is the objective to remove the state from the Helm chart such that there is no longer any check for drift for this value?

Modifying controller code to meet the needs of CI/CD code does seem a bit backwards. I am also wary of potentially manually overriding user intent on what could be a very salient field (is there some reason why insecureSkipTLSVerify is set? Will un-setting it cause an outage? Can we be sure the answer will be identical for all projects that consume cert-controller?)

As for other options, maybe a post-install hook on the Helm chart to modify the value?

Here is an example of a post-install hook G8r uses:

https://github.com/open-policy-agent/gatekeeper/blob/master/charts/gatekeeper/templates/namespace-post-install.yaml

I'm a bit confused here. You're saying that having insecureSkipTLSVerify: false in the Helm chart causes CI/CD pipelines to detect drift because the API server strips default values, correct?

Yes, as it's the default value, it is removed automatically. If you deploy an apiservice with explicitly set insecureSkipTLSVerify: false and then you get it from the cluster, you will see that your set value insecureSkipTLSVerify: false has been chopped from the manifest. This triggers out-of-sync on CD systems like ArgoCD.
You can just try with this:

apiVersion: apiregistration.k8s.io/v1 kind: APIService metadata: labels: app.kubernetes.io/name: v1beta1.external.metrics.k8s.io app.kubernetes.io/version: latest app.kubernetes.io/part-of: keda-operator name: v1beta1.custom.metrics.k8s.io # nosemgrep: yaml.kubernetes.security.skip-tls-verify-service.skip-tls-verify-service spec: service: name: keda-metrics-apiserver namespace: keda group: external.metrics.k8s.io version: v1beta1 groupPriorityMinimum: 100 versionPriority: 100 insecureSkipTLSVerify: false

I'm curious what makes cert-controller different in this regard, since it's merely another thing essentially running kubectl apply. Would cert-controller also be subject to the same default-stripping issue, leaving you right back where you began?

Yes, it's removed exactly in the same way, but if it's cert-controller, we don't need to set it in the helm chart, so external tools aren't affected.

Or is the objective to remove the state from the Helm chart such that there is no longer any check for drift for this value?

The objective is to not have to explicitly set the value to default value in general. (not only from helm but from raw manifests)

Modifying controller code to meet the needs of CI/CD code does seem a bit backwards. I am also wary of potentially manually overriding user intent on what could be a very salient field (is there some reason why insecureSkipTLSVerify is set? Will un-setting it cause an outage? Can we be sure the answer will be identical for all projects that consume cert-controller?)

Actually, if cert-controller tries to patch the apiservice to add the caBundle but the apiservice has set insecureSkipTLSVerify: true, cert-controller will fail because insecureSkipTLSVerify: true and caBundle are mutually exclusive. If the value has been set to true and then cert-controller has been enabled, cert-controller won't work at all, printing errors all the time due to this mutual exclusivity.

As for other options, maybe a post-install hook on the Helm chart to modify the value?

We don't need this feature anymore (nor for helm nor for raw yamls) as we just needed it during some releases after the migration from random certs generated by the pods to real certs, stored in Kubernetes and managed by cert-controller/cert-manager. I mean, we don't need this feature anymore for KEDA xD

I've just pushed it instead of just closing the PR because this project is awesome and has helped us a lot and I assume that we aren't the only one who faces with this and the code was already done.

I can extend any other doubt that you have, said that, if you think that this doesn't make sense, we can close the PR

Actually, if cert-controller tries to patch the apiservice to add the caBundle but the apiservice has set insecureSkipTLSVerify: true, cert-controller will fail because insecureSkipTLSVerify: true and caBundle are mutually exclusive. If the value has been set to true and then cert-controller has been enabled, cert-controller won't work at all, printing errors all the time due to this mutual exclusivity.

This makes sense, but if a user has intentionally skipped TLS verification, will cert-controller overriding it break their use case? If so, consumers would be unable to use that version of cert-controller due to the mandatory override. It comes down to whether its better for cert-controller to return errors or for it to override user intent without context as to why it was set.

I've just pushed it instead of just closing the PR because this project is awesome and has helped us a lot and I assume that we aren't the only one who faces with this and the code was already done.

I very much appreciate the consideration and am glad the project has helped you!

I can extend any other doubt that you have, said that, if you think that this doesn't make sense, we can close the PR

I think if the override was optional, such that projects could opt-in, I would be less concerned. This would probably boil down to adding a new option to the rotator:

cert-controller/pkg/rotator/rotator.go

Lines 219 to 259 in b1a35eb

// CertRotator contains cert artifacts and a channel to close when the certs are ready.

type CertRotator struct {

reader SyncingReader

writer client.Writer

SecretKey types.NamespacedName

CertDir string

CAName string

CAOrganization string

DNSName string

ExtraDNSNames []string

IsReady chan struct{}

Webhooks []WebhookInfo

// FieldOwner is the optional fieldmanager of the webhook updated fields.

FieldOwner string

RestartOnSecretRefresh bool

ExtKeyUsages *[]x509.ExtKeyUsage

// RequireLeaderElection should be set to true if the CertRotator needs to

// be run in the leader election mode.

RequireLeaderElection bool

// CaCertDuration sets how long a CA cert will be valid for.

CaCertDuration time.Duration

// ServerCertDuration sets how long a server cert will be valid for.

ServerCertDuration time.Duration

// RotationCheckFrequency sets how often the rotation is executed

RotationCheckFrequency time.Duration

// LookaheadInterval sets how long before the certificate is renewed

LookaheadInterval time.Duration

// CertName and Keyname override certificate path

CertName string

KeyName string

certsMounted chan struct{}

certsNotMounted chan struct{}

wasCAInjected *atomic.Bool

caNotInjected chan struct{}

// testNoBackgroundRotation doesn't actually start the rotator in the background.

// This should only be used for testing.

testNoBackgroundRotation bool

}

Which defaults to off so that behavior would be backwards compatible for users on upgrade.

Definitely up to you if the extra work is worth your time. I apologize for the slow review cycle.

Kindly reminder @maxsmythe 😄

I think we probably want to test both cases, but I hear you about not wanting to propagate the expected assertion everywhere.

Maybe we can use functional options to make this a backwards-compatible change to testWebhook().

Functional options blog: https://commandcenter.blogspot.com/2014/01/self-referential-functions-and-design.html

f**k! I missed your comment 🤦
I'll review the PR this week 😄

I've added another split test for this specific feature. PTAL when you have some time @maxsmythe

Kindly reminder @maxsmythe 😄

pkg/rotator/rotator_test.go

Signed-off-by: Jorge Turrado <jorge_turrado@hotmail.es>

codecov-commenter · 2023-11-30T23:36:40Z

Codecov Report

Attention: Patch coverage is 60.00000% with 4 lines in your changes are missing coverage. Please review.

Project coverage is 60.13%. Comparing base (190188d) to head (b6466d0).

Files	Patch %	Lines
pkg/rotator/rotator.go	60.00%	2 Missing and 2 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #160      +/-   ##
==========================================
+ Coverage   57.16%   60.13%   +2.97%     
==========================================
  Files           1        1              
  Lines         572      577       +5     
==========================================
+ Hits          327      347      +20     
+ Misses        181      170      -11     
+ Partials       64       60       -4

Flag	Coverage Δ
unittests	`60.13% <60.00%> (+2.97%)`	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

JorTurFer · 2023-12-12T08:50:59Z

Hello!
Any update about this?

JorTurFer · 2024-01-14T15:52:58Z

Hello @ritazh @maxsmythe
Sorry for linking you directly, but this PR hasn't got any answer for a month. Could you review it please?

ritazh

LGTM

JorTurFer · 2024-02-06T23:51:44Z

Hello ✋
Can this be merged?

JaydipGabani · 2024-02-07T01:28:13Z

@JorTurFer We are waiting on lgtm from @maxsmythe as well.

Signed-off-by: Jorge Turrado <jorge.turrado@scrm.lidl>

Signed-off-by: Jorge Turrado <jorge_turrado@hotmail.es>

JorTurFer · 2024-03-31T15:53:53Z

pkg/rotator/rotator.go

@@ -778,7 +788,7 @@ func (r *ReconcileWH) Reconcile(ctx context.Context, request reconcile.Request)
 		artifacts, err := buildArtifactsFromSecret(secret)
 		if err != nil {
 			crLog.Error(err, "secret is not well-formed, cannot update webhook configurations")
-			return reconcile.Result{}, nil
+			return reconcile.Result{}, err


I've found this error too. If the reconciliation has failed, we should reschedule it manually or just return the error for auto rescheduling it

feat: Ensure that insecureSkipTLSVerify is false before setting caBun…

711ab42

…dle on APIService Signed-off-by: Jorge Turrado <jorge.turrado@scrm.lidl>

JorTurFer changed the title ~~feat: Ensure that insecureSkipTLSVerify is false before setting caBundle on APIService~~ fix: Ensure that insecureSkipTLSVerify is false before setting caBundle on APIService Nov 21, 2023

JorTurFer mentioned this pull request Nov 21, 2023

remove not required insecureSkipTLSVerify kedacore/charts#564

Merged

4 tasks

zroubalik approved these changes Nov 27, 2023

View reviewed changes

acpana reviewed Nov 30, 2023

View reviewed changes

add a comment for the test case

b6658e1

Signed-off-by: Jorge Turrado <jorge_turrado@hotmail.es>

JaydipGabani approved these changes Jan 22, 2024

View reviewed changes

ritazh approved these changes Jan 23, 2024

View reviewed changes

Merge branch 'master' into set-insecure

ab23e8f

ritazh requested a review from maxsmythe January 23, 2024 04:24

make replace optional

944182b

Signed-off-by: Jorge Turrado <jorge.turrado@scrm.lidl>

JorTurFer force-pushed the set-insecure branch from aea7e07 to 944182b Compare February 7, 2024 08:58

JorTurFer and others added 2 commits March 31, 2024 17:40

add a test

64b274a

Signed-off-by: Jorge Turrado <jorge_turrado@hotmail.es>

Merge branch 'master' into set-insecure

b6466d0

JorTurFer commented Mar 31, 2024

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix: Ensure that insecureSkipTLSVerify is false before setting caBundle on APIService #160

fix: Ensure that insecureSkipTLSVerify is false before setting caBundle on APIService #160

JorTurFer commented Nov 21, 2023

acpana Nov 30, 2023

JorTurFer Nov 30, 2023

maxsmythe Feb 7, 2024

JorTurFer Feb 7, 2024

maxsmythe Feb 7, 2024

JorTurFer Feb 13, 2024

maxsmythe Feb 15, 2024

JorTurFer Mar 24, 2024

JorTurFer Mar 31, 2024

JorTurFer May 27, 2024

codecov-commenter commented Nov 30, 2023 •

edited

Loading

JorTurFer commented Dec 12, 2023

JorTurFer commented Jan 14, 2024

ritazh left a comment

JorTurFer commented Feb 6, 2024

JaydipGabani commented Feb 7, 2024

JorTurFer Mar 31, 2024

	// CertRotator contains cert artifacts and a channel to close when the certs are ready.
	type CertRotator struct {
	reader SyncingReader
	writer client.Writer

	SecretKey types.NamespacedName
	CertDir string
	CAName string
	CAOrganization string
	DNSName string
	ExtraDNSNames []string
	IsReady chan struct{}
	Webhooks []WebhookInfo
	// FieldOwner is the optional fieldmanager of the webhook updated fields.
	FieldOwner string
	RestartOnSecretRefresh bool
	ExtKeyUsages *[]x509.ExtKeyUsage
	// RequireLeaderElection should be set to true if the CertRotator needs to
	// be run in the leader election mode.
	RequireLeaderElection bool
	// CaCertDuration sets how long a CA cert will be valid for.
	CaCertDuration time.Duration
	// ServerCertDuration sets how long a server cert will be valid for.
	ServerCertDuration time.Duration
	// RotationCheckFrequency sets how often the rotation is executed
	RotationCheckFrequency time.Duration
	// LookaheadInterval sets how long before the certificate is renewed
	LookaheadInterval time.Duration
	// CertName and Keyname override certificate path
	CertName string
	KeyName string

	certsMounted chan struct{}
	certsNotMounted chan struct{}
	wasCAInjected *atomic.Bool
	caNotInjected chan struct{}

	// testNoBackgroundRotation doesn't actually start the rotator in the background.
	// This should only be used for testing.
	testNoBackgroundRotation bool
	}

fix: Ensure that insecureSkipTLSVerify is false before setting caBundle on APIService #160

Are you sure you want to change the base?

fix: Ensure that insecureSkipTLSVerify is false before setting caBundle on APIService #160

Conversation

JorTurFer commented Nov 21, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov-commenter commented Nov 30, 2023 • edited Loading

Codecov Report

JorTurFer commented Dec 12, 2023

JorTurFer commented Jan 14, 2024

ritazh left a comment

Choose a reason for hiding this comment

JorTurFer commented Feb 6, 2024

JaydipGabani commented Feb 7, 2024

Choose a reason for hiding this comment

codecov-commenter commented Nov 30, 2023 •

edited

Loading