MON-4608: add interrupts to NodeExporterCollectorConfig CRD types - #2955

In response to this:

Summary

Add NodeExporterCollectorInterruptsConfig to the NodeExporterCollectorConfig CRD type, enabling users to configure the node-exporter interrupts collector via the ClusterMonitoring
CR.

Unlike simpler collectors (softirqs, zoneinfo) that use a plain collectionPolicy on/off switch, the interrupts collector follows the systemd pattern — a union with collectionPolicy
as the discriminator and an optional collect sub-struct containing an include list of regex patterns. This design matches the CMO implementation on the
node-exporter-collector-interrupts branch, which passes the include patterns to
--collector.interrupts.name-include.

Changes

• Added Interrupts field to NodeExporterCollectorConfig
• Added NodeExporterCollectorInterruptsConfig (union: collectionPolicy + collect)
• Added NodeExporterCollectorInterruptsCollectConfig with include list (+listType=set, max 50 items)
• Added NodeExporterInterruptsIncludePattern validated string type (1–1024 chars)
• XValidation rule: collect is forbidden when collectionPolicy is not Collect
• Regenerated CRDs, deepcopy, OpenAPI, swagger docs, and model names

Cross-reference

• CMO branch: node-exporter-collector-interrupts — uses Interrupts.Include []string internally; config_merge.go will translate from this CRD's collectionPolicy/collect.include
  to the internal representation when vendoring.
• Sibling PR pattern: follows the same framework as the zoneinfo branch (node-exporter-collector-zoneinfo), extended with the union pattern from the systemd collector.

Verification

• make update (full regeneration) ✅
• make build ✅
• make -C config/v1alpha1 test — 262 tests passed ✅
• make lint — 0 issues ✅

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the openshift-eng/jira-lifecycle-plugin repository.

Walkthrough

Adds the disabled-by-default interrupts node-exporter collector to the ClusterMonitoring API and CRD. It supports collection policies and optional regex filtering for /proc/interrupts lines. Validation enforces policy, length, printable-ASCII, uniqueness, and list-size rules. Updates netClass and systemd validation messages and test expectations.

Suggested reviewers: joelspeed, marioferh

• Add interrupts collector configuration to the ClusterMonitoring Node Exporter config.
• Model interrupts as a union with optional collect.include regex filters.
• Regenerate CRDs and OpenAPI artifacts to publish the new schema.

graph TD
  A["ClusterMonitoring CR"] --> B["NodeExporterCollectorConfig"] --> C["Interrupts union config"] --> D["CMO config merge"] --> E["node-exporter args"] --> F["interrupts collector"]
  F --> G["/proc/interrupts"]
  subgraph Legend
    direction LR
    _cr["CRD/CR"] ~~~ _svc(["Controller/Service"]) ~~~ _fs[["Host file"]]
  end

Loading

The following are alternative approaches to this PR:

1. Flat on/off only (collectionPolicy without collect/include)

• ➕ Simpler schema and fewer validation rules
• ➕ Lower risk of user misconfiguration around regex patterns
• ➖ Cannot bound metric cardinality when enabled
• ➖ Does not match node-exporter’s interrupts collector filtering capabilities

2. Single string regex (nameInclude) instead of list of patterns

• ➕ Direct mapping to a single node-exporter flag value
• ➕ Simpler merging logic than set semantics
• ➖ Harder for users to manage multiple patterns safely
• ➖ Loses uniqueness/set behavior and per-pattern validation ergonomics

Recommendation: The chosen union design (collectionPolicy discriminator + optional collect.include set) is the best fit: it matches the established 'systemd pattern' in this API, allows optional filtering to control cardinality, and enforces correctness with XValidation (collect forbidden unless Collect). The added list constraints (+listType=set, max 50, per-item length) provide reasonable guardrails while keeping the UX flexible.

Enhancement (3) +118 / -0

types_cluster_monitoring.goAdd interrupts collector types and wire into NodeExporterCollectorConfig +69/-0

Add interrupts collector types and wire into NodeExporterCollectorConfig

• Introduces the 'Interrupts' field on 'NodeExporterCollectorConfig'. Adds 'NodeExporterCollectorInterruptsConfig' as a union keyed by 'collectionPolicy', with an optional 'collect.include' set of regex patterns and an XValidation rule forbidding 'collect' unless 'Collect'. Defines a validated string type for include patterns.

config/v1alpha1/types_cluster_monitoring.go

zz_generated.deepcopy.goRegenerate deepcopy implementations for interrupts config types +39/-0

Regenerate deepcopy implementations for interrupts config types

• Adds deepcopy methods for 'NodeExporterCollectorInterruptsConfig' and 'NodeExporterCollectorInterruptsCollectConfig', and updates 'NodeExporterCollectorConfig' deepcopy to copy the new 'Interrupts' field.

config/v1alpha1/zz_generated.deepcopy.go

zz_generated.model_name.goRegister OpenAPI model names for interrupts config types +10/-0

Register OpenAPI model names for interrupts config types

• Adds OpenAPI model name functions for the new interrupts config structs so they are emitted consistently in generated OpenAPI.

config/v1alpha1/zz_generated.model_name.go

Documentation (1) +20 / -0

zz_generated.swagger_doc_generated.goRegenerate swagger docs for interrupts collector fields and types +20/-0

Regenerate swagger docs for interrupts collector fields and types

• Extends SwaggerDoc maps to document the new 'interrupts' field and the new interrupts config types, including the union behavior and include semantics.

config/v1alpha1/zz_generated.swagger_doc_generated.go

Other (5) +334 / -1

0000_10_config-operator_01_clustermonitorings.crd.yamlRegenerate CRD manifest to include interrupts collector schema +66/-0

Regenerate CRD manifest to include interrupts collector schema

• Publishes the new 'interrupts' field in the ClusterMonitoring CRD, including union semantics, 'collect.include' constraints (set, min/max items), and the XValidation rule tying 'collect' to 'collectionPolicy=Collect'.

config/v1alpha1/zz_generated.crd-manifests/0000_10_config-operator_01_clustermonitorings.crd.yaml

ClusterMonitoringConfig.yamlRegenerate featuregated CRD manifest with interrupts schema +66/-0

Regenerate featuregated CRD manifest with interrupts schema

• Mirrors the CRD schema changes for feature-gated manifests, ensuring the 'interrupts' collector config and validations are present in the featuregated output.

config/v1alpha1/zz_generated.featuregated-crd-manifests/clustermonitorings.config.openshift.io/ClusterMonitoringConfig.yaml

zz_generated.openapi.goEmit OpenAPI schemas for interrupts union and include set +86/-1

Emit OpenAPI schemas for interrupts union and include set

• Adds OpenAPI definitions for 'NodeExporterCollectorInterruptsConfig' and 'NodeExporterCollectorInterruptsCollectConfig', wires 'interrupts' into 'NodeExporterCollectorConfig', and includes union metadata ('x-kubernetes-unions') and list-type extensions.

openapi/generated_openapi/zz_generated.openapi.go

openapi.jsonRegenerate JSON OpenAPI output with interrupts schemas +50/-0

Regenerate JSON OpenAPI output with interrupts schemas

• Updates the published OpenAPI JSON to include the new interrupts-related definitions and the 'interrupts' property on the collector config.

openapi/openapi.json

0000_10_config-operator_01_clustermonitorings.crd.yamlUpdate payload CRD manifest with interrupts collector schema +66/-0

Update payload CRD manifest with interrupts collector schema

• Propagates the regenerated CRD schema into the shipped payload manifests so clusters receive the new 'interrupts' configuration surface.

payload-manifests/crds/0000_10_config-operator_01_clustermonitorings.crd.yaml

Description

The new optional InterruptsConfig.Collect field is documented as optional, but does not explain
what happens when it is omitted (defaults/behavior). This violates the requirement that new optional
fields document omit behavior.

Code

config/v1alpha1/types_cluster_monitoring.go[R721-728]

+	// collect contains configuration options that apply only when the interrupts collector is actively collecting metrics
+	// (i.e. when collectionPolicy is Collect).
+	// collect is optional and may be omitted even when collectionPolicy is Collect.
+	// collect may only be set when collectionPolicy is Collect.
+	// When set, at least one field must be specified within collect.
+	// +unionMember
+	// +optional
+	Collect NodeExporterCollectorInterruptsCollectConfig `json:"collect,omitzero,omitempty"`

Relevance

●● Moderate

Team often documents omit behavior, but a similar “optional/nullable omit behavior” doc request was
previously rejected.

PR-#2851
PR-#2697

ⓘ Recommendations generated based on similar findings in past PRs

Evidence

PR Compliance ID 1507 requires every new optional field to explicitly document what happens when it
is not provided. The Collect field is +optional but its comment only states it may be omitted,
without describing the resulting behavior/defaults.

config/v1alpha1/types_cluster_monitoring.go[721-728]
Skill: api-review

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
`Collect` is marked `+optional` but its comment does not describe the behavior when the field is omitted (e.g., what defaults apply / what is collected).

## Issue Context
Optional-field documentation must clearly state omit behavior ("When omitted...", "Defaults to..."), especially for API/CRD fields.

## Fix Focus Areas
- config/v1alpha1/types_cluster_monitoring.go[721-728]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

Description

The interrupts.collect.include field is documented as allowing an empty list to mean “collect all
interrupts,” but it is also constrained with +kubebuilder:validation:MinItems=1, which rejects
include: [] during CRD validation. This doc/schema mismatch creates misleading API documentation
and can confuse users (including breaking templating/merge workflows that intentionally emit empty
lists).

Code

config/v1alpha1/types_cluster_monitoring.go[R735-751]

+	// include is a list of regular expression patterns that select which interrupt lines to collect.
+	// include is optional.
+	// Each line in /proc/interrupts is matched against the same string node-exporter uses:
+	// the IRQ name, info, and devices fields joined with ";", for example "LOC;77;IO-APIC 2-edge ...".
+	// Patterns are combined with OR into a single expression anchored on both ends,
+	// so each pattern must match the entire string (use ".*" where needed).
+	// An empty list or omitting include collects all interrupt lines.
+	// Each entry must be at least 1 character and at most 1024 characters.
+	// Maximum length for this list is 50.
+	// Minimum length for this list is 1.
+	// Entries in this list must be unique.
+	// +kubebuilder:validation:MaxItems=50
+	// +kubebuilder:validation:MinItems=1
+	// +listType=set
+	// +optional
+	Include []NodeExporterInterruptsIncludePattern `json:"include,omitempty"`
+}

Relevance

● Weak

Very similar schema-vs-doc contradiction cleanup was previously rejected; likely they won’t change
MinItems/comment alignment.

PR-#2645

ⓘ Recommendations generated based on similar findings in past PRs

Evidence

PR Compliance ID 1509 disallows documentation that describes behavior for values that are made
invalid by kubebuilder validation markers. In the Go type, the comment for Include states that an
empty list (or omitting the field) collects all interrupt lines, yet the kubebuilder marker
MinItems=1 requires at least one item; this flows into the generated CRD where the same
description appears alongside minItems: 1, so include: [] is invalid in practice and the
documented behavior cannot be exercised.

config/v1alpha1/types_cluster_monitoring.go[735-751]
config/v1alpha1/types_cluster_monitoring.go[731-751]
config/v1alpha1/zz_generated.crd-manifests/0000_10_config-operator_01_clustermonitorings.crd.yaml[2187-2233]
Skill: api-review

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
`NodeExporterCollectorInterruptsCollectConfig.Include` is documented as supporting an empty list (e.g., “An empty list … collects all interrupt lines”), but the field is annotated with `+kubebuilder:validation:MinItems=1`, which makes `include: []` invalid and rejected by CRD validation.

## Issue Context
This is a user-visible API contract mismatch and violates the requirement (PR Compliance ID 1509) that documentation must not describe behavior for values that are invalidated by kubebuilder markers. As implemented and generated, YAML such as `collect: { include: [] }` will fail validation despite the docs implying it is valid; this can also disrupt templating/merge workflows that emit empty lists to explicitly clear patterns.

Align the intended behavior between documentation and schema:
- If an empty list should be allowed to mean “collect all,” remove `MinItems=1` (keeping `+optional` as appropriate) and ensure the generated CRDs reflect the change.
- If an empty list should not be allowed, update the docs to say only omission collects all lines, and remove the “empty list” statement.

## Fix Focus Areas
- config/v1alpha1/types_cluster_monitoring.go[735-751]
- config/v1alpha1/zz_generated.crd-manifests/0000_10_config-operator_01_clustermonitorings.crd.yaml[2187-2233]

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

In response to this:

/override ci/prow/verify-hypershift-integration

Fix pending for this job, but this PR shouldn't affect it anyway

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository.

• OWNERS [everettraven]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

In response to this:

@JoelSpeed / @everettraven

Change Overview

Commit: 42ec1d006 [MON-4608](https://redhat.atlassian.net/browse/MON-4608): add interrupts to NodeExporterCollectorConfig CRD types

What Changed

1. New Go types in config/v1alpha1/types_cluster_monitoring.go:

• NodeExporterCollectorInterruptsConfig — union type with collectionPolicy (discriminator) and collect (union member)
• NodeExporterCollectorInterruptsCollectConfig — holds include list of regex patterns
• NodeExporterInterruptsIncludePattern — string type with validation (MinLength=1, MaxLength=1024, printable ASCII only)

2. New field Interrupts added to NodeExporterCollectorConfig struct

3. Updated error messages for NodeExporterCollectorNetClassConfig and NodeExporterCollectorSystemdConfig XValidation rules (from "collect is forbidden when collectionPolicy is not Collect" to "collect may be set when collectionPolicy is Collect, and forbidden otherwise")

4. Generated files updated: deepcopy, swagger docs, OpenAPI definitions, CRD manifests, model names

Validation Rules

Rule	Description
Union discriminator	collectionPolicy is required, valid values: Collect, DoNotCollect
XValidation	When collectionPolicy=Collect, collect field is required; when DoNotCollect, collect is forbidden
MinProperties=1	collect object must have at least one field
MinItems=1	include list must have at least 1 entry
MaxItems=50	include list may have at most 50 entries
listType=set	No duplicate entries in include
MinLength=1	Each include pattern must be at least 1 character
MaxLength=1024	Each include pattern must be at most 1024 characters
XValidation (regex)	Each pattern must contain only printable ASCII (regex: ^[\x20-\x7E]+$)

Cluster Environment

Property	Value
OCP Version	4.22.5
Kubernetes Version	v1.35.5
Feature Set	Default
oc Client Version	5.0.0-ec.4
Nodes	3 (control-plane/master/worker)
Node Names	hub-ctlplane-{0,1,2}.hub.5g-deployment.lab

Pre-Cluster Test Results

Integration Tests (make -C config/v1alpha1 test)

Ran 279 of 6589 Specs in 18.875 seconds
SUCCESS! -- 279 Passed | 0 Failed | 0 Pending | 6383 Skipped

Result: PASS

Unit Tests (make test-unit)

All unit tests passed. No failures detected.

Result: PASS

Codegen Verification (make verify-codegen-crds)

Exit code: 0. All generated CRDs are up to date.

Result: PASS

Lint (make lint)

Failed with a pre-existing local toolchain issue (plugin.Open error on kube-api-linter.so). Verified the same failure occurs on master without any changes — not related to this PR.

Result: N/A (pre-existing toolchain issue)

Cluster Functional Tests

CRD Application

The clustermonitorings.config.openshift.io CRD was applied using server-side apply:

oc apply -f config/v1alpha1/zz_generated.crd-manifests/0000_10_config-operator_01_clustermonitorings.crd.yaml --server-side --force-conflicts

CRD status: Established=True, NamesAccepted=True

Test Results

#	Test Name	Expected	Actual	Result
1	Create with interrupts Collect + include list	Created	clustermonitoring.config.openshift.io/cluster created	PASS
2	Reject interrupts Collect without collect field	Rejected	collect is required when collectionPolicy is Collect, and forbidden otherwise	PASS
3	Accept interrupts DoNotCollect without collect	Accepted	configured — spec shows {"collectionPolicy":"DoNotCollect"}	PASS
4	Reject interrupts with invalid collectionPolicy	Rejected	Unsupported value: "Invalid": supported values: "Collect", "DoNotCollect"	PASS
5	Reject interrupts with empty object (required field)	Rejected	spec.nodeExporterConfig.collectors.interrupts.collectionPolicy: Required value	PASS
6	Reject DoNotCollect with collect set	Rejected	collect is required when collectionPolicy is Collect, and forbidden otherwise	PASS
7	Reject Collect with empty collect object	Rejected	should have at least 1 properties + include: Required value	PASS
8	Accept Collect with wildcard include ".*"	Accepted	spec shows {"collect":{"include":[".*"]},"collectionPolicy":"Collect"}	PASS
9	Reject include with control characters (\t)	Rejected	must contain only printable ASCII characters (no control characters)	PASS
10	Accept full nodeExporterConfig with interrupts + other collectors	Accepted	All collectors and settings persisted correctly	PASS
11	Accept Collect with multiple include patterns	Accepted	5 patterns stored: ["LOC.*","NMI.*","RES.*","CAL.*","TLB.*"]	PASS
12	Reject empty include list (MinItems=1)	Rejected	should have at least 1 items	PASS
13	Reject duplicate include patterns (listType=set)	Rejected	Duplicate value: "LOC.*"	PASS
14	Reject include pattern exceeding MaxLength=1024	Rejected	may not be more than 1024 bytes	PASS
15	Accept include pattern at exactly MaxLength=1024	Accepted	1024-char pattern stored successfully	PASS
16	Reject empty include pattern (MinLength=1)	Rejected	should be at least 1 chars long + ASCII validation	PASS
17	Reject include list exceeding MaxItems=50	Rejected	Too many: 51: must have at most 50 items	PASS
18	Accept include list at exactly MaxItems=50	Accepted	50 patterns stored successfully	PASS
19	Update from Collect to DoNotCollect	Accepted	spec shows {"collectionPolicy":"DoNotCollect"}	PASS
20	Update from DoNotCollect back to Collect with include	Accepted	spec shows include with "LOC.*"	PASS
21	Update include patterns (add more)	Accepted	3 patterns stored: ["LOC.*","NMI.*","RES.*"]	PASS
22	Reject update to Collect without collect	Rejected	collect is required when collectionPolicy is Collect, and forbidden otherwise	PASS
23	Remove interrupts entirely (leave other collector)	Accepted	Only cpuFreq remains in collectors	PASS
24	Verify netClass error message change	Correct	collect may be set when collectionPolicy is Collect, and forbidden otherwise	PASS
25	Verify systemd error message change	Correct	collect may be set when collectionPolicy is Collect, and forbidden otherwise	PASS
26	Verify valid printable ASCII pattern accepted	Accepted	["valid-pattern-123.*"] stored	PASS

All 26 functional tests: PASS

Reconciliation Stability

Test Methodology

After applying the CRD and creating a ClusterMonitoring CR with interrupts configuration, the resourceVersion of both the CRD and the CR were recorded, then checked after waiting periods.

Results

Check	CRD resourceVersion	CR resourceVersion	Duration
Before wait	120045	126357	—
After 60 seconds	120045	126357	60s
After 120 seconds	120045	128449*	120s

*Note: The CR resourceVersion changed between the 60s and 120s checks because the test itself applied a new configuration (Test 10+). After the final configuration was applied at resourceVersion 128449, it remained stable for the full 120-second observation window.

Operator Health During Test

Operator	Available	Progressing	Degraded
config-operator	True	False	False
monitoring	True	False	False

No reconciliation events detected on the CRD.

Result: PASS — The CRD and CR were not reconciled or overwritten by any cluster operator.

Conclusion

All changes introduced by commit 42ec1d006 (MON-4608) pass functional validation:

1. Integration tests: 279/279 passed
2. Unit tests: All passed
3. Codegen verification: Up to date (exit code 0)
4. Cluster functional tests: 26/26 passed — all validation rules enforced correctly on a live OCP 4.22.5 cluster
5. Reconciliation stability: No drift detected over 120-second observation — CRD and CR stable
6. Operator health: config-operator and monitoring operator remain Available and non-Degraded
7. Error message updates: netClass and systemd XValidation messages correctly updated

The interrupts collector configuration follows the established union pattern used by other collectors (netClass, systemd), with the additional safety requirement that collect.include is mandatory when collectionPolicy=Collect to prevent uncontrolled metric cardinality.

/verified by @midu16

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the openshift-eng/jira-lifecycle-plugin repository.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository. I understand the commands that are listed here.