roachprod/failure-injection: add failure-injection registry

This registry will allow for future usage of the failure
injection library through the CLI and the failure injection
planner/controller.

pkg/roachprod/failureinjection/failures/BUILD.bazel

@@ -5,6 +5,7 @@ go_library(
     srcs = [
         "failure.go",
         "network_partition.go",
+        "registry.go",
     ],
     importpath = "github.com/cockroachdb/cockroach/pkg/roachprod/failureinjection/failures",
     visibility = ["//visibility:public"],

pkg/roachprod/failureinjection/failures/network_partition.go

@@ -65,6 +65,10 @@ func MakeIPTablesPartitionNode(
 
 const IPTablesNetworkPartitionName = "iptables-network-partition"
 
+func registerIPTablesPartitionNode(r *FailureRegistry) {
+	r.add(IPTablesNetworkPartitionName, NetworkPartitionArgs{}, MakeIPTablesPartitionNode)
+}
+
 func (f *IPTablesPartitionNode) Description() string {
 	return IPTablesNetworkPartitionName
 }

pkg/roachprod/failureinjection/failures/registry.go

@@ -0,0 +1,72 @@
+// Copyright 2025 The Cockroach Authors.
+//
+// Use of this software is governed by the CockroachDB Software License
+// included in the /LICENSE file.
+
+package failures
+
+import (
+	"fmt"
+	"regexp"
+
+	"github.com/cockroachdb/cockroach/pkg/roachprod/logger"
+)
+
+type failureSpec struct {
+	makeFailureFunc func(clusterName string, l *logger.Logger, secure bool) (FailureMode, error)
+	args            FailureArgs
+}
+type FailureRegistry struct {
+	failures map[string]failureSpec
+}
+
+func NewFailureRegistry() *FailureRegistry {
+	return &FailureRegistry{
+		failures: make(map[string]failureSpec),
+	}
+}
+
+func (r *FailureRegistry) Register() {
+	registerIPTablesPartitionNode(r)
+}
+
+func (r *FailureRegistry) add(
+	failureName string,
+	args FailureArgs,
+	makeFailureFunc func(clusterName string, l *logger.Logger, secure bool) (FailureMode, error),
+) {
+	if _, ok := r.failures[failureName]; ok {
+		panic(fmt.Sprintf("failure %s already exists", failureName))
+	}
+	r.failures[failureName] = failureSpec{
+		makeFailureFunc: makeFailureFunc,
+		args:            args,
+	}
+}
+
+func (r *FailureRegistry) List(regex string) []string {
+	var filter *regexp.Regexp
+	if regex == "" {
+		filter = regexp.MustCompile(`.`)
+	} else {
+		filter = regexp.MustCompile(regex)
+	}
+
+	var matches []string
+	for name := range r.failures {
+		if filter.MatchString(name) {
+			matches = append(matches, name)
+		}
+	}
+	return matches
+}
+
+func (r *FailureRegistry) GetFailure(
+	clusterName, failureName string, l *logger.Logger, secure bool,
+) (FailureMode, error) {
+	spec, ok := r.failures[failureName]
+	if !ok {
+		return nil, fmt.Errorf("unknown failure %s", failureName)
+	}
+	return spec.makeFailureFunc(clusterName, l, secure)
+}