roachprod/cli: add disk-stall chaos command with cgroup support

pkg/cmd/roachprod/cli/BUILD.bazel

@@ -5,6 +5,7 @@ go_library(
     srcs = [
         "bash_completion.go",
         "chaos.go",
+        "chaos_disk.go",
         "chaos_network.go",
         "commands.go",
         "flags.go",

pkg/cmd/roachprod/cli/chaos.go

@@ -121,6 +121,7 @@ Default: all`)
 	// Add subcommands
 	chaosCmd.AddCommand(cr.buildChaosNetworkPartitionCmd())
 	chaosCmd.AddCommand(cr.buildChaosNetworkLatencyCmd())
+	chaosCmd.AddCommand(cr.buildChaosDiskStallCmd())
 
 	return chaosCmd
 }

pkg/cmd/roachprod/cli/chaos_disk.go

@@ -0,0 +1,137 @@
+// Copyright 2025 The Cockroach Authors.
+//
+// Use of this software is governed by the CockroachDB Software License
+// included in the /LICENSE file.
+
+package cli
+
+import (
+	"context"
+	"time"
+
+	"github.com/cockroachdb/cockroach/pkg/roachprod/config"
+	"github.com/cockroachdb/cockroach/pkg/roachprod/failureinjection/failures"
+	"github.com/cockroachdb/errors"
+	"github.com/spf13/cobra"
+)
+
+// buildChaosDiskStallCmd creates the disk-stall chaos command
+func (cr *commandRegistry) buildChaosDiskStallCmd() *cobra.Command {
+	var nodes []int32
+	var stallType string
+	var restartNodes bool
+	var cycle bool
+	var cycleStallDuration, cycleUnstallDuration time.Duration
+
+	cmd := &cobra.Command{
+		Use:   "disk-stall <cluster>",
+		Short: "Injects disk stalls to test storage resilience",
+		Long: `Injects disk stalls using cgroup to throttle disk I/O.
+
+Disk stall type:
+  cgroup: Uses cgroups v2 to throttle disk I/O operations
+
+By default, cgroup stalls write operations completely. This simulates scenarios where
+the disk becomes slow or unresponsive, allowing you to test how CockroachDB handles
+degraded storage performance.
+
+The stall can be temporary (default 5 minutes) or you can use --run-forever to keep
+it active until manually stopped with Ctrl+C.
+
+Examples:
+  # Stall disk I/O on nodes 1,2,3 for 5 minutes (default)
+  roachprod chaos disk-stall mycluster --type cgroup --nodes 1,2,3
+
+  # Cycle stalling on/off with custom durations
+  roachprod chaos disk-stall mycluster --type cgroup --nodes 1 \
+    --cycle --cycle-stall-duration 10s --cycle-unstall-duration 5s
+
+  # Stall until manually interrupted
+  roachprod chaos disk-stall mycluster --type cgroup --nodes 1 --run-forever
+
+  # Stall for 15 minutes before recovery
+  roachprod chaos disk-stall mycluster --type cgroup --nodes 1 --wait-before-cleanup 15m
+
+	# Stall for 30s before recovery on an insecure cluster
+	roachprod chaos disk-stall mycluster --type cgroup --nodes 1 --wait-before-cleanup 30s --insecure
+`,
+		Args: cobra.ExactArgs(1),
+		Run: Wrap(func(cmd *cobra.Command, args []string) error {
+			clusterName := args[0]
+			nodeList := parseInt32SliceToNodes(nodes)
+
+			// Validate type
+			var failureName string
+			var failureArgs failures.DiskStallArgs
+
+			switch stallType {
+			case "cgroup":
+				failureName = failures.CgroupsDiskStallName
+				// Cgroup defaults: stall writes completely
+				// These can be exposed as flags in the future:
+				// - StallReads (default: false)
+				// - StallWrites (default: true)
+				// - StallLogs (default: false)
+				// - Throughput (default: 0 = full stall, can be 2+ for throttling)
+				failureArgs = failures.DiskStallArgs{
+					StallReads:           false,
+					StallWrites:          true,
+					StallLogs:            false,
+					Throughput:           0, // Full stall
+					RestartNodes:         restartNodes,
+					Nodes:                nodeList,
+					Cycle:                cycle,
+					CycleStallDuration:   cycleStallDuration,
+					CycleUnstallDuration: cycleUnstallDuration,
+				}
+				config.Logger.Printf("Stalling disk I/O on nodes %s using cgroup",
+					formatNodeList(nodeList))
+
+			default:
+				return errors.Newf("--type must be 'cgroup', got: %s", stallType)
+			}
+
+			opts, err := getGlobalChaosOpts()
+			if err != nil {
+				return err
+			}
+
+			// Create failer from registry
+			failer, err := createFailer(
+				clusterName,
+				failureName,
+				opts,
+				getClusterOptions()...,
+			)
+			if err != nil {
+				return err
+			}
+
+			return runFailureLifecycle(
+				context.Background(),
+				failer,
+				failureArgs,
+				opts,
+			)
+		}),
+	}
+
+	cmd.Flags().Int32SliceVarP(&nodes, "nodes", "n", nil,
+		"nodes on which to inject the disk stall")
+	cmd.Flags().StringVar(&stallType, "type", "",
+		"disk staller type: cgroup (uses cgroups v2 to throttle I/O)")
+	cmd.Flags().BoolVar(&restartNodes, "restart-nodes", true,
+		"allow the failure mode to restart nodes as needed")
+	cmd.Flags().BoolVar(&cycle, "cycle", false,
+		"repeatedly stall and unstall the disk until recovery")
+	cmd.Flags().DurationVar(&cycleStallDuration, "cycle-stall-duration", 5*time.Second,
+		"duration of each stall cycle")
+	cmd.Flags().DurationVar(&cycleUnstallDuration, "cycle-unstall-duration", 5*time.Second,
+		"duration of each unstall cycle")
+
+	_ = cmd.MarkFlagRequired("nodes")
+	_ = cmd.MarkFlagRequired("type")
+	initFlagInsecureForCmd(cmd)
+
+	return cmd
+}