Chaos Types Overview

Chaos Mesh supports 14 different chaos types for comprehensive fault injection across Kubernetes infrastructure and applications.

Overview

Each chaos type is implemented as a Kubernetes Custom Resource Definition (CRD) with its own controller and injection mechanisms. CRD Definitions: api/v1alpha1/
Controller Implementations: controllers/chaosimpl/
Daemon Implementations: pkg/chaosdaemon/

Pod-Level Chaos

PodChaos

Inject faults into pod lifecycle operations.

PodChaos

CRD: podchaos_types.go
Actions: pod-kill, pod-failure, container-kill

Actions:

pod-kill
pod-failure
container-kill

Kills entire pods by deleting them from Kubernetes.Use Cases:

Test pod restart and recovery
Validate StatefulSet resilience
Test readiness probes

Parameters:

gracePeriod: Seconds before forced deletion (default: 0)

Example:

action: pod-kill
gracePeriod: 30

Makes pods unavailable without killing them by injecting a failure condition.Use Cases:

Test service degradation
Validate circuit breakers
Test rolling updates under failure

Parameters:

duration: How long pod stays failed (required)

Example:

action: pod-failure
duration: 30s

Kills specific containers within a pod.Use Cases:

Test container restart policies
Validate sidecar resilience
Test init container dependencies

Parameters:

containerNames: List of containers to kill

Example:

action: container-kill
containerNames: ["app", "sidecar"]

Source: api/v1alpha1/podchaos_types.go:43-54

Network Chaos

NetworkChaos

Inject network-level faults including latency, packet loss, and partitions.

NetworkChaos

CRD: networkchaos_types.go
Actions: netem, delay, loss, duplicate, corrupt, partition, bandwidth

Actions:

delay - Network Latency

Add latency to network packets.Parameters:

latency: Delay duration (e.g., “100ms”)
jitter: Variation in delay (e.g., “10ms”)
correlation: Correlation percentage (0-100)

Example:

action: delay
delay:
  latency: "100ms"
  jitter: "10ms"
  correlation: "50"

Implementation: pkg/chaosdaemon/tc_server.go, pkg/chaosdaemon/netem/

loss - Packet Loss

Drop network packets randomly.Parameters:

loss: Percentage of packets to drop (0-100)
correlation: Correlation percentage

Example:

action: loss
loss:
  loss: "25"
  correlation: "25"

duplicate - Packet Duplication

Duplicate network packets.Parameters:

duplicate: Percentage of packets to duplicate
correlation: Correlation percentage

Example:

action: duplicate
duplicate:
  duplicate: "10"
  correlation: "25"

corrupt - Packet Corruption

Corrupt packet data.Parameters:

corrupt: Percentage of packets to corrupt
correlation: Correlation percentage

Example:

action: corrupt
corrupt:
  corrupt: "5"

partition - Network Partition

Block network traffic between pods.Parameters:

direction: to, from, or both
target: Target pod selector

Example:

action: partition
direction: both
target:
  selector:
    namespaces: ["default"]
    labelSelectors:
      app: "target-app"

bandwidth - Bandwidth Limitation

Limit network bandwidth.Parameters:

rate: Bandwidth limit (e.g., “1mbps”)
limit: Queue size in bytes
buffer: Token bucket buffer size

Example:

action: bandwidth
bandwidth:
  rate: "1mbps"
  limit: 20000
  buffer: 10000

netem - Combined Network Effects

Combine multiple network chaos effects.Parameters: Merge of delay, loss, duplicate, corrupt specsExample:

action: netem
delay:
  latency: "100ms"
loss:
  loss: "10"

Source: api/v1alpha1/networkchaos_types.go:48-73

I/O Chaos

IOChaos

Inject file system I/O faults.

IOChaos

CRD: iochaos_types.go
Actions: latency, fault, attrOverride, mistake

Actions:

latency

Add delay to I/O operations.Parameters:

delay: Duration (e.g., “100ms”)
path: File path pattern
methods: I/O methods to affect
percent: Percentage of operations (0-100)

action: latency
delay: "100ms"
path: "/var/data/*"
percent: 50

fault

Return errors on I/O operations.Parameters:

errno: Error code (e.g., 5 for EIO)
path: File path pattern
methods: I/O methods to affect

action: fault
errno: 5
path: "/data/file.txt"

attrOverride

Override file attributes.Parameters:

attr: Attribute overrides
path: File path pattern

action: attrOverride
attr:
  perm: 0000

mistake

Inject incorrect data into I/O.Parameters:

mistake: Mistake specification
path: File path pattern

action: mistake
path: "/data/*"

Implementation: FUSE-based interception (pkg/chaosdaemon/iochaos_server.go) Source: api/v1alpha1/iochaos_types.go:50-51

Stress Chaos

StressChaos

Generate CPU or memory stress on pods.

StressChaos

CRD: stresschaos_types.go
Stressors: CPU, Memory

Stressor Types:

CPU Stress
Memory Stress

Stress CPU cores.Parameters:

workers: Number of CPU workers
load: Load percentage per worker (0-100)
options: Additional stress-ng options

Example:

stressors:
  cpu:
    workers: 2
    load: 50

Source: api/v1alpha1/stresschaos_types.go:183-196

Stress memory allocation.Parameters:

workers: Number of memory workers
size: Memory size per worker (e.g., “256MB”, “50%”)
oomScoreAdj: OOM score adjustment (-1000 to 1000)
options: Additional stress-ng options

Example:

stressors:
  memory:
    workers: 4
    size: "256MB"

Source: api/v1alpha1/stresschaos_types.go:160-181

Implementation: Uses stress-ng (pkg/chaosdaemon/stress_server_linux.go)

Time Chaos

TimeChaos

Simulate clock skew by offsetting system time.

TimeChaos

CRD: timechaos_types.go
Mechanism: vDSO clock interception

Parameters:

timeOffset: Offset duration (e.g., “-1h”, “30m”, “5s”)
clockIds: Clock IDs to affect (CLOCK_REALTIME, CLOCK_MONOTONIC, etc.)

Example:

timeOffset: "-1h"
clockIds:
  - CLOCK_REALTIME
  - CLOCK_MONOTONIC

Use Cases:

Test time-sensitive logic
Validate timeout handling
Test distributed system clock drift
Certificate expiration testing

Implementation: Syscall interception (pkg/chaosdaemon/time_server_linux.go) Source: api/v1alpha1/timechaos_types.go:48-56

HTTP Chaos

HTTPChaos

Manipulate HTTP requests and responses.

HTTPChaos

CRD: httpchaos_types.go
Target: Request or Response

Actions:

Abort: Return error responses
Delay: Add latency
Replace: Modify request/response body
Patch: Modify headers

Matching Criteria:

port: Target port
path: URI path pattern
method: HTTP method (GET, POST, etc.)
code: Response status code
requestHeaders: Request header matchers
responseHeaders: Response header matchers

Example:

target: Request
port: 8080
path: "/api/*"
method: GET
abort: true

Implementation: Transparent proxy (pkg/chaosdaemon/httpchaos_server.go) Source: api/v1alpha1/httpchaos_types.go:40-86

DNS Chaos

DNSChaos

Inject DNS resolution errors.

DNSChaos

CRD: dnschaos_type.go
Actions: error, random

Actions:

error
random

Return DNS resolution errors.Example:

action: error
patterns:
  - "google.com"
  - "github.*"

Return random IP addresses.Example:

action: random
patterns:
  - "*.example.com"

Pattern Matching:

Exact match: "google.com"
Wildcard suffix: "github.*" (matches github.com, github.io)
Placeholder: "chaos-mes?.org" (matches chaos-mesh.org)

Implementation: Custom DNS server (pkg/chaosdaemon/dns_server.go) Source: api/v1alpha1/dnschaos_type.go:26-34

JVM Chaos

JVMChaos

Inject faults into JVM applications using Byteman.

JVMChaos

CRD: jvmchaos_types.go
Actions: latency, return, exception, stress, gc, ruleData, mysql

Actions:

latency - Method Latency

Add delay to Java method invocations.Parameters:

class: Java class name
method: Method name
latency: Delay in milliseconds

action: latency
class: "com.example.Service"
method: "getData"
latency: 1000

return - Override Return Value

Override method return values.Parameters:

class: Java class name
method: Method name
returnValue: Value to return

action: return
class: "com.example.Service"
method: "isEnabled"
returnValue: "false"

exception - Throw Exception

Throw exceptions from methods.Parameters:

class: Java class name
method: Method name
exception: Exception to throw

action: exception
class: "com.example.Service"
method: "process"
exception: "java.io.IOException"

stress - JVM Stress

Generate CPU or memory stress within JVM.Parameters:

cpuCount: Number of CPU cores to stress
memoryType: “stack” or “heap”

action: stress
cpuCount: 2
memoryType: "heap"

gc - Trigger GC

Trigger garbage collection.

action: gc

ruleData - Custom Byteman Rule

Execute custom Byteman rule.Parameters:

ruleData: Raw Byteman rule

action: ruleData
ruleData: |
  RULE custom rule
  CLASS com.example.Service
  METHOD process
  AT ENTRY
  IF true
  DO traceln("Method called")
  ENDRULE

mysql - MySQL JDBC Faults

Inject faults into MySQL JDBC operations.Parameters:

mysqlConnectorVersion: “5” or “8”
database: Database name pattern
table: Table name pattern
sqlType: SQL type (select, insert, update, delete, replace)
exception: Exception to throw
latency: Delay in milliseconds

action: mysql
mysqlConnectorVersion: "8"
database: "test"
table: "users"
sqlType: "select"
latency: 1000

Implementation: Byteman agent (pkg/chaosdaemon/jvm_server.go) Source: api/v1alpha1/jvmchaos_types.go:44-69

Kernel Chaos

KernelChaos

Inject faults into kernel functions using BPF.

KernelChaos

CRD: kernelchaos_types.go
Mechanism: BPF-based fault injection

Fail Types:

0: slab allocation failures (kmalloc)
1: page allocation failures
2: bio (block I/O) failures

Parameters:

failtype: What to fail (0, 1, or 2)
headers: Required kernel headers
callchain: Specific call chain to target
probability: Percentage (0-100)
times: Maximum failure count

Example:

failKernRequest:
  failtype: 0  # slab allocation
  probability: 50
  times: 100
  callchain:
    - funcname: "ext4_mount"
      predicate: 'STRNCMP(name->name, "bananas", 8)'

Use Cases:

Test memory allocation failures
Validate error handling in file system operations
Test resilience to kernel-level faults

Source: api/v1alpha1/kernelchaos_types.go:58-98

Block Chaos

BlockChaos

Inject delays into block device I/O operations.

BlockChaos

CRD: blockchaos_types.go
Actions: delay

Parameters:

volumeName: Name of the volume
delay.latency: I/O delay duration
delay.jitter: Jitter amount
delay.correlation: Correlation percentage

Example:

action: delay
volumeName: "data-volume"
delay:
  latency: "100ms"
  jitter: "10ms"
  correlation: "50"

Use Cases:

Test application behavior under slow disks
Validate timeout handling
Test I/O-bound applications

Implementation: pkg/chaosdaemon/blockchaos_server_linux.go Source: api/v1alpha1/blockchaos_types.go:35-73

Cloud Provider Chaos

AWSChaos

Inject faults into AWS infrastructure.

AWSChaos

CRD: awschaos_types.go
Actions: ec2-stop, ec2-restart, detach-volume

Actions:

ec2-stop
ec2-restart
detach-volume

Stop EC2 instances.Parameters:

awsRegion: AWS region
ec2Instance: Instance ID
secretName: AWS credentials secret

action: ec2-stop
awsRegion: "us-west-2"
ec2Instance: "i-1234567890abcdef0"

Restart EC2 instances.

action: ec2-restart
awsRegion: "us-west-2"
ec2Instance: "i-1234567890abcdef0"

Detach EBS volumes.Parameters:

volumeID: EBS volume ID
deviceName: Device name

action: detach-volume
awsRegion: "us-west-2"
ec2Instance: "i-1234567890abcdef0"
volumeID: "vol-1234567890abcdef0"
deviceName: "/dev/sdf"

Source: api/v1alpha1/awschaos_types.go:43-76

GCPChaos

Inject faults into GCP infrastructure.

GCPChaos

CRD: gcpchaos_types.go
Actions: node-stop, node-reset, disk-loss

Actions:

node-stop: Stop Compute Engine instances
node-reset: Reset Compute Engine instances
disk-loss: Detach persistent disks

Parameters:

project: GCP project ID
zone: GCP zone
instance: Instance name
deviceNames: Disk device names (for disk-loss)

Example:

action: disk-loss
project: "my-project"
zone: "us-central1-a"
instance: "my-instance"
deviceNames: ["disk-1", "disk-2"]

Source: api/v1alpha1/gcpchaos_types.go:44-77

AzureChaos

Inject faults into Azure infrastructure.

AzureChaos

CRD: azurechaos_types.go
Actions: vm-stop, vm-restart, disk-detach

Actions:

vm-stop: Stop Azure VMs
vm-restart: Restart Azure VMs
disk-detach: Detach managed disks

Parameters:

subscriptionID: Azure subscription ID
resourceGroupName: Resource group name
vmName: VM name
diskName: Disk name (for disk-detach)
lun: Logical unit number (for disk-detach)

Example:

action: disk-detach
subscriptionID: "sub-123"
resourceGroupName: "my-rg"
vmName: "my-vm"
diskName: "data-disk"
lun: 0

Source: api/v1alpha1/azurechaos_types.go:43-68

Physical Machine Chaos

PhysicalMachineChaos

Inject faults into physical machines or VMs outside Kubernetes.

PhysicalMachineChaos

CRD: physical_machine_chaos_types.go
Actions: 40+ actions across network, disk, process, JVM, and more

Action Categories:

Stress Actions

stress-cpu: CPU stress
stress-mem: Memory stress

Disk Actions

disk-read-payload: Disk read stress
disk-write-payload: Disk write stress
disk-fill: Fill disk space

Network Actions

network-corrupt: Packet corruption
network-duplicate: Packet duplication
network-loss: Packet loss
network-delay: Network latency
network-partition: Network partition
network-dns: DNS chaos
network-bandwidth: Bandwidth limitation
network-flood: Network flooding
network-down: Disable network interface

Process Actions

process: Kill or signal processes

JVM Actions

jvm-exception: Throw exceptions
jvm-gc: Trigger GC
jvm-latency: Method latency
jvm-return: Override return values
jvm-stress: JVM stress
jvm-rule-data: Custom Byteman rules
jvm-mysql: MySQL JDBC faults

Clock Actions

clock: Clock offset

Redis Actions

redis-expiration: Set key expiration
redis-penetration: Cache penetration
redis-cacheLimit: Limit cache size
redis-restart: Restart Redis Sentinel
redis-stop: Stop Redis Sentinel

Kafka Actions

kafka-fill: Fill Kafka topic
kafka-flood: Flood Kafka
kafka-io: Kafka I/O faults

HTTP Actions

http-abort: Abort HTTP requests
http-delay: HTTP latency
http-config: HTTP config manipulation
http-request: Send HTTP requests

File Actions

file-create: Create files
file-modify: Modify file permissions
file-delete: Delete files
file-rename: Rename files
file-append: Append to files
file-replace: Replace file content

VM Actions

vm: VM-level operations

Custom Actions

user_defined: Execute custom commands

Source: api/v1alpha1/physical_machine_chaos_types.go:22-69

PhysicalMachineChaos requires installing chaosd agent on target machines. It operates outside the Kubernetes cluster.

Common Fields

All chaos types share common fields:

Selector

Defines which pods/resources to target. See Selectors for details.

Mode

How many targets to affect:

one: Random single target
all: All matching targets
fixed: Fixed number (set value)
fixed-percent: Fixed percentage (set value)
random-max-percent: Random up to max percentage (set value)

Duration

How long the chaos lasts:

duration: "30s"  # 30 seconds
duration: "5m"   # 5 minutes
duration: "1h"   # 1 hour

Omit for indefinite duration (manual deletion required).

Remote Cluster

Target pods in remote clusters:

remoteCluster: "cluster-2"

Status Tracking

All chaos experiments track status with: Conditions (api/v1alpha1/common_types.go:45-52):

Selected: Targets have been selected
AllInjected: All targets are injected
AllRecovered: All targets are recovered
Paused: Experiment is paused

Records (api/v1alpha1/common_types.go:78-88):

Track injection/recovery events
Count successful operations
Record timestamps and error messages

Architecture

How chaos types are implemented

Selectors

Target selection mechanisms

Components

Component responsibilities

Documentation Index

​Overview

​Pod-Level Chaos

​PodChaos

PodChaos

​Network Chaos

​NetworkChaos

NetworkChaos

​I/O Chaos

​IOChaos

IOChaos

latency

fault

attrOverride

mistake

​Stress Chaos

​StressChaos

StressChaos

​Time Chaos

​TimeChaos

TimeChaos

​HTTP Chaos

​HTTPChaos

HTTPChaos

​DNS Chaos

​DNSChaos

DNSChaos

​JVM Chaos

​JVMChaos

JVMChaos

​Kernel Chaos

​KernelChaos

KernelChaos

​Block Chaos

​BlockChaos

BlockChaos

​Cloud Provider Chaos

​AWSChaos

AWSChaos

​GCPChaos

GCPChaos

​AzureChaos

AzureChaos

​Physical Machine Chaos

​PhysicalMachineChaos

PhysicalMachineChaos

​Common Fields

​Selector

​Mode

​Duration

​Remote Cluster

​Status Tracking

​Related Resources

Architecture

Selectors

Components

Overview

Pod-Level Chaos

PodChaos

Network Chaos

NetworkChaos

I/O Chaos

IOChaos

Stress Chaos

StressChaos

Time Chaos

TimeChaos

HTTP Chaos

HTTPChaos

DNS Chaos

DNSChaos

JVM Chaos

JVMChaos

Kernel Chaos

KernelChaos

Block Chaos

BlockChaos

Cloud Provider Chaos

AWSChaos

GCPChaos

AzureChaos

Physical Machine Chaos

PhysicalMachineChaos

Common Fields

Selector

Mode

Duration

Remote Cluster

Status Tracking

Related Resources