Simulate Network Faults

This document describes how to simulate network faults using NetworkChaos in Chaos Mesh.

NetworkChaos introduction

NetworkChaos is a fault type in Chaos Mesh. By creating a NetworkChaos experiment, you can simulate a network fault scenario for a cluster. Currently, NetworkChaos supports the following fault types:

• Partition: network disconnection and partition.
• Net Emulation: poor network conditions, such as high delays, high packet loss rate, packet reordering, and so on.
• Bandwidth: limit the communication bandwidth between nodes.

Notes

Before creating NetworkChaos experiments, ensure the following:

1. During the network injection process, make sure that the connection between Controller Manager and Chaos Daemon works, otherwise the NetworkChaos cannot be restored anymore.
2. If you want to simulate Net Emulation fault, make sure the NET_SCH_NETEM module is installed in the Linux kernel. If you are using CentOS, you can install the module through the kernel-modules-extra package. Most other Linux distributions have installed the module already by default.

Create experiments using Chaos Dashboard

1. Open Chaos Dashboard, and click NEW EXPERIMENT on the page to create a new experiment:

   

2. In the Choose a Target area, choose NETWORK ATTACK and select a specific behavior, such as LOSS. Then fill out specific configuration.

   

   For details of specific configuration fields, refer to Field description.

3. Fill out the experiment information, and specify the experiment scope and the scheduled experiment duration.

   

4. Submit the experiment information.

Create experiments using the YAML files

Net emulation example

1. Write the experiment configuration to the network-delay.yaml file, as shown below:

   apiVersion: chaos-mesh.org/v1alpha1
   kind: NetworkChaos
   metadata:
     name: delay
   spec:
     action: delay
     mode: one
     selector:
       namespaces:
         - default
       labelSelectors:
         'app': 'web-show'
     delay:
       latency: '10ms'
       correlation: '100'
       jitter: '0ms'

   This configuration causes a latency of 10 milliseconds in the network connections of the target Pods. In addition to latency injection, Chaos Mesh supports packet loss and packet reordering injection. For details, see field description.

2. After the configuration file is prepared, use kubectl to create an experiment:

   kubectl apply -f ./network-delay.yaml

Partition example

1. Write the experiment configuration to the network-partition.yaml file, as shown below:

   apiVersion: chaos-mesh.org/v1alpha1
   kind: NetworkChaos
   metadata:
     name: partition
   spec:
     action: partition
     mode: all
     selector:
       namespaces:
         - default
       labelSelectors:
         'app': 'app1'
     direction: to
     target:
       mode: all
       selector:
         namespaces:
           - default
         labelSelectors:
           'app': 'app2'

   This configuration blocks the connection created from app1 to app2. The value for the direction field can be to, from or both. For details, refer to Field description.

2. After the configuration file is prepared, use kubectl to create the experiment:

   kubectl apply -f ./network-partition.yaml

Bandwidth example

1. Write the experiment configuration to the network-bandwidth.yaml file, as shown below:

   apiVersion: chaos-mesh.org/v1alpha1
   kind: NetworkChaos
   metadata:
     name: bandwidth
   spec:
     action: bandwidth
     mode: all
     selector:
       namespaces:
         - default
       labelSelectors:
         'app': 'app1'
     bandwidth:
       rate: '1mbps'
       limit: 20971520
       buffer: 10000

   This configuration limits the bandwidth of app1 to 1 mbps.

2. After the configuration file is prepared, use kubectl to create the experiment:

   kubectl apply -f ./network-bandwidth.yaml

Field description

Parameter	Type	Description	Default value	Required	Example
action	string	Indicates the specific fault type. Available types include: netem, delay (network delay), loss (packet loss), duplicate (packet duplicating), corrupt (packet corrupt), partition (network partition), and bandwidth (network bandwidth limit). After you specify action field, refer to Description for action-related fields for other necessary field configuration.	None	Yes	Partition
target	Selector	Used in combination with direction, making Chaos only effective for some packets.	None	No
direction	enum	Indicates the direction of target packets. Available values include from (the packets from target), to (the packets to target), and both ( the packets from or to target). This parameter makes Chaos only take effect for a specific direction of packets.	to	No	both
mode	string	Specifies the mode of the experiment. The mode options include one (selecting a random Pod), all (selecting all eligible Pods), fixed (selecting a specified number of eligible Pods), fixed-percent (selecting a specified percentage of Pods from the eligible Pods), and random-max-percent (selecting the maximum percentage of Pods from the eligible Pods).	None	Yes	one
value	string	Provides a parameter for the mode configuration, depending on mode. For example, when mode is set to fixed-percent, value specifies the percentage of Pods.	None	No	1
selector	struct	Specifies the target Pod. For details, refer to Define the experiment scope.	None	Yes
externalTargets	[]string	Indicates the network targets except for Kubernetes, which can be IPv4 addresses or domains. This parameter only works with direction: to.	None	No	1.1.1.1, google.com
device	string	Specifies the affected network interface	None	No	"eth0"

Description for action-related fields

For the Net Emulation and Bandwidth fault types, you can further configure the action related parameters according to the following description.

• Net Emulation type: delay, loss, duplicated, corrupt
• Bandwidth type: bandwidth

delay

Setting action to delay means simulating network delay fault. You can also configure the following parameters.

Parameter	Type	Description	Required	Required	Example
latency	string	Indicates the network latency	No	No	2ms
correlation	string	Indicates the correlation between the current latency and the previous one. Range of value: [0, 100]	No	No	50
jitter	string	Indicates the range of the network latency	No	No	1ms
reorder	Reorder(#Reorder)	Indicates the status of network packet reordering		No

The computational model for correlation is as follows:

1. Generate a random number whose distribution is related to the previous value:

   rnd = value * (1-corr) + last_rnd * corr

   rnd is the random number. corr is the correlation you fill out before.

2. Use this random number to determine the delay of the current packet:

   ((rnd % (2 * sigma)) + mu) - sigma

   In the above command, sigma is jitter and mu is latency.

reorder

Setting action to reorder means simulating network packet reordering fault. You can also configure the following parameters.

Parameter	Type	Description	Default value	Required	Example
reorder	string	Indicates the probability to reorder	0	No	0.5
correlation	string	Indicates the correlation between this time's length of delay time and the previous time's length of delay time. Range of value: [0, 100]	0	No	50
gap	int	Indicates the gap before and after packet reordering	0	No	5

loss

Setting action to loss means simulating packet loss fault. You can also configure the following parameters.

Parameter	Type	Description	Default value	Required	Example
loss	string	Indicates the probability of packet loss. Range of value: [0, 100]	0	No	50
correlation	string	Indicates the correlation between the probability of current packet loss and the previous time's packet loss. Range of value: [0, 100]	0	No	50

duplicate

Set action to duplicate, meaning simulating package duplication. At this point, you can also set the following parameters.

Parameter	Type	Description	Default value	Required	Example
duplicate	string	Indicates the probability of packet duplicating. Range of value: [0, 100]	0	No	50
correlation	string	Indicates the correlation between the probability of current packet duplicating and the previous time's packet duplicating. Range of value: [0, 100]	0	No	50

corrupt

Setting action to corrupt means simulating package corruption fault. You can also configure the following parameters.

Parameter	Type	Description	Default value	Required	Example
corrupt	string	Indicates the probability of packet corruption. Range of value: [0, 100]	0	No	50
correlation	string	Indicates the correlation between the probability of current packet corruption and the previous time's packet corruption. Range of value: [0, 100]	0	No	50

For occasional events such as reorder, loss, duplicate, and corrupt, the correlation is more complicated. For specific model description, refer to NetemCLG.

bandwidth

Setting action to bandwidth means simulating bandwidth limit fault. You also need to configure the following parameters.

Parameter	Type	Description	Default value	Required	Example
rate	string	Indicates the rate of bandwidth limit		Yes	1mbps
limit	uint32	Indicates the number of bytes waiting in queue		Yes	1
buffer	uint32	Indicates the maximum number of bytes that can be sent instantaneously		Yes	1
peakrate	uint64	Indicates the maximum consumption of bucket (usually not set)		No	1
minburst	uint32	Indicates the size of peakrate bucket (usually not set)		No	1

For more details of these fields, you can refer to tc-tbf document. The limit is suggested to set to at least 2 * rate * latency, where the latency is the estimated latency between source and target, and it can be estimated through ping command. Too small limit can cause high loss rate and impact the throughput of the tcp connection.