Ctrlk
Sign UpGet a DemoContact Us
For the complete documentation index, see llms.txt. This page is also available as Markdown.

Linux network loss

Drop a percentage of network packets leaving a target Linux machine for a configurable duration so you can test how the workload behaves when packets are lost.

Linux network loss is a chaos fault that drops NETWORK_PACKET_LOSS_PERCENTAGE percent of packets leaving the target Linux machine on NETWORK_INTERFACES for DURATION, then restores normal connectivity. Loss is restricted to traffic destined for DESTINATION_HOSTS/DESTINATION_IPS and the configured port filters; SSH ports stay reachable when WHITELIST_SSH is true. The fault runs through the Linux Chaos Infrastructure (LCI) systemd service installed on the target VM.

Use this fault to test how a workload behaves when the network is unreliable: whether application clients honor their own timeouts, whether retries amplify the loss, whether circuit breakers fire correctly, and whether monitoring detects the connectivity degradation within the alerting SLA.

RUN YOUR FIRST EXPERIMENT

If you have not installed the Linux Chaos Infrastructure yet, go to Linux Chaos Infrastructure to install the agent and connect the VM to the control plane.

Use cases

Run this fault when you want to answer concrete questions like:

  • Packet-loss tolerance: When NETWORK_PACKET_LOSS_PERCENTAGE of packets to DESTINATION_HOSTS are dropped, do application clients honor their own timeouts cleanly?

  • Retry storms: Do retries amplify the lost traffic, or do exponential backoff and jitter prevent it?

  • Circuit breakers: Does the service-mesh or application circuit breaker open within the configured threshold?

  • Monitoring fidelity: Do alerts on connection errors, retransmits (node_netstat_TcpExt_TCPLostRetransmit), and end-to-end p99 fire within the alerting SLA?

Prerequisites

  • Linux Chaos Infrastructure installed: The linux-chaos-infrastructure systemd service is active on the target VM and the infrastructure is in CONNECTED state. Go to Linux Chaos Infrastructure to install it.

  • Target interface exists: Each entry in NETWORK_INTERFACES exists on the target VM. Confirm with ip -br link.

  • tc available: The fault uses Linux Traffic Control (tc) with the netem qdisc to apply loss. Provided by iproute2, installed on all supported distributions.

Supported environments

The fault has been tested on the following Linux distributions. Go to Linux fault requirements to see the full compatibility matrix.

Platform
Support status

Ubuntu 16+, Debian 10+

Supported

CentOS 7+, RHEL 7+, Fedora 30+

Supported

openSUSE LEAP 15.4+ / SUSE Linux Enterprise 15+

Supported

Permissions required

This fault is classified as an Advanced Linux fault. It requires the Linux Chaos Infrastructure systemd service to run with the root user and root user group on the target VM so it can manage the tc qdisc. No cloud credentials are needed.

Fault tunables

Configure the following fault parameters when you add Linux network loss to an experiment in Chaos Studio. Defaults are shown for reference.

Chaos parameters

Tunable
Description
Default

DURATION

Total duration of the fault. Accepts [hours]h[minutes]m[seconds]s format.

30s

NETWORK_PACKET_LOSS_PERCENTAGE

Percentage of packets to drop (0 to 100).

100

NETWORK_INTERFACES

Comma-separated network interfaces to apply chaos on (for example, eth0,ens192).

eth0

RAMP_TIME

Wait period in seconds before and after the fault. Go to ramp time to read how it is applied.

0

Target filters (provide at least one host or IP to limit blast radius)

Tunable
Description
Default

DESTINATION_HOSTS

Comma-separated destination host names to target (for example, google.com,api.example.com).

""

DESTINATION_IPS

Comma-separated destination IPs to target. Per-IP ports can be specified using the ip|port format (for example, 1.1.1.1,35.24.108.92|3000|8080).

""

SOURCE_PORTS

Comma-separated source ports to target. Prefix with ! to whitelist instead of target (for example, !22,2222 excludes SSH).

""

DESTINATION_PORTS

Comma-separated destination ports to target. Prefix with ! to whitelist instead of target.

""

WHITELIST_SSH

Keep SSH ports (22,2222) excluded from chaos to preserve management access.

true

When neither DESTINATION_HOSTS nor DESTINATION_IPS is set, the fault applies to all destinations on the interface.

Tunables that apply to every fault are documented in common tunables for all faults.

Fault execution in brief

Adds a netem qdisc on NETWORK_INTERFACES that drops NETWORK_PACKET_LOSS_PERCENTAGE percent of egress packets matching the configured destination and port filters for DURATION, then removes the qdisc.

Expected behavior during fault execution

  • The configured percentage of egress packets to matched destinations is dropped for the duration of the fault.

  • TCP connections see retransmissions (TCPLostRetransmit); UDP traffic loses datagrams silently.

  • Application calls to the matched destinations time out or fail; retries may succeed when they hit the surviving packets.

  • After the duration ends, the qdisc is removed and packet flow returns to baseline.

WHEN THE FAULT ENDS

The chaos pod removes the netem qdisc. Packets flow normally on the next send; in-flight retransmissions may take a few RTTs to settle.

Signals to watch

Attach resilience probes to assert each layer:

  • TCP retransmits: Use a Prometheus probe on node_netstat_TcpExt_TCPLostRetransmit and assert it rose during the chaos window.

  • Application timeouts: Use a Prometheus probe on application timeout/connection-error counters.

  • End-to-end availability: Use an HTTP probe on a user-visible endpoint that depends on the matched destinations.

Verify the fault execution effect

While the experiment is running, confirm packets were dropped and then recovered:

  1. Inspect the active qdisc.

    A netem qdisc with the configured loss parameter should be present during the chaos window and removed afterwards.

  2. Run ping or curl against a matched destination.

    Packet loss and curl timeouts should match approximately NETWORK_PACKET_LOSS_PERCENTAGE.

  3. Inspect Linux Chaos Infrastructure logs.

    Look for the fault start, the resolved targets, and the fault end markers.

Recovery and cleanup

  • End of duration: The chaos pod removes the netem qdisc when DURATION elapses; packets flow normally.

  • Abort the experiment: Stopping the experiment from Chaos Studio also removes the qdisc.

  • Manual recovery: If the qdisc survives an abort, remove it with sudo tc qdisc del dev <interface> root (or parent ffff: for ingress qdiscs created by the fault).

  • Workload recovery: Applications resume normal throughput as soon as the qdisc is removed; long-lived connections that broke must be re-established by the client.

Limitations

  • Egress only: The fault drops packets leaving the VM on the configured interfaces; inbound packets are not directly dropped.

  • Single VM scope: Each fault run targets one VM (the VM hosting the selected Linux Chaos Infrastructure).

  • Loss is uniform random: Drops are random per packet; the actual drop rate approximates NETWORK_PACKET_LOSS_PERCENTAGE over many packets.

  • SSH whitelisting: When WHITELIST_SSH is false, the fault may sever your SSH session to the target VM if the SSH ports match the filter.

  • Existing qdisc: If a custom qdisc is already attached to the interface, the fault adds a child or fails; review with tc qdisc show before running.

Troubleshooting

Linux network loss fault shows no packet loss in Harness Chaos Engineering

Confirm the destination filter matches your test traffic and that you are sending egress traffic on the configured NETWORK_INTERFACES. Run tc -s qdisc show dev during the chaos window to see the netem statistics.

SSH session dropped during the experiment

WHITELIST_SSH defaults to true to keep ports 22 and 2222 reachable. If SSH dropped, your session may be on a custom port. Add that port to SOURCE_PORTS with the ! prefix to whitelist it (for example, !22,2222,2200).

Packet loss persists after the experiment ends

If the netem qdisc was not removed (for example, the experiment aborted ungracefully), remove it manually with sudo tc qdisc del dev root. Inspect with tc qdisc show dev first.

Related faults

Last updated

Was this helpful?