Skip to main content

Linux API latency

Last updated on

Linux API latency is a chaos fault that starts a local proxy on the target Linux machine, redirects traffic to/from TARGET_SERVICE_PORT through the proxy on PROXY_PORT, and adds LATENCY of delay to matching API requests for DURATION. DATA_DIRECTION chooses whether the delay is applied to the request, response, or both. Filters and HTTPS interception controls mirror Linux API block. The fault runs through the Linux Chaos Infrastructure (LCI) systemd service installed on the target VM.

Use this fault to test how callers behave when an API gets slow: whether they honor their own timeouts, whether retries amplify the delay, whether thread pools exhaust, and whether monitoring detects the elevated tail latency 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:

  • Caller timeouts: When the API takes LATENCY longer than usual, do callers honor their own timeouts cleanly?
  • Thread-pool starvation: Do upstream worker pools back up when responses get slow?
  • Retry storms: Do retries amplify the slow traffic, or do exponential backoff and jitter contain it?
  • Monitoring fidelity: Do alerts on end-to-end p99, queue depth, and timeout counters 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 service port reachable: TARGET_SERVICE_PORT is the port the application listens on (ingress) or the upstream service port (egress).
  • Proxy port free: PROXY_PORT (default 20000) is not bound by another process on the target VM.
  • HTTPS material if HTTPS_ENABLED=true: Provide a CA bundle in HTTPS_ROOT_CERT_FILE and additional certificates in CUSTOM_CERTIFICATES.
  • /tmp is exec-mountable: Verify with findmnt -l | grep noexec | grep /tmp. If /tmp is mounted noexec, remount with sudo mount /tmp -o remount,exec.

Supported environments​

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

PlatformSupport 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 install iptables redirects and bind the proxy port. No cloud credentials are needed.

Fault tunables​

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

Chaos parameters

TunableDescriptionDefault
DURATIONTotal duration of the fault. Accepts [hours]h[minutes]m[seconds]s format.30s
LATENCYLatency to inject per matching request (for example, 2s, 500ms).2s
TARGET_SERVICE_PORTPort of the application (ingress) or upstream service (egress) to intercept.80
PROXY_PORTPort on which the local proxy listens.20000
NETWORK_INTERFACENetwork interface used for ingress redirection.eth0
DATA_DIRECTIONDirection of data to delay. Accepts request, response, or both.both
RAMP_TIMEWait period in seconds before and after the fault. Go to ramp time to read how it is applied.0

Request filters (combine as needed)

TunableDescriptionDefault
PATH_FILTERAPI path glob to match. Leave empty to match all paths.""
HEADERS_FILTERSHTTP headers filter (comma-separated Header=value pairs).""
METHODSComma-separated HTTP methods to match.""

Direction and source/destination filters

TunableDescriptionDefault
SERVICE_DIRECTIONingress (delay requests arriving at the VM) or egress (delay requests leaving the VM).ingress
SOURCE_HOSTSComma-separated source host names to match (ingress only).""
SOURCE_IPSComma-separated source IPs to match (ingress only).""
DESTINATION_HOSTSComma-separated destination host names to match (egress only).""
DESTINATION_IPSComma-separated destination IPs to match (egress only).""
DESTINATION_PORTSComma-separated destination ports for egress traffic.""

HTTPS interception (egress)

TunableDescriptionDefault
HTTPS_ENABLEDEnable HTTPS interception. Required for matching HTTPS traffic by PATH_FILTER, headers, or methods.false
HTTPS_ROOT_CERT_FILEPath to the root CA bundle used by the proxy to terminate egress HTTPS.""
CUSTOM_CERTIFICATESBase64-encoded custom certificates added to the proxy trust store.""

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

Fault execution in brief​

Adds an iptables redirect on NETWORK_INTERFACE that routes TARGET_SERVICE_PORT traffic through the local proxy on PROXY_PORT for DURATION. Matching requests are held by LATENCY (in DATA_DIRECTION) before being forwarded; non-matching requests pass through unchanged. On exit, the iptables redirect and proxy are removed.

Expected behavior during fault execution​

  • Matching API calls take LATENCY longer than usual for the duration of the fault.
  • Callers see slower responses; tail latency (p99) shifts upward by approximately LATENCY.
  • Callers with shorter timeouts than LATENCY see clean timeout errors.
  • After the duration ends, the iptables redirect is removed and the proxy stops; traffic resumes its original path with no added latency.
When the fault ends

The chaos pod removes the iptables redirect and stops the local proxy. The next request flows directly with no added latency.

Signals to watch​

Attach resilience probes to assert each layer:

  • End-to-end latency: Use a Prometheus probe on application p95/p99 metrics.
  • Caller timeout counters: Use a Prometheus probe on caller timeout/connection-error counters.
  • HTTP probe: Use an HTTP probe on a user-visible endpoint that exercises the slowed API.

Verify the fault execution effect​

  1. Send a matching request from the VM.

    curl -w '%{time_total}\n' -o /dev/null -s http://localhost:<TARGET_SERVICE_PORT><PATH_FILTER-example>

    Latency should rise by approximately LATENCY during the chaos window.

  2. Inspect iptables redirects.

    sudo iptables -t nat -L -n -v | grep <PROXY_PORT>

  3. Inspect Linux Chaos Infrastructure logs.

    sudo journalctl -u linux-chaos-infrastructure -n 100 --no-pager

Recovery and cleanup​

  • End of duration: The chaos pod removes the iptables redirect and stops the proxy when DURATION elapses.
  • Abort the experiment: Stopping the experiment from Chaos Studio also removes the redirect and stops the proxy.
  • Manual recovery: If the redirect survives an abort, inspect with sudo iptables -t nat -L -n -v and delete the rule added by the fault; kill the proxy with sudo lsof -i :<PROXY_PORT> followed by sudo kill <pid>.

Limitations​

  • HTTP/3 (QUIC) not intercepted.
  • HTTPS requires certificates: Without HTTPS_ROOT_CERT_FILE, HTTPS traffic cannot be matched on path/header/method.
  • Single VM scope: Each fault run targets one VM.
  • Per-request delay: LATENCY is applied per matching request; high-request-rate workloads accumulate large delays.
  • /tmp exec required.

Troubleshooting​

Linux API latency fault did not slow any request in Harness Chaos Engineering

Verify PATH_FILTER, METHODS, and HEADERS_FILTERS match your test traffic. For HTTPS endpoints, set HTTPS_ENABLED=true and provide HTTPS_ROOT_CERT_FILE. Confirm the iptables redirect is in place with sudo iptables -t nat -L -n -v.

Proxy port already in use

PROXY_PORT (default 20000) conflicted with another listener. Change PROXY_PORT to a free port and re-run.

Latency persists after the experiment ends

If the iptables redirect was not removed, inspect with sudo iptables -t nat -L -n -v and delete the rule added by the fault. Kill any orphan proxy on PROXY_PORT.