Fault types reference

rcfg-sim supports four fault types, enabled by name via --fault-types. Each fires per relevant event at the configured --fault-rate.

auth_fail

Rejects the SSH authentication as if the password were wrong — indistinguishable from a real auth failure. Use it to test retry logic, credential-rotation handling, and how your tool reports unreachable devices.

disconnect_mid

Streams a partial config and then hard-resets the TCP connection mid-transfer. The cut happens after 20–40% of the config bytes have been written (a uniform random point in that window). The command’s duration is still recorded before the connection is dropped, so your metrics reflect the work done. This is the fault for testing partial-read handling, resume logic, and integrity checks on collected configs.

slow_response

Multiplies the per-command response delay by a random 10–50×, capped at 60 seconds. A 10 ms floor is applied first so the multiplier is observable even when --response-delay-ms-max is 0. Use it to exercise timeouts, slow-loris tolerance, and concurrency back-pressure when a slice of the fleet drags.

malformed

Corrupts the streamed config output. The bytes before and after the corruption keep the zero-copy path, so this is realistic and cheap. Use it to test parser robustness, validation, and how your tool handles configs that don’t round-trip cleanly.

Combining faults

Enable several at once; each rolls independently per event:

--fault-rate 0.1 --fault-types "auth_fail,disconnect_mid,slow_response,malformed"

At --fault-rate 0.1, roughly one in ten relevant events triggers each enabled fault. Tune the rate to model anything from a mostly-healthy fleet to a chaos test.

Metrics

Each activation increments rcfgsim_faults_injected_total{type="auth_fail|disconnect_mid|slow_response|malformed"}. The label set is closed to exactly these four values. See the metrics reference.