Skip to content

Device profiles

Faults are random per event across the whole fleet. A device profile is the opposite: a fixed way one device misbehaves on every session. A test can say “these 3 devices hang” and then check what your tooling did with exactly those devices.

Profile What it simulates What the client sees
hang:<duration> A device that stops responding mid-session Login succeeds, then the first command gets no output for <duration> (e.g. 3600s, 2h). Commands sent meanwhile are answered, in order, once the hang ends. Every new session hangs again.
latency:<min>-<max> A slow device or a long WAN path Every command waits uniform [min, max] (e.g. 200-800ms, 1s-3s, or a fixed 500ms). This replaces --response-delay-ms-min/max for that device.
refuse A dead or rate-limiting SSH daemon TCP connects, then closes before the SSH handshake
disconnect A device that always drops mid-transfer Every config stream (e.g. show running-config) is cut part way, as disconnect_mid at rate 1

In a latency range, a bare number on the left takes the unit on the right, so 200-800ms means 200 ms to 800 ms.

Profiles work for every driver: Cisco IOS, Junos and the TL1 personalities.

Assigning profiles to a share of the fleet

Section titled “Assigning profiles to a share of the fleet”

--device-profiles takes comma-separated profile@percent items:

Terminal window
./bin/rcfg-sim \
--listen-ip 127.0.0.1 --port-start 12000 --port-count 100 \
--manifest /tmp/rcfg-test/manifest.csv --host-key /tmp/rcfg-test/hostkey \
--device-profiles "hang:3600s@3%,latency:200-800ms@10%" \
--profile-seed 42
Flag Default Purpose
--device-profiles "" (off) profile@percent,... — which profiles, and what share of the fleet gets each
--profile-seed 42 Decides which devices are picked

Which devices get which profile depends only on --profile-seed and the device’s hostname. The same seed picks the same devices on every run and on every simulator instance, however the fleet is split across IPs. Percentages may add up to at most 100, and no device gets two profiles.

At startup the simulator logs one device profile active line per profile, with the device count and up to 50 hostnames:

{"level":"INFO","msg":"device profile active","profile":"hang:1h0m0s","devices":3,"hostnames":"rtr-msp-hq-1012,rtr-lax-br-1047,rtr-nyc-dc-1083","hostnames_truncated":false}

Use these hostnames to find the devices in your tooling. hostnames_truncated=true means more than 50 devices have that profile.

The manifest takes an optional 11th column, profile, holding one profile without a percentage:

hostname,ip,port,vendor,template,username,password,enable_password,config_file,size_bucket,profile
rtr-msp-hq-1000,127.0.0.1,12000,Cisco,cisco_ios,admin,admin,enable123,/tmp/rcfg-test/configs/device-00000.cfg,md,hang:7200s
rtr-msp-hq-1001,127.0.0.1,12001,Cisco,cisco_ios,admin,admin,enable123,/tmp/rcfg-test/configs/device-00001.cfg,md,
  • A value in the column overrides --device-profiles for that device.
  • An empty value leaves the device to --device-profiles.
  • Manifests without the column load as before. The generator does not write it; add it by hand.
  • An invalid value stops the simulator at startup, naming the device.

See the manifest schema.

The simulator closes every session after --max-session (default 5m). For a hang device the cap is extended by the hang, so a 2-hour hang is not cut at 5 minutes.

Every other session keeps the cap, including slow latency devices. If a slow device’s whole session runs past 5 minutes, raise the cap or the simulator, not the client under test, ends the session:

Terminal window
--device-profiles "latency:2s-5s@10%" --max-session 15m

A hang also ends early when the client disconnects or the simulator shuts down.

Profiles reuse the existing metrics. No new labels are added:

Profile Shows up as
refuse rcfgsim_sessions_total{result="error"}
disconnect rcfgsim_sessions_total{result="disconnect"} and rcfgsim_faults_injected_total{type="disconnect_mid"}
hang Long sessions in rcfgsim_session_duration_seconds
latency Slow commands in rcfgsim_command_duration_seconds

See the metrics reference.

You want Use
A realistic random share of failures across the fleet --fault-rate / --fault-types
The same named devices misbehaving on every run, to assert on them Device profiles

The two combine: a profiled device still rolls any enabled faults, except a refuse device, which closes before any fault can fire.