Troubleshooting

invalid scenario: ... is "token" extracted in an earlier step? A template uses a variable before any step extracts it, or the name is misspelled. Variables only flow forward within an iteration.

... is a public host whose ownership is not verified, so load is capped Verify the target (see safety) or test a private address.

Every request fails with blocked by safety The request goes to a public host other than the target. Allow it with --allow-host (CLI) or the target’s allowed hosts (server).

dropped iterations in an open-model run No virtual user was free when an iteration was due. Raise maxVUs, or the target is so slow that the planned rate cannot be served. Dropped iterations show the generator could not keep the schedule; they are not target errors.

The verdict is generator-limited A worker, or the machine running stampede run, was saturated (CPU, scheduling lag, dropped iterations, GC, file descriptors, ephemeral ports or network). Add workers, give them more CPU, or lower the load per worker. Running out of ephemeral ports usually means connections are not reused: keep-alive is off, or every request opens a new connection.

data.users reads a file, which needs the server to be started with --data-dir On a server, CSV and JSON feeders must live in its data directory. In the Compose stack ShopLab’s data/ folder is mounted there.

too many open files / local resource limit errors at high concurrency Raise the file-descriptor limit (ulimit -n) on the generating machine. Each open connection uses one.

Workers do not appear in stampede workers Check that the worker’s --token matches the server’s --join-token, that it can reach the server’s worker port (8081), and --insecure versus TLS matches the server.

stampede compare says not comparable The runs differ in scenario, load plan, duration or worker count, or one was generator-limited. Compare like with like.