> For the complete documentation index, see [llms.txt](https://blackprince001.gitbook.io/flowbench-documentation/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://blackprince001.gitbook.io/flowbench-documentation/guide/targets-and-safety.md). # Targets and safety rails A load test is a weapon pointed at something, so where it may point and how hard it may fire are configuration a run is gated on — not conventions a careful author remembers. The target config owns both, and every check happens before any load is generated. ```yaml # targets/staging.yaml name: staging base_urls: - https://staging.shop.internal max_vus: 500 max_rps: 1000 request_timeout: 5s agent_addr: staging.shop.internal:9090 disallowed_modes: [stress, soak] ``` Target configs never carry credentials — auth material comes from the environment through `{{ env.* }}` ([authentication](/flowbench-documentation/guide/auth.md)). ## The allow-list `base_urls` is both the base URL requests resolve against and the **host allow-list**: a step whose URL resolves outside it — including a templated URL, an OAuth2 `token_url`, or a WebSocket handshake — is refused. The flow says what to do; the target says what it is allowed to touch. Unknown keys in a target file fail loudly rather than being ignored, for the same reason. ## Ceilings `max_vus` caps the schedule's peak VU count; `max_rps` caps a declared `arrival_cap`. A scenario that would exceed either is refused pre-run with exit `2` — finding out at plan time beats finding out 10,000 goroutines in. `disallowed_modes` refuses whole profiles against a target: a shared staging environment can permit load but never stress or soak. `request_timeout` bounds every single call against this target (default 30s), which is also what turns a hung endpoint into a clean, classified timeout failure instead of a stuck VU. ## Resolving targets `flowbench run` takes `--target` as either a name resolved in `--targets-dir` (default `targets/`) or a direct path to a file: ```bash flowbench run flow.yaml --target staging # targets/staging.yaml flowbench run flow.yaml --target ./local.yaml # explicit path ``` With no target at all, `--target local` falls back to `http://localhost:8080` — the local dev loop needs no ceremony. `flowbench target ` prints the resolved config as JSON; it exists both for humans and as the single source of truth the [Python SDK](/flowbench-documentation/guide/python-sdk.md) shells out to. ## Attaching an agent `agent_addr` names a [`flowbench-agent`](/flowbench-documentation/guide/agent.md) on the target host. When set, every load-family run polls it once a second and saves the target's CPU/memory series beside the run — the data that separates target saturation from generator saturation, and the input the [stress knee classification](/flowbench-documentation/guide/profiles-and-thresholds.md) needs. Polling is fail-open: an unreachable or dying agent never blocks or fails a run. ## The kill switch Every run can be stopped cleanly: Ctrl-C cancels the run context and unwinds every VU, and the partial run is still summarized and saved — a run that had to be killed is exactly the run worth reading. Under `--watch`, the live view adds a server-side abort button with the same semantics ([results server](/flowbench-documentation/guide/results-server.md)). --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://blackprince001.gitbook.io/flowbench-documentation/guide/targets-and-safety.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.