CPU, memory and disk

These faults put pressure on the machine litmus-lite runs on, from inside the litmus-lite process. They answer "what happens to my service when it has to share this box with something greedy?", so they only tell you something about a service on the same machine.

All three are bounded and undone at the end of the fault.

CPU

cpu.hog keeps workers goroutines busy (default 1, at most 16).

In CursorIn a terminal

Burn 4 CPU workers for 10 seconds and tell me whether the API on port 8080 keeps its p95 under 300 ms.

litmus-lite hub import cpu.hog -beside ./api

faults:
  - { name: burn, kind: cpu.hog, duration: 10s, params: { workers: 4 } }
hypotheses:
  - { name: stays fast, metric: p95, operator: "<", value: 300ms }

Memory

memory.hog allocates and holds size (default 32MB, at most 256MB) for the fault's duration.

faults:
  - { name: eat-ram, kind: memory.hog, duration: 8s, params: { size: 128MB } }

Disk

disk.fill writes a file of size (default 16MB, at most 256MB) and deletes it when the fault ends. It needs -yes, never overwrites an existing file, does not follow symlinks and refuses a filesystem root.

In CursorIn a terminal

Fill 200 MB under /tmp for 8 seconds and tell me whether the upload endpoint still answers.

yes is refused through MCP unless you enabled it with LITMUS_LITE_MCP_ALLOW=yes; the agent should give you the command instead.

faults:
  - name: fill-tmp
    kind: disk.fill
    duration: 8s
    params: { size: 200MB, path: /tmp/litmus-lite-fill.bin }

litmus-lite run -yes api/disk-fill.chaos.yaml

Without path, the file goes in a new name under the system temp directory. Do not point it at a volume you care about.

Parameters in full: Faults reference.