cmpatino's picture
cmpatino HF Staff
deploy controller (2e85dec)
3ed7ad0 verified
Raw History Blame Contribute Delete
9.27 kB
"""`flock` CLI: a thin client over the controller API.
Config: FLOCK_URL and FLOCK_TOKEN. Inside an agent the worker sets FLOCK_AGENT_TOKEN (that
agent's scoped token, which wins over FLOCK_TOKEN), FLOCK_AGENT and FLOCK_PARENT; see
docs/spawn.md for what an agent may call.
"""
from __future__ import annotations
import json
import os
import time
from pathlib import Path
from typing import Any
import httpx
import typer
import yaml
from flock.models import AGENT_NAME_RE, SwarmSpec
app = typer.Typer(help="Host coding-harness agent swarms on HF Jobs.", no_args_is_help=True)
def client() -> httpx.Client:
"""Overridden in tests to point at an in-process app."""
url = os.environ.get("FLOCK_URL", "http://localhost:7860")
token = os.environ.get("FLOCK_AGENT_TOKEN") or os.environ.get("FLOCK_TOKEN", "")
return httpx.Client(base_url=url, headers={"Authorization": f"Bearer {token}"}, timeout=60)
def _call(method: str, path: str, **kw) -> Any:
with client() as c:
r = c.request(method, path, **kw)
if r.status_code >= 400:
typer.echo(f"error {r.status_code}: {r.text}", err=True)
raise typer.Exit(1)
return r.json()
def _as_agent() -> bool:
"""Running inside an agent with its scoped token: names are agents, never swarms."""
return bool(os.environ.get("FLOCK_AGENT_TOKEN"))
def _is_swarm(name: str) -> bool:
"""Only a name shaped like an agent's (`...-NNNN`) costs a lookup."""
if not AGENT_NAME_RE.match(name):
return True
with client() as c:
return c.get(f"/swarms/{name}").status_code == 200
def _table(rows: list[list[Any]], header: list[str]) -> None:
cells = [header] + [[("" if v is None else str(v)) for v in row] for row in rows]
widths = [max(len(r[i]) for r in cells) for i in range(len(header))]
for r in cells:
typer.echo(" ".join(v.ljust(w) for v, w in zip(r, widths, strict=True)).rstrip())
def _states(d: dict[str, int]) -> str:
return " ".join(f"{k}:{v}" for k, v in sorted(d.items())) or "-"
def _print_status(st: dict) -> None:
typer.echo(
f"{st['swarm']} agents={st['agents_total']} workers={st['workers_total']} "
f"pending={st['pending_prompts']} cost=${st['cost_usd']:.2f}"
)
rows = [
[
t["name"],
t["health"],
t["replicas"],
_states(t["agents_by_state"]),
_states(t["workers_by_state"]),
f"${t['cost_usd']:.2f}",
]
for t in st["types"].values()
]
_table(rows, ["type", "health", "replicas", "agents", "workers", "cost"])
def load_spec(spec_file: str) -> SwarmSpec:
"""Read a spec and inline each type's system_prompt_file (relative to the spec file)."""
path = Path(spec_file)
spec = SwarmSpec.model_validate(yaml.safe_load(path.read_text(encoding="utf-8")))
for name, ts in spec.types.items():
if ts.system_prompt_file and ts.system_prompt is None:
src = Path(ts.system_prompt_file)
src = src if src.is_absolute() else path.parent / src
if not src.is_file():
typer.echo(f"type {name}: system_prompt_file {src} not found", err=True)
raise typer.Exit(1)
ts.system_prompt = src.read_text(encoding="utf-8")
return spec
@app.command()
def apply(spec_file: str) -> None:
"""Apply a swarm spec (YAML)."""
spec = load_spec(spec_file)
st = _call("PUT", f"/swarms/{spec.swarm}", json=spec.model_dump(mode="json"))
_print_status(st)
@app.command()
def status(swarm: str = typer.Argument(None)) -> None:
"""Swarm status: counts by type and state. Given an agent name (or inside an agent), that
agent's status as JSON."""
if not swarm and _as_agent():
swarm = os.environ.get("FLOCK_AGENT")
if swarm and (_as_agent() or not _is_swarm(swarm)):
typer.echo(json.dumps(_call("GET", f"/agents/{swarm}"), indent=2))
return
if swarm:
_print_status(_call("GET", f"/swarms/{swarm}")["status"])
return
swarms = _call("GET", "/swarms")
if not swarms:
typer.echo("no swarms")
for st in swarms:
_print_status(st)
typer.echo("")
@app.command()
def agents(
swarm: str,
type: str = typer.Option(None, "--type"),
state: str = typer.Option(None, "--state"),
parent: str = typer.Option(None, "--parent", help="only the children of this agent"),
) -> None:
"""List a swarm's agents."""
params = {k: v for k, v in (("type", type), ("state", state), ("parent", parent)) if v}
rows, cursor = [], None
while True:
page = _call(
"GET",
f"/swarms/{swarm}/agents",
params={**params, "cursor": cursor} if cursor else params,
)
for a in page["agents"]:
rows.append(
[
a["name"],
a["state"],
a["worker"],
a["turns"],
a["pending"],
f"${a['cost_usd']:.2f}",
a["last_heartbeat"],
]
)
cursor = page["cursor"]
if not cursor:
break
_table(rows, ["name", "state", "worker", "turns", "pending", "cost", "last_heartbeat"])
@app.command()
def agent(name: str) -> None:
"""One agent's status as JSON."""
typer.echo(json.dumps(_call("GET", f"/agents/{name}"), indent=2))
@app.command()
def prompt(
name: str,
text: str,
timeout: str = typer.Option(None, "--timeout"),
reply_to: str = typer.Option(None, "--reply-to"),
) -> None:
"""Send a prompt to an agent; prints the ack."""
body = {"text": text, "timeout": timeout, "reply_to": reply_to}
typer.echo(json.dumps(_call("POST", f"/agents/{name}/prompt", json=body)))
@app.command()
def events(
name: str,
after: int = typer.Option(0, "--after"),
follow: bool = typer.Option(False, "--follow", "-f"),
wait: bool = typer.Option(False, "--wait", help="block until an event after --after"),
timeout: float = typer.Option(600, "--timeout", help="seconds --wait blocks at most"),
) -> None:
"""Print an agent's event stream (JSONL); --follow keeps long-polling; --wait returns once
something new arrived (the last line on stderr is the cursor for the next --after)."""
deadline = time.monotonic() + timeout
while True:
poll = follow or wait
page = _call(
"GET", f"/agents/{name}/events", params={"after": after, "wait": 25 if poll else 0}
)
for ev in page["events"]:
typer.echo(json.dumps(ev))
after = page["next"]
if wait and (page["events"] or time.monotonic() >= deadline):
typer.echo(f"next: {after}", err=True)
break
if not poll:
break
@app.command()
def scale(swarm: str, type: str, replicas: int) -> None:
"""Set a type's replicas (edits the stored spec and re-applies it)."""
spec = SwarmSpec.model_validate(_call("GET", f"/swarms/{swarm}")["spec"])
if type not in spec.types:
typer.echo(f"no type {type!r} in swarm {swarm!r}", err=True)
raise typer.Exit(1)
spec.types[type].replicas = replicas
_print_status(_call("PUT", f"/swarms/{swarm}", json=spec.model_dump(mode="json")))
@app.command()
def remove(swarm: str) -> None:
"""Drain and remove a swarm. Given a spawned agent's name (or inside an agent), remove that
agent and everything it spawned."""
if _as_agent() or not _is_swarm(swarm):
typer.echo(json.dumps(_call("DELETE", f"/agents/{swarm}")))
return
typer.echo(json.dumps(_call("DELETE", f"/swarms/{swarm}")))
@app.command()
def spawn(
type: str,
prompt: str = typer.Option(None, "--prompt", help="first prompt for the child"),
key: str = typer.Option(None, "--key", help="idempotency key: a repeat returns the same child"),
parent: str = typer.Option(None, "--parent", help="default: $FLOCK_AGENT (yourself)"),
as_json: bool = typer.Option(False, "--json", help="print the whole ack"),
) -> None:
"""Spawn a child agent of TYPE; prints its name."""
parent = parent or os.environ.get("FLOCK_AGENT")
if not parent:
typer.echo("no parent: pass --parent or set FLOCK_AGENT", err=True)
raise typer.Exit(1)
body = {"type": type, "prompt": prompt, "key": key}
ack = _call("POST", f"/agents/{parent}/spawn", json=body)
typer.echo(json.dumps(ack) if as_json else ack["name"])
@app.command()
def whoami() -> None:
"""Who the configured token acts as: the controller, or an agent with its lineage."""
typer.echo(json.dumps(_call("GET", "/whoami"), indent=2))
@app.command()
def workers(swarm: str) -> None:
"""List a swarm's workers."""
rows = [
[
w["id"],
w["type"],
w["state"],
w["job_id"],
len(w["agents"]),
w["crashes"],
w["deadline"],
w["last_heartbeat"],
]
for w in _call("GET", "/workers", params={"swarm": swarm})
]
_table(rows, ["id", "type", "state", "job", "agents", "crashes", "deadline", "last_heartbeat"])
if __name__ == "__main__":
app()