feat(backends,api,ui): process trees with cpu and stderr per session, interrupt, close and kill from the system page

tests/test_api.py

@@ -396,6 +396,54 @@ async def test_describe_snapshots_open_tools_and_records_rate_limit(
     assert page["offset"] == 0 and len(page["entries"]) == 1
 
 
+async def test_sessions_show_their_processes_and_take_orders(world: World) -> None:
+    world.backend._factory = ToolClient  # noqa: SLF001
+    api = Api(world)
+    api.store.grant("api-only", "weak", scope="api")
+    weak = {"Authorization": "Bearer weak"}
+    conv = await world.conversations.create(kind="master", agent="a", origin="test")
+    key = conv.external_id
+    ScriptedClient.hold = asyncio.Event()
+    await world.conversations.post(conv, "go")
+    await asyncio.sleep(0.3)
+    census = await api.get("/sessions")
+    (row,) = census["sessions"]
+    assert row["key"] == key and row["busy"]
+    assert row["processes"] == [] and row["cpu"] is None
+    assert isinstance(census["others"], list)
+    live = world.pool.get(key)
+    assert live is not None
+    live.stderr.append("warning: something")
+    stderr = await api.get(f"/sessions/{key}/stderr")
+    assert stderr["lines"] == ["warning: something"]
+    for path in (f"/sessions/{key}/close", "/processes/1/kill"):
+        assert (await api.http.post(path, headers=weak)).status_code == 403
+    assert (
+        await api.http.get(f"/sessions/{key}/stderr", headers=weak)
+    ).status_code == 403
+
+    closed = await api.http.post(f"/sessions/{key}/close", headers=HEADERS)
+    assert closed.json() == {"key": key, "closed": True}
+    assert world.pool.get(key) is None
+    deadline = asyncio.get_running_loop().time() + 5
+    while ("user", "failed") not in await world.statuses(conv):
+        assert asyncio.get_running_loop().time() < deadline, "the cut turn hangs"
+        await asyncio.sleep(0.02)
+    again = await api.http.post(f"/sessions/{key}/close", headers=HEADERS)
+    assert again.status_code == 404
+    gone = await api.http.post("/processes/999999999/kill", headers=HEADERS)
+    assert gone.status_code == 404
+
+    ScriptedClient.hold = asyncio.Event()
+    await world.conversations.post(conv, "once more")
+    await asyncio.sleep(0.3)
+    interrupted = await api.http.post(f"/sessions/{key}/interrupt", headers=weak)
+    assert interrupted.json() == {"key": key, "interrupted": True}
+    assert ScriptedClient.instances[-1].interrupted
+    kinds = [r["kind"] for r in (await api.get("/audit"))["records"]]
+    assert kinds == ["session_interrupt", "session_close"]
+
+
 async def test_tokens_and_audit_need_admin_scope(world: World) -> None:
     api = Api(world)
     api.store.grant("api-only", "weak", scope="api")

tests/test_processes.py

@@ -0,0 +1,92 @@
+import os
+import subprocess
+import sys
+import time
+
+from beaver_gateway.backends.processes import Processes, _command, subtree
+
+SLEEP = "import time; time.sleep(60)"
+NEST = (
+    "import subprocess, sys, time; "
+    f"subprocess.Popen([sys.executable, '-c', {SLEEP!r}]); time.sleep(60)"
+)
+
+
+def spawn(code: str) -> subprocess.Popen[bytes]:
+    return subprocess.Popen([sys.executable, "-c", code])
+
+
+def rows_under(processes: Processes, pid: int, count: int) -> list[dict]:
+    deadline = time.monotonic() + 5
+    while time.monotonic() < deadline:
+        rows = subtree(processes.tree_parents_first(), pid)
+        if len(rows) >= count:
+            return rows
+        time.sleep(0.05)
+    msg = f"{pid} never grew to {count} processes"
+    raise AssertionError(msg)
+
+
+def test_the_tree_shows_what_the_gateway_started_and_what_it_costs() -> None:
+    child = spawn(SLEEP)
+    try:
+        processes = Processes()
+        (row,) = rows_under(processes, child.pid, 1)
+        assert row["parent"] == os.getpid()
+        assert row["depth"] == 0
+        assert row["rss"] > 0
+        assert row["cpu"] is None
+        assert "time.sleep" in row["command"]
+        (again,) = rows_under(processes, child.pid, 1)
+        assert again["cpu"] is not None
+    finally:
+        child.kill()
+        child.wait()
+
+
+def test_kill_takes_the_branch_and_nothing_the_gateway_did_not_start() -> None:
+    parent = spawn(NEST)
+    try:
+        processes = Processes()
+        top, nested = rows_under(processes, parent.pid, 2)
+        assert (top["depth"], nested["depth"]) == (0, 1)
+        assert not processes.kill_own_descendant(os.getpid())
+        assert not processes.kill_own_descendant(os.getppid())
+        assert processes.kill_own_descendant(parent.pid)
+        parent.wait(timeout=5)
+        alive = {row["pid"] for row in processes.tree_parents_first()}
+        assert not alive & {top["pid"], nested["pid"]}
+    finally:
+        parent.kill()
+        parent.wait()
+
+
+def test_a_subtree_ends_at_the_next_sibling() -> None:
+    rows = [
+        {"pid": 1, "depth": 0},
+        {"pid": 2, "depth": 1},
+        {"pid": 3, "depth": 2},
+        {"pid": 4, "depth": 1},
+        {"pid": 5, "depth": 0},
+    ]
+    assert [(r["pid"], r["depth"]) for r in subtree(rows, 2)] == [(2, 0), (3, 1)]
+    assert [r["pid"] for r in subtree(rows, 1)] == [1, 2, 3, 4]
+    assert subtree(rows, 9) == []
+    assert subtree(rows, None) == []
+
+
+def test_a_command_line_keeps_prompts_and_configs_to_itself() -> None:
+    argv = [
+        "/opt/claude/claude",
+        "--model",
+        "m",
+        "--system-prompt",
+        "you are a beaver",
+        "--mcp-config",
+        '{"gateway": {"headers": {"Authorization": "Bearer abc"}}}',
+        "x" * 100,
+    ]
+    assert _command(argv) == (
+        "claude --model m --system-prompt … --mcp-config … " + "x" * 80 + "…"
+    )
+    assert _command([]) == ""