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

This commit is contained in:
hh committed 2026-10-13 01:22:41 +02:00
1 parent 9875f08016
commit 712c24de4e
12 files changed
+996 -127

No files matched your search

+24 -12
View File
@@ -20,6 +20,7 @@ import tempfile
import time import time
import uuid import uuid
import warnings import warnings
from collections import deque
from collections.abc import Mapping from collections.abc import Mapping
from dataclasses import dataclass, field from dataclasses import dataclass, field
from pathlib import Path from pathlib import Path
@@ -49,7 +50,12 @@ from claude_agent_sdk import (
from beaver_gateway.agents import policy as policy_mod from beaver_gateway.agents import policy as policy_mod
from beaver_gateway.agents import prompts as prompt_assembly from beaver_gateway.agents import prompts as prompt_assembly
from beaver_gateway.backends.capture import TurnCapture, TurnUsage from beaver_gateway.backends.capture import TurnCapture, TurnUsage
from beaver_gateway.backends.sessions import Session, SessionClient, SessionPool from beaver_gateway.backends.sessions import (
STDERR_LINES,
Session,
SessionClient,
SessionPool,
)
from beaver_gateway.backends.transcript import ( from beaver_gateway.backends.transcript import (
build_entries, build_entries,
close_open_tool_uses, close_open_tool_uses,
@@ -71,6 +77,7 @@ from beaver_gateway.events.stream import (
build_thinking_delta, build_thinking_delta,
build_tool_use_block_start, build_tool_use_block_start,
) )
from beaver_gateway.security.redact import redact
if TYPE_CHECKING: if TYPE_CHECKING:
from collections.abc import ( from collections.abc import (
@@ -262,12 +269,7 @@ class ClaudeSdkBackend:
await self._pool.close(key) await self._pool.close(key)
async def interrupt(self, key: str) -> bool: async def interrupt(self, key: str) -> bool:
live = self._pool.get(key) return await self._pool.interrupt(key)
if live is None or not live.busy:
return False
live.interrupt_requested = True
await live.client.interrupt()
return True
def live(self, key: str) -> Session | None: def live(self, key: str) -> Session | None:
return self._pool.get(key) return self._pool.get(key)
@@ -674,7 +676,8 @@ class ClaudeSdkBackend:
async def _spawn( async def _spawn(
self, resume: str | None, *, key: str, spec: _SessionSpec self, resume: str | None, *, key: str, spec: _SessionSpec
) -> Session: ) -> Session:
options = self._build_options(resume, key=key, spec=spec) stderr: deque[str] = deque(maxlen=STDERR_LINES)
options = self._build_options(resume, key=key, spec=spec, stderr=stderr)
client = self._factory(options) client = self._factory(options)
await client.connect() await client.connect()
_log.info( _log.info(
@@ -693,6 +696,7 @@ class ClaudeSdkBackend:
session_id=resume, session_id=resume,
resumed=resume is not None, resumed=resume is not None,
pinned=spec.pinned, pinned=spec.pinned,
stderr=stderr,
) )
live.reader = asyncio.create_task(self._read(live)) live.reader = asyncio.create_task(self._read(live))
return live return live
@@ -701,7 +705,12 @@ class ClaudeSdkBackend:
return _RunnerClient(options, uid=self._uid) return _RunnerClient(options, uid=self._uid)
def _build_options( def _build_options(
self, resume: str | None, *, key: str, spec: _SessionSpec self,
resume: str | None,
*,
key: str,
spec: _SessionSpec,
stderr: deque[str] | None = None,
) -> ClaudeAgentOptions: ) -> ClaudeAgentOptions:
agent = self._agent agent = self._agent
opt = agent.options opt = agent.options
@@ -747,11 +756,14 @@ class ClaudeSdkBackend:
plugins=cast("Any", plugins), plugins=cast("Any", plugins),
skills="all" if plugins else None, skills="all" if plugins else None,
max_turns=opt.max_turns, max_turns=opt.max_turns,
stderr=lambda line: _log.warning( stderr=lambda line: self._stderr(line.rstrip(), stderr),
"claude[%s]: %s", agent.name, line.rstrip()
),
) )
def _stderr(self, line: str, tail: deque[str] | None) -> None:
_log.warning("claude[%s]: %s", self._agent.name, line)
if tail is not None:
tail.append(redact(line))
def _can_use_tool( def _can_use_tool(
self, key: str self, key: str
) -> Callable[ ) -> Callable[
+118
View File
@@ -0,0 +1,118 @@
"""What runs under the gateway: the process tree with memory and CPU, and the kill."""
from __future__ import annotations
import contextlib
import os
import time
from dataclasses import dataclass, field
from pathlib import Path
from typing import Any
import psutil
from beaver_gateway.security.redact import redact
__all__ = ["Processes", "subtree"]
_VALUE_HIDDEN = frozenset(
{
"--system-prompt",
"--append-system-prompt",
"--mcp-config",
"--settings",
"--agents",
}
)
_ARG_CLIP = 80
_COMMAND_CLIP = 240
@dataclass(slots=True)
class Processes:
root: int = field(default_factory=os.getpid)
_cpu_seen: dict[int, tuple[float, float]] = field(default_factory=dict)
def tree_parents_first(self) -> list[dict[str, Any]]:
try:
found = psutil.Process(self.root).children(recursive=True)
except psutil.Error:
return []
now = time.monotonic()
rows: dict[int, dict[str, Any]] = {}
for process in found:
with contextlib.suppress(psutil.Error), process.oneshot():
spent = process.cpu_times()
rows[process.pid] = {
"pid": process.pid,
"parent": process.ppid(),
"name": process.name(),
"command": _command(process.cmdline()),
"rss": process.memory_info().rss,
"cpu": self._rate(process.pid, spent.user + spent.system, now),
"age_seconds": round(time.time() - process.create_time(), 1),
}
self._cpu_seen = {p: seen for p, seen in self._cpu_seen.items() if p in rows}
children: dict[int, list[int]] = {}
for pid, row in rows.items():
parent = row["parent"] if row["parent"] in rows else self.root
children.setdefault(parent, []).append(pid)
ordered: list[dict[str, Any]] = []
def walk(parent: int, depth: int) -> None:
for pid in sorted(children.get(parent, ())):
ordered.append({**rows[pid], "depth": depth})
walk(pid, depth + 1)
walk(self.root, 0)
return ordered
def kill_own_descendant(self, pid: int) -> bool:
try:
owned = psutil.Process(self.root).children(recursive=True)
target = next((p for p in owned if p.pid == pid), None)
if target is None:
return False
doomed = [*target.children(recursive=True), target]
except psutil.Error:
return False
for process in doomed:
with contextlib.suppress(psutil.Error):
process.kill()
return True
def _rate(self, pid: int, spent: float, now: float) -> float | None:
before = self._cpu_seen.get(pid)
self._cpu_seen[pid] = (spent, now)
if before is None or now <= before[1]:
return None
return round(100 * max(spent - before[0], 0.0) / (now - before[1]), 1)
def subtree(rows: list[dict[str, Any]], pid: int | None) -> list[dict[str, Any]]:
start = next((i for i, row in enumerate(rows) if row["pid"] == pid), None)
if start is None:
return []
base = rows[start]["depth"]
out = [{**rows[start], "depth": 0}]
for row in rows[start + 1 :]:
if row["depth"] <= base:
break
out.append({**row, "depth": row["depth"] - base})
return out
def _command(argv: list[str]) -> str:
if not argv:
return ""
parts = [Path(argv[0]).name]
hide = False
for arg in argv[1:]:
if hide:
parts.append("…")
hide = False
continue
hide = arg in _VALUE_HIDDEN
parts.append(arg if len(arg) <= _ARG_CLIP else f"{arg[:_ARG_CLIP]}…")
command = redact(" ".join(parts))
return command if len(command) <= _COMMAND_CLIP else f"{command[:_COMMAND_CLIP]}…"
+66 -33
View File
@@ -13,12 +13,15 @@ import contextlib
import logging import logging
import os import os
import time import time
from collections import deque
from dataclasses import dataclass, field from dataclasses import dataclass, field
from pathlib import Path from pathlib import Path
from typing import TYPE_CHECKING, Any, Protocol from typing import TYPE_CHECKING, Any, Protocol
import psutil import psutil
from beaver_gateway.backends.processes import Processes, subtree
if TYPE_CHECKING: if TYPE_CHECKING:
from collections.abc import ( from collections.abc import (
AsyncIterable, AsyncIterable,
@@ -43,6 +46,7 @@ DEFAULT_TTL: Mapping[str, float | None] = {
_EVICT_ORDER = {"fork": 0, "job": 0, "deep": 1, "branch": 2, "master": 3} _EVICT_ORDER = {"fork": 0, "job": 0, "deep": 1, "branch": 2, "master": 3}
_RSS_HEADROOM = 0.8 _RSS_HEADROOM = 0.8
STDERR_LINES = 200
class SessionClient(Protocol): class SessionClient(Protocol):
@@ -86,6 +90,7 @@ class Session:
"""Summary of the last task notification, the pending turn's prompt.""" """Summary of the last task notification, the pending turn's prompt."""
observer: Callable[[Any], None] | None = None observer: Callable[[Any], None] | None = None
"""The last turn's observer; sees subagent traffic between turns.""" """The last turn's observer; sees subagent traffic between turns."""
stderr: deque[str] = field(default_factory=lambda: deque(maxlen=STDERR_LINES))
@property @property
def busy(self) -> bool: def busy(self) -> bool:
@@ -121,6 +126,7 @@ class SessionPool:
self._max_live = max_live self._max_live = max_live
self._ttl = dict(ttl) self._ttl = dict(ttl)
self._reap_interval = reap_interval self._reap_interval = reap_interval
self._processes = Processes()
def __len__(self) -> int: def __len__(self) -> int:
return len(self._sessions) return len(self._sessions)
@@ -168,19 +174,6 @@ class SessionPool:
total += child.memory_info().rss total += child.memory_info().rss
return total return total
@staticmethod
def rss_of(session: Session) -> int | None:
pid = session.pid
if pid is None:
return None
try:
process = psutil.Process(pid)
return process.memory_info().rss + sum(
c.memory_info().rss for c in process.children(recursive=True)
)
except psutil.Error:
return None
def over_limit(self) -> bool: def over_limit(self) -> bool:
if self._rss_limit is not None: if self._rss_limit is not None:
return self.rss() > self._rss_limit * _RSS_HEADROOM return self.rss() > self._rss_limit * _RSS_HEADROOM
@@ -214,11 +207,21 @@ class SessionPool:
_log.info("closing session %s (%s, %s)", session.session_id, session.kind, key) _log.info("closing session %s (%s, %s)", session.session_id, session.kind, key)
if session.reader is not None and session.reader is not asyncio.current_task(): if session.reader is not None and session.reader is not asyncio.current_task():
session.reader.cancel() session.reader.cancel()
if session.attached:
session.inbox.put_nowait(RuntimeError(f"session {key} was closed"))
try: try:
await session.client.disconnect() await session.client.disconnect()
except Exception: # noqa: BLE001 except Exception: # noqa: BLE001
_log.exception("disconnect failed for session %s", session.session_id) _log.exception("disconnect failed for session %s", session.session_id)
async def interrupt(self, key: str) -> bool:
session = self._sessions.get(key)
if session is None or not session.busy:
return False
session.interrupt_requested = True
await session.client.interrupt()
return True
async def close_all(self, *, agent: str | None = None) -> None: async def close_all(self, *, agent: str | None = None) -> None:
for session in list(self._sessions.values()): for session in list(self._sessions.values()):
if agent is None or session.agent == agent: if agent is None or session.agent == agent:
@@ -246,27 +249,57 @@ class SessionPool:
_log.exception("session reaper failed") _log.exception("session reaper failed")
def snapshot(self) -> list[dict[str, Any]]: def snapshot(self) -> list[dict[str, Any]]:
return self.census()["sessions"]
def census(self) -> dict[str, Any]:
now = time.monotonic() now = time.monotonic()
return [ rows = self._processes.tree_parents_first()
{ owned: set[int] = set()
"key": s.key, sessions: list[dict[str, Any]] = []
"agent": s.agent, for s in self._sessions.values():
"kind": s.kind, tree = subtree(rows, s.pid)
"session_id": s.session_id, owned.update(row["pid"] for row in tree)
"pid": s.pid, sessions.append(
"rss": self.rss_of(s), {
"idle_seconds": round(now - s.last_used, 1), "key": s.key,
"age_seconds": round(now - s.created_at, 1), "agent": s.agent,
"turns": s.turns, "kind": s.kind,
"busy": s.busy, "session_id": s.session_id,
"running_turn": s.running_turn, "pid": s.pid,
"pending": len(s.pending), "rss": sum(row["rss"] for row in tree) if tree else None,
"pending_question": s.pending_question, "cpu": _cpu(tree),
"pinned": s.pinned, "idle_seconds": round(now - s.last_used, 1),
"dirty": s.dirty, "age_seconds": round(now - s.created_at, 1),
} "turns": s.turns,
for s in self._sessions.values() "busy": s.busy,
] "running_turn": s.running_turn,
"pending": len(s.pending),
"pending_question": s.pending_question,
"pinned": s.pinned,
"dirty": s.dirty,
"processes": tree,
}
)
return {
"rss": sum(row["rss"] for row in rows),
"rss_limit": self._rss_limit,
"cpu": _cpu(rows),
"sessions": sessions,
"others": [row for row in rows if row["pid"] not in owned],
}
async def kill(self, pid: int) -> bool:
if not self._processes.kill_own_descendant(pid):
return False
for session in list(self._sessions.values()):
if session.pid == pid:
await self.close(session.key)
return True
def _cpu(rows: list[dict[str, Any]]) -> float | None:
rates = [row["cpu"] for row in rows if row["cpu"] is not None]
return round(sum(rates), 1) if rates else None
def cgroup_limit() -> int | None: def cgroup_limit() -> int | None:
+66 -6
View File
@@ -55,6 +55,7 @@ if TYPE_CHECKING:
from collections.abc import AsyncIterator, Iterable, Sequence from collections.abc import AsyncIterator, Iterable, Sequence
from pathlib import Path from pathlib import Path
from beaver_gateway.backends.sessions import Session, SessionPool
from beaver_gateway.conversations.service import Conversations from beaver_gateway.conversations.service import Conversations
from beaver_gateway.frontends.base import GatewayRuntime from beaver_gateway.frontends.base import GatewayRuntime
from beaver_gateway.jobs.scheduler import Job, Scheduler from beaver_gateway.jobs.scheduler import Job, Scheduler
@@ -633,15 +634,74 @@ def build_app( # noqa: PLR0915
await require_token(request, runtime, scope=SCOPE) await require_token(request, runtime, scope=SCOPE)
return _sse(runtime, conversation_id=None) return _sse(runtime, conversation_id=None)
def pool_of() -> SessionPool:
if runtime.pool is None:
raise HTTPException(status.HTTP_404_NOT_FOUND, "no session pool")
return runtime.pool
def session_of(key: str) -> Session:
session = pool_of().get(key)
if session is None:
raise HTTPException(status.HTTP_404_NOT_FOUND, f"no live session {key}")
return session
@app.get("/sessions") @app.get("/sessions")
async def sessions(request: Request) -> dict[str, Any]: async def sessions(request: Request) -> dict[str, Any]:
await require_token(request, runtime, scope=SCOPE) await require_token(request, runtime, scope=SCOPE)
pool = runtime.pool if runtime.pool is None:
return { return {
"rss": pool.rss() if pool is not None else None, "rss": None,
"rss_limit": pool.rss_limit if pool is not None else None, "rss_limit": None,
"sessions": pool.snapshot() if pool is not None else [], "cpu": None,
} "sessions": [],
"others": [],
}
return runtime.pool.census()
@app.get("/sessions/{key}/stderr")
async def session_stderr(key: str, request: Request) -> dict[str, Any]:
await require_token(request, runtime, scope=ADMIN_SCOPE)
return {"key": key, "lines": list(session_of(key).stderr)}
@app.post("/sessions/{key}/interrupt")
async def session_interrupt(key: str, request: Request) -> dict[str, Any]:
token = await require_token(request, runtime, scope=SCOPE)
session = session_of(key)
interrupted = await pool_of().interrupt(key)
await audit.log(
runtime,
actor=f"token:{token}",
kind="session_interrupt",
agent_name=session.agent,
session=key,
interrupted=interrupted,
)
return {"key": key, "interrupted": interrupted}
@app.post("/sessions/{key}/close")
async def session_close(key: str, request: Request) -> dict[str, Any]:
token = await require_token(request, runtime, scope=ADMIN_SCOPE)
session = session_of(key)
await pool_of().close(key)
await audit.log(
runtime,
actor=f"token:{token}",
kind="session_close",
agent_name=session.agent,
session=key,
busy=session.busy,
)
return {"key": key, "closed": True}
@app.post("/processes/{pid}/kill")
async def process_kill(pid: int, request: Request) -> dict[str, Any]:
token = await require_token(request, runtime, scope=ADMIN_SCOPE)
if not await pool_of().kill(pid):
raise HTTPException(
status.HTTP_404_NOT_FOUND, f"no process {pid} under the gateway"
)
await audit.log(runtime, actor=f"token:{token}", kind="process_kill", pid=pid)
return {"pid": pid, "killed": True}
@app.get("/schedules") @app.get("/schedules")
async def schedules(request: Request) -> dict[str, Any]: async def schedules(request: Request) -> dict[str, Any]:
+48
View File
@@ -396,6 +396,54 @@ async def test_describe_snapshots_open_tools_and_records_rate_limit(
assert page["offset"] == 0 and len(page["entries"]) == 1 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: async def test_tokens_and_audit_need_admin_scope(world: World) -> None:
api = Api(world) api = Api(world)
api.store.grant("api-only", "weak", scope="api") api.store.grant("api-only", "weak", scope="api")
+92
View File
@@ -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([]) == ""
+16
View File
@@ -313,6 +313,22 @@ export class ApiClient {
return this.get("/api/sessions"); return this.get("/api/sessions");
} }
sessionStderr(key: string): Promise<{ key: string; lines: string[] }> {
return this.get(`/api/sessions/${encodeURIComponent(key)}/stderr`);
}
interruptSession(key: string): Promise<{ interrupted: boolean }> {
return this.post(`/api/sessions/${encodeURIComponent(key)}/interrupt`);
}
closeSession(key: string): Promise<{ closed: boolean }> {
return this.post(`/api/sessions/${encodeURIComponent(key)}/close`);
}
killProcess(pid: number): Promise<{ killed: boolean }> {
return this.post(`/api/processes/${pid}/kill`);
}
jobs(): Promise<JobsResponse> { jobs(): Promise<JobsResponse> {
return this.get("/api/jobs"); return this.get("/api/jobs");
} }
+18
View File
@@ -106,7 +106,9 @@ export interface ContentBlock {
export interface HistoryMessage { export interface HistoryMessage {
content: string | ContentBlock[]; content: string | ContentBlock[];
id?: string | null;
role: "user" | "assistant"; role: "user" | "assistant";
service?: string;
// When the first block of this message was written; absent for old rows. // When the first block of this message was written; absent for old rows.
ts?: string | null; ts?: string | null;
} }
@@ -144,17 +146,31 @@ export interface AgentsResponse {
mcps: { name: string; kind: string }[]; mcps: { name: string; kind: string }[];
} }
export interface ProcessRow {
age_seconds: number;
command: string;
cpu: number | null;
depth: number;
name: string;
parent: number;
pid: number;
rss: number;
}
export interface SessionRow { export interface SessionRow {
age_seconds: number; age_seconds: number;
agent: string; agent: string;
busy: boolean; busy: boolean;
cpu: number | null;
dirty: boolean; dirty: boolean;
idle_seconds: number; idle_seconds: number;
key: string; key: string;
kind: string; kind: string;
pending: number;
pending_question: boolean; pending_question: boolean;
pid: number | null; pid: number | null;
pinned: boolean; pinned: boolean;
processes: ProcessRow[];
rss: number | null; rss: number | null;
running_turn: string | null; running_turn: string | null;
session_id: string | null; session_id: string | null;
@@ -162,6 +178,8 @@ export interface SessionRow {
} }
export interface SessionsResponse { export interface SessionsResponse {
cpu: number | null;
others: ProcessRow[];
rss: number | null; rss: number | null;
rss_limit: number | null; rss_limit: number | null;
sessions: SessionRow[]; sessions: SessionRow[];
+6
View File
@@ -68,6 +68,12 @@ export function fmtSeconds(seconds: number | null | undefined): string {
: fmtDuration(seconds * THOUSAND); : fmtDuration(seconds * THOUSAND);
} }
export function fmtCpu(percent: number | null | undefined): string {
return percent === null || percent === undefined
? "–"
: `${Math.round(percent)}%`;
}
export function fmtBytes(bytes: number | null | undefined): string { export function fmtBytes(bytes: number | null | undefined): string {
if (bytes === null || bytes === undefined) { if (bytes === null || bytes === undefined) {
return "–"; return "–";
+68
View File
@@ -0,0 +1,68 @@
<script lang="ts">
import type { ProcessRow } from "$lib/api/types";
import { Button } from "$lib/components/ui/button";
import { fmtBytes, fmtCpu, fmtSeconds } from "$lib/format";
let {
rows,
onKill,
}: { rows: ProcessRow[]; onKill: (row: ProcessRow) => void } = $props();
const INDENT_REM = 0.875;
</script>
<div class="overflow-x-auto">
<table class="w-full text-sm">
<thead class="text-muted-foreground text-xs">
<tr class="border-b text-left">
<th class="py-1.5 pr-3 font-medium">pid</th>
<th class="w-full py-1.5 pr-3 font-medium">command</th>
<th class="py-1.5 pr-3 text-right font-medium">cpu</th>
<th class="py-1.5 pr-3 text-right font-medium">rss</th>
<th class="hidden py-1.5 pr-3 text-right font-medium sm:table-cell">
age
</th>
<th class="py-1.5"><span class="sr-only">actions</span></th>
</tr>
</thead>
<tbody>
{#each rows as row (row.pid)}
<tr class="border-b last:border-b-0">
<td class="tabular py-1 pr-3 text-muted-foreground text-xs">
{row.pid}
</td>
<td class="max-w-0 py-1 pr-3">
<span
class="block truncate font-mono text-xs"
title={row.command}
style:padding-left="{row.depth * INDENT_REM}rem"
>
{row.command || row.name}
</span>
</td>
<td class="tabular py-1 pr-3 text-right text-xs">
{fmtCpu(row.cpu)}
</td>
<td class="tabular whitespace-nowrap py-1 pr-3 text-right text-xs">
{fmtBytes(row.rss)}
</td>
<td
class="tabular hidden whitespace-nowrap py-1 pr-3 text-right text-muted-foreground text-xs sm:table-cell"
>
{fmtSeconds(row.age_seconds)}
</td>
<td class="py-0.5 text-right">
<Button
class="text-muted-foreground hover:text-destructive max-sm:h-10"
onclick={() => onKill(row)}
size="xs"
variant="ghost"
>
Kill
</Button>
</td>
</tr>
{/each}
</tbody>
</table>
</div>
+150
View File
@@ -0,0 +1,150 @@
<script lang="ts">
import { onMount, untrack } from "svelte";
import { base } from "$app/paths";
import type { ApiClient } from "$lib/api/client";
import type { ProcessRow, SessionRow } from "$lib/api/types";
import { Button } from "$lib/components/ui/button";
import { shortId } from "$lib/format";
import { ConversationFeed } from "$lib/panel/conversation.svelte";
import TurnCard from "$lib/panel/turn-card.svelte";
import ProcessTable from "./process-table.svelte";
let {
client,
row,
busy,
note,
onInterrupt,
onClose,
onKill,
}: {
client: ApiClient;
row: SessionRow;
busy: boolean;
note: string;
onInterrupt: (row: SessionRow) => void;
onClose: (row: SessionRow) => void;
onKill: (process: ProcessRow) => void;
} = $props();
const TICK_MS = 1000;
const feed = new ConversationFeed(
() => client,
untrack(() => row.key)
);
let stderr = $state<string[] | null>(null);
let now = $state(Date.now());
const turn = $derived(feed.model.running);
onMount(() => {
feed.start();
const timer = setInterval(() => {
now = Date.now();
}, TICK_MS);
return () => {
feed.stop();
clearInterval(timer);
};
});
$effect(() => {
const { key, idle_seconds: polled } = row;
if (polled >= 0) {
client
.sessionStderr(key)
.then((tail) => {
stderr = tail.lines;
})
.catch(() => undefined);
}
});
</script>
<div class="flex flex-col gap-4 px-3 pt-2 pb-4">
<div class="flex flex-wrap items-center gap-x-4 gap-y-2">
<p class="tabular min-w-0 flex-1 text-muted-foreground text-xs">
session
{shortId(row.session_id)}
{#if row.pid}
· pid {row.pid}
{/if}
{#if row.pending > 0}
· {row.pending}
buffered
{/if}
</p>
<div class="flex flex-wrap items-center gap-2">
{#if note}
<span class="text-muted-foreground text-xs" role="status">{note}</span>
{/if}
{#if !feed.error}
<Button
class="max-sm:h-10"
href="{base}/conversations/{row.key}"
size="sm"
variant="ghost"
>
Open thread
</Button>
{/if}
<Button
class="max-sm:h-10"
disabled={busy || !row.busy}
onclick={() => onInterrupt(row)}
size="sm"
variant="outline"
>
Interrupt turn
</Button>
<Button
class="max-sm:h-10"
disabled={busy}
onclick={() => onClose(row)}
size="sm"
variant="destructive"
>
Close session
</Button>
</div>
</div>
<section class="flex flex-col gap-1">
<h4 class="label-quiet">Now</h4>
{#if turn}
<TurnCard {now} {turn} />
{:else}
<p class="text-muted-foreground text-sm">
{row.busy
? "A turn is running; its events have not reached this page yet."
: "Nothing in flight."}
</p>
{/if}
</section>
<section class="flex flex-col gap-1">
<h4 class="label-quiet">Process tree</h4>
{#if row.processes.length === 0}
<p class="text-muted-foreground text-sm">
The process is not visible to the gateway.
</p>
{:else}
<ProcessTable {onKill} rows={row.processes} />
{/if}
</section>
<section class="flex flex-col gap-1">
<h4 class="label-quiet">stderr</h4>
{#if stderr?.length === 0}
<p class="text-muted-foreground text-sm">
Nothing on stderr since this process started.
</p>
{:else if stderr}
<pre
class="max-h-64 overflow-auto whitespace-pre-wrap break-all rounded-md bg-muted/40 p-2 text-xs"
>{stderr.join(
"\n"
)}</pre>
{/if}
</section>
</div>
+324 -76
View File
@@ -1,16 +1,39 @@
<script lang="ts"> <script lang="ts">
import ChevronRightIcon from "@lucide/svelte/icons/chevron-right";
import { onMount } from "svelte"; import { onMount } from "svelte";
import { base } from "$app/paths"; import type {
import type { SessionsResponse } from "$lib/api/types"; ProcessRow,
SessionRow,
SessionsResponse,
} from "$lib/api/types";
import ErrorNote from "$lib/components/error-note.svelte"; import ErrorNote from "$lib/components/error-note.svelte";
import { fmtBytes, fmtSeconds, shortId } from "$lib/format"; import { Button } from "$lib/components/ui/button";
import * as Dialog from "$lib/components/ui/dialog";
import { fmtBytes, fmtCpu, fmtSeconds, shortId } from "$lib/format";
import { gateway } from "$lib/gateway.svelte";
import { session } from "$lib/session.svelte"; import { session } from "$lib/session.svelte";
import KindMark from "$lib/shell/kind-mark.svelte"; import KindMark from "$lib/shell/kind-mark.svelte";
import ProcessTable from "$lib/system/process-table.svelte";
import SessionDetail from "$lib/system/session-detail.svelte";
import { cn } from "$lib/utils"; import { cn } from "$lib/utils";
const POLL_MS = 10_000; interface Order {
confirm: string;
lines: string[];
run: () => Promise<unknown>;
title: string;
}
const POLL_MS = 3000;
const SECOND_LOOK_FOR_CPU_MS = 1000;
const COLUMNS = 8;
let sessions = $state<SessionsResponse | null>(null); let sessions = $state<SessionsResponse | null>(null);
let failure = $state<string | null>(null); let failure = $state<string | null>(null);
let unfolded = $state<Set<string>>(new Set());
let order = $state<Order | null>(null);
let orderFailure = $state<string | null>(null);
let notes = $state<Record<string, string>>({});
let busy = $state(false);
async function load() { async function load() {
const { client } = session; const { client } = session;
@@ -25,90 +48,315 @@
} }
} }
function toggle(key: string) {
const next = new Set(unfolded);
if (!next.delete(key)) {
next.add(key);
}
unfolded = next;
}
function ask(next: Order) {
orderFailure = null;
order = next;
}
async function carryOut(pending: Order) {
busy = true;
try {
await pending.run();
order = null;
await load();
} catch (cause) {
orderFailure = cause instanceof Error ? cause.message : String(cause);
} finally {
busy = false;
}
}
async function interrupt(row: SessionRow) {
const { client } = session;
if (!client) {
return;
}
busy = true;
try {
const { interrupted } = await client.interruptSession(row.key);
notes = {
...notes,
[row.key]: interrupted ? "" : "Nothing was running.",
};
await load();
} catch (cause) {
const message = cause instanceof Error ? cause.message : String(cause);
notes = { ...notes, [row.key]: message };
} finally {
busy = false;
}
}
function askClose(row: SessionRow) {
const { client } = session;
if (!client) {
return;
}
const lines = [
"The claude process exits. The conversation stays and resumes from its transcript on the next message.",
];
if (row.busy) {
lines.push("A turn is running: it is cut and its reply is lost.");
}
if (row.pinned) {
lines.push("This session is pinned: the gateway never closes it itself.");
}
ask({
confirm: "Close session",
lines,
run: () => client.closeSession(row.key),
title: `Close the ${row.kind} session?`,
});
}
function askKill(process: ProcessRow) {
const { client } = session;
if (!client) {
return;
}
const owner = sessions?.sessions.find((row) => row.pid === process.pid);
const lines = [
`SIGKILL to ${process.name} and everything it started. Nothing gets to clean up.`,
];
if (owner) {
lines.push(
"It is a session's claude: the session is dropped, and the conversation resumes from its transcript on the next message."
);
}
ask({
confirm: "Kill",
lines,
run: () => client.killProcess(process.pid),
title: `Kill process ${process.pid}?`,
});
}
const titles = $derived(
new Map(gateway.conversations.map((row) => [row.id, row.title]))
);
onMount(() => { onMount(() => {
load(); load();
const second = setTimeout(load, SECOND_LOOK_FOR_CPU_MS);
const poll = setInterval(load, POLL_MS); const poll = setInterval(load, POLL_MS);
return () => clearInterval(poll); return () => {
clearTimeout(second);
clearInterval(poll);
};
}); });
</script> </script>
<section class="flex flex-col gap-3"> {#snippet state(s: SessionRow)}
<div class="flex items-baseline gap-3"> <span class={cn(s.busy ? "text-signal" : "text-muted-foreground")}>
<h2 class="font-medium text-sm">Live claude processes</h2> {s.busy ? "busy" : "idle"}
{#if sessions} </span>
<span class="tabular text-muted-foreground text-xs"> {#if s.pinned}
{sessions.sessions.length} <span class="ml-1 text-muted-foreground">pinned</span>
· {fmtBytes(sessions.rss)} {/if}
{#if sessions.rss_limit} {#if s.dirty}
of {fmtBytes(sessions.rss_limit)} <span class="ml-1 text-warn">dirty</span>
{/if} {/if}
</span> {#if s.pending_question}
{/if} <span class="ml-1 text-attention-foreground">question</span>
</div> {/if}
{/snippet}
<div class="flex flex-wrap items-baseline gap-3">
<h2 class="font-medium text-sm">Live claude processes</h2>
{#if sessions}
<span class="tabular text-muted-foreground text-xs">
{sessions.sessions.length}
{sessions.sessions.length === 1 ? "session" : "sessions"}
· {fmtBytes(sessions.rss)}
{#if sessions.rss_limit}
of {fmtBytes(sessions.rss_limit)}
{/if}
· cpu {fmtCpu(sessions.cpu)}
</span>
{/if}
</div>
<div class="flex flex-col gap-6">
{#if failure} {#if failure}
<ErrorNote message={failure} retry={load} /> <ErrorNote message={failure} retry={load} />
{:else if !sessions} {:else if !sessions}
<p class="text-muted-foreground text-sm">Loading…</p> <p class="text-muted-foreground text-sm">Loading…</p>
{:else if sessions.sessions.length === 0}
<p class="text-muted-foreground text-sm">
No claude processes alive. Sessions spawn on the first turn and are reaped
by idle time or memory pressure.
</p>
{:else} {:else}
<div class="overflow-x-auto rounded-lg border bg-strip"> <section class="flex flex-col gap-2">
<table class="w-full text-sm"> <h3 class="label-quiet">Sessions</h3>
<thead class="text-muted-foreground text-xs"> {#if sessions.sessions.length === 0}
<tr class="border-b text-left"> <p class="text-muted-foreground text-sm">
<th class="px-3 py-2 font-medium">conversation</th> No claude processes alive. Sessions spawn on the first turn and are
<th class="px-3 py-2 font-medium">agent</th> reaped by idle time or memory pressure.
<th class="px-3 py-2 font-medium">state</th> </p>
<th class="px-3 py-2 text-right font-medium">rss</th> {:else}
<th class="px-3 py-2 text-right font-medium">idle</th> <div class="overflow-x-auto rounded-lg border bg-strip">
<th class="px-3 py-2 text-right font-medium">age</th> <table class="w-full text-sm">
<th class="px-3 py-2 text-right font-medium">turns</th> <thead class="text-muted-foreground text-xs">
</tr> <tr class="border-b text-left">
</thead> <th class="px-3 py-2 font-medium">conversation</th>
<tbody> <th class="hidden px-3 py-2 font-medium sm:table-cell">
{#each sessions.sessions as s (s.key)} agent
<tr class="row-hover border-b last:border-b-0"> </th>
<td class="px-3 py-1.5"> <th class="hidden px-3 py-2 font-medium sm:table-cell">
<a state
class="flex items-center gap-2 hover:underline" </th>
href="{base}/conversations/{s.key}" <th class="px-3 py-2 text-right font-medium">cpu</th>
<th class="px-3 py-2 text-right font-medium">rss</th>
<th
class="hidden px-3 py-2 text-right font-medium sm:table-cell"
> >
<KindMark kind={s.kind} /> idle
<span class="tabular text-xs">{shortId(s.key)}</span> </th>
</a> <th
</td> class="hidden px-3 py-2 text-right font-medium sm:table-cell"
<td class="px-3 py-1.5">{s.agent}</td>
<td class="px-3 py-1.5 text-xs">
<span
class={cn(s.busy ? "text-signal" : "text-muted-foreground")}
> >
{s.busy ? "busy" : "idle"} age
</span> </th>
{#if s.pinned} <th
<span class="ml-1 text-muted-foreground">pinned</span> class="hidden px-3 py-2 text-right font-medium sm:table-cell"
>
turns
</th>
</tr>
</thead>
<tbody>
{#each sessions.sessions as s (s.key)}
{@const expanded = unfolded.has(s.key)}
<tr class={cn("border-b", !expanded && "last:border-b-0")}>
<td class="max-w-0 px-1.5 py-1 sm:w-2/5">
<button
aria-expanded={expanded}
class="row-hover flex max-w-full items-center gap-2 rounded-md px-1.5 py-1 text-left max-sm:min-h-10"
onclick={() => toggle(s.key)}
type="button"
>
<ChevronRightIcon
class={cn(
"size-3.5 text-icon transition-transform duration-150",
expanded && "rotate-90"
)}
/>
<KindMark kind={s.kind} />
<span class="flex min-w-0 flex-col">
<span class="truncate font-medium">
{titles.get(s.key) ?? shortId(s.key)}
</span>
<span class="text-xs sm:hidden">
{@render state(s)}
</span>
</span>
</button>
</td>
<td class="hidden px-3 py-1.5 sm:table-cell">{s.agent}</td>
<td
class="hidden whitespace-nowrap px-3 py-1.5 text-xs sm:table-cell"
>
{@render state(s)}
</td>
<td class="tabular px-3 py-1.5 text-right">
{fmtCpu(s.cpu)}
</td>
<td class="tabular whitespace-nowrap px-3 py-1.5 text-right">
{fmtBytes(s.rss)}
</td>
<td
class="tabular hidden px-3 py-1.5 text-right sm:table-cell"
>
{fmtSeconds(s.idle_seconds)}
</td>
<td
class="tabular hidden px-3 py-1.5 text-right sm:table-cell"
>
{fmtSeconds(s.age_seconds)}
</td>
<td
class="tabular hidden px-3 py-1.5 text-right sm:table-cell"
>
{s.turns}
</td>
</tr>
{#if expanded && session.client}
<tr class="border-b last:border-b-0">
<td class="bg-background/60" colspan={COLUMNS}>
<SessionDetail
{busy}
client={session.client}
note={notes[s.key] ?? ""}
onClose={askClose}
onInterrupt={interrupt}
onKill={askKill}
row={s}
/>
</td>
</tr>
{/if} {/if}
{#if s.dirty} {/each}
<span class="ml-1 text-warn">dirty</span> </tbody>
{/if} </table>
{#if s.pending_question} </div>
<span class="ml-1 text-attention-foreground">question</span> {/if}
{/if} </section>
</td>
<td class="tabular px-3 py-1.5 text-right">{fmtBytes(s.rss)}</td> <section class="flex flex-col gap-2">
<td class="tabular px-3 py-1.5 text-right"> <h3 class="label-quiet">Outside any session</h3>
{fmtSeconds(s.idle_seconds)} {#if sessions.others.length === 0}
</td> <p class="text-muted-foreground text-sm">
<td class="tabular px-3 py-1.5 text-right"> Nothing else runs under the gateway.
{fmtSeconds(s.age_seconds)} </p>
</td> {:else}
<td class="tabular px-3 py-1.5 text-right">{s.turns}</td> <ProcessTable onKill={askKill} rows={sessions.others} />
</tr> {/if}
{/each} </section>
</tbody>
</table>
</div>
{/if} {/if}
</section> </div>
<Dialog.Root
onOpenChange={(next) => {
if (!next) {
order = null;
}
}}
open={order !== null}
>
<Dialog.Content>
{#if order}
{@const pending = order}
<Dialog.Header>
<Dialog.Title>{pending.title}</Dialog.Title>
{#each pending.lines as line (line)}
<Dialog.Description>{line}</Dialog.Description>
{/each}
{#if orderFailure}
<p class="text-destructive text-sm" role="alert">{orderFailure}</p>
{/if}
</Dialog.Header>
<Dialog.Footer>
<Button
disabled={busy}
onclick={() => {
order = null;
}}
variant="ghost"
>
Cancel
</Button>
<Button
disabled={busy}
onclick={() => carryOut(pending)}
variant="destructive"
>
{pending.confirm}
</Button>
</Dialog.Footer>
{/if}
</Dialog.Content>
</Dialog.Root>