Files
openchamber/packages/web/server/lib/opencode/managed-process-registry.js
T
Bohdan Triapitsyn 2ff5428c69 feat(opencode): never leave orphaned OpenCode server processes
OpenChamber spawns the OpenCode server as an external child binary (detached
on Unix), so a hard crash, SIGKILL, or Ctrl+C of the host before graceful
teardown could leave it running. Orphaned servers then accumulate and contend
on the shared SQLite DB, causing severe startup slowdowns.

Add a per-process registry plus a startup reaper, mirroring the pattern
OpenCode's own CLI daemon uses for its detached server:

- One file per spawned process at
  ~/.config/openchamber/managed-opencode/<pid>.json. Per-process files avoid
  the read-modify-write clobber race between concurrent runtimes/windows that a
  single shared file would suffer.
- On spawn, record the child (pid, owner pid, port, binary, host runtime).
- On graceful close/restart, delete the record.
- On startup, reap only our own, verified, genuinely-orphaned processes:
  recorded by us AND still a live `opencode serve` on the recorded port AND
  whose spawner is provably gone (reparented to pid 1, or recorded owner dead).
  It never touches a process a live instance is using, the user's standalone
  server, the official desktop app, or the TUI.

Wire it into every runtime that spawns the server:

- web/desktop via the OpenCode lifecycle (register on spawn, unregister on
  close/restart, reap at startup). The restart-for-config-change flow inherits
  this automatically through the same kill/spawn paths.
- VS Code carries a parity implementation (it does not bundle the web package)
  that reads/writes the same registry directory and uses the same algorithm.
- Tag the actual host runtime (desktop/web/ssh-remote/vscode) for observability.

Also tighten teardown so the registry stays accurate and orphans die promptly
instead of only on the next start:

- The web server now also handles SIGHUP and SIGUSR2 (terminal close and the
  nodemon restart used by dev:server:watch / dev:web:hmr).
- Electron now installs SIGINT/SIGTERM/SIGHUP handlers that run the same
  background teardown as a normal quit, covering Ctrl+C on electron:dev.

External OpenCode servers (OPENCODE_SKIP_START) are intentionally excluded: we
never manage or kill processes we did not spawn.
2026-06-24 16:51:17 +03:00

252 lines
9.1 KiB
JavaScript

// Managed OpenCode process registry + orphan reaper.
//
// OpenChamber spawns the OpenCode server as an EXTERNAL child binary (on Unix
// with `detached: true`, so it leads its own process group). That binary can
// therefore outlive its parent if the parent is hard-killed/crashes/`Ctrl+C`ed
// before graceful teardown runs — leaving an orphaned `opencode serve` that
// then contends on the shared SQLite DB and slows everything down.
//
// We cannot tie an arbitrary external binary to the parent's death portably
// (Electron's `utilityProcess` would, but it only runs JS entrypoints, not a
// standalone binary). So we use the same pattern OpenCode's own CLI daemon uses
// for its detached server: an on-disk record of the pids WE spawned, plus a
// startup reaper that kills ONLY our own, verified, genuinely-orphaned
// processes — never a process a live instance (another desktop window, a VS
// Code host, the user's standalone `opencode`) is actively using.
//
// Storage: ONE FILE PER SPAWNED PROCESS in a registry directory, named
// `<childPid>.json`. Multiple runtimes (web/desktop/VS Code) and multiple
// windows all run concurrently; a single shared JSON file would be corrupted by
// the read-modify-write race (last writer wins, clobbering another instance's
// entry). Per-process files mean every instance only ever writes/deletes its
// OWN file, so there is no write contention at all.
//
// Safety model (why this never kills the wrong thing):
// 1. The reaper only ever considers pids THIS product recorded. The user's
// standalone CLI server, the official desktop app, and the TUI are never
// recorded, so they are never even candidates.
// 2. Before killing, it re-verifies the live pid is still an `opencode serve`
// matching the recorded port (guards against the OS recycling a dead pid
// onto an unrelated process).
// 3. It kills only when the spawning owner is provably gone — the child has
// been reparented to init/pid 1, or the recorded owner pid is dead. A
// child still owned by a live instance is left untouched.
//
// The VS Code extension cannot import this module (it does not bundle the web
// package); it carries a parity implementation that reads/writes the SAME dir.
import fs from 'node:fs';
import os from 'node:os';
import path from 'node:path';
import { spawnSync } from 'node:child_process';
const resolveRegistryDir = () => {
const override = process.env.OPENCHAMBER_MANAGED_PROCESS_REGISTRY;
if (override && override.trim()) return override.trim();
return path.join(os.homedir(), '.config', 'openchamber', 'managed-opencode');
};
const entryFilePath = (pid) => path.join(resolveRegistryDir(), `${pid}.json`);
const writeEntryFile = (entry) => {
const dir = resolveRegistryDir();
try {
fs.mkdirSync(dir, { recursive: true });
const filePath = path.join(dir, `${entry.pid}.json`);
const tmp = `${filePath}.tmp-${process.pid}`;
fs.writeFileSync(tmp, JSON.stringify(entry, null, 2));
fs.renameSync(tmp, filePath);
} catch {
// Best-effort: a failed registry write must never break spawn/shutdown.
}
};
const readAllEntries = () => {
const dir = resolveRegistryDir();
let names = [];
try {
names = fs.readdirSync(dir).filter((name) => name.endsWith('.json'));
} catch {
return [];
}
const out = [];
for (const name of names) {
const filePath = path.join(dir, name);
try {
const entry = JSON.parse(fs.readFileSync(filePath, 'utf8'));
if (entry && Number.isInteger(entry.pid)) {
out.push({ entry, filePath });
} else {
fs.rmSync(filePath, { force: true });
}
} catch {
// Corrupt/partial file — drop it.
try { fs.rmSync(filePath, { force: true }); } catch {}
}
}
return out;
};
/** Record an OpenCode process WE spawned so a future run can reap it if orphaned. */
export const registerManagedProcess = ({ pid, ownerPid, port, binary, runtime } = {}) => {
if (!Number.isInteger(pid)) return;
writeEntryFile({
pid,
ownerPid: Number.isInteger(ownerPid) ? ownerPid : process.pid,
port: Number.isInteger(port) ? port : null,
binary: typeof binary === 'string' ? binary : null,
runtime: typeof runtime === 'string' ? runtime : 'web',
startedAt: new Date().toISOString(),
});
};
/** Drop a pid from the registry (after we have killed/closed it ourselves). */
export const unregisterManagedProcess = (pid) => {
if (!Number.isInteger(pid)) return;
try {
fs.rmSync(entryFilePath(pid), { force: true });
} catch {
}
};
const isPidAlive = (pid) => {
if (!Number.isInteger(pid)) return false;
try {
process.kill(pid, 0);
return true;
} catch (error) {
// EPERM = process exists but we lack permission to signal it → still alive.
return error?.code === 'EPERM';
}
};
const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));
// Returns { ppid, command } for a live pid on Unix, or null if it can't be read.
const readUnixProcInfo = (pid) => {
try {
const result = spawnSync('ps', ['-p', String(pid), '-o', 'ppid=,command='], {
encoding: 'utf8',
timeout: 3000,
windowsHide: true,
});
const line = (result.stdout || '').trim();
if (!line) return null;
const match = line.match(/^\s*(\d+)\s+(.*)$/);
if (!match) return null;
return { ppid: Number.parseInt(match[1], 10), command: match[2] };
} catch {
return null;
}
};
// Windows image name for a pid (e.g. "opencode.exe"), or null.
const readWindowsImageName = (pid) => {
try {
const result = spawnSync('tasklist', ['/FI', `PID eq ${pid}`, '/FO', 'CSV', '/NH'], {
encoding: 'utf8',
timeout: 3000,
windowsHide: true,
});
return (result.stdout || '').trim() || null;
} catch {
return null;
}
};
const commandIdentifiesOurServer = (command, entry) => {
if (typeof command !== 'string') return false;
const lower = command.toLowerCase();
if (!lower.includes('opencode') || !lower.includes('serve')) return false;
// Tie to the exact server we registered when we know its port, so a recycled
// pid running a *different* opencode server is never mistaken for ours.
if (Number.isInteger(entry.port) && !command.includes(String(entry.port))) return false;
return true;
};
const killOrphan = async (pid) => {
if (process.platform === 'win32') {
try {
spawnSync('taskkill', ['/PID', String(pid), '/T', '/F'], { stdio: 'ignore', timeout: 5000, windowsHide: true });
} catch {
}
return;
}
const signalTree = (signal) => {
try { process.kill(-pid, signal); } catch {}
try { process.kill(pid, signal); } catch {}
};
signalTree('SIGTERM');
for (let waited = 0; waited < 1500 && isPidAlive(pid); waited += 150) {
await sleep(150);
}
if (isPidAlive(pid)) {
signalTree('SIGKILL');
await sleep(300);
}
};
// Decide+act on a single registry entry. Returns true if it was reaped.
const processEntry = async (entry, { log }) => {
// Dead pid → nothing to do (caller drops the file).
if (!isPidAlive(entry.pid)) return false;
const ownerGone = Number.isInteger(entry.ownerPid) && !isPidAlive(entry.ownerPid);
if (process.platform === 'win32') {
const image = readWindowsImageName(entry.pid);
const looksLikeOpencode = typeof image === 'string' && image.toLowerCase().includes('opencode');
// Windows lacks reliable reparent-to-1 semantics (job objects usually kill
// children with the parent), so we reap only when the owner is provably dead
// AND the image still looks like opencode.
if (looksLikeOpencode && ownerGone) {
await killOrphan(entry.pid);
log?.(`[lifecycle] reaped orphaned OpenCode pid ${entry.pid} (owner ${entry.ownerPid} gone)`);
return true;
}
return false;
}
const info = readUnixProcInfo(entry.pid);
// Can't verify identity (or it's not our server) → leave it alone.
if (!info || !commandIdentifiesOurServer(info.command, entry)) return false;
const orphaned = info.ppid === 1 || ownerGone;
if (!orphaned) return false; // still owned by a live instance
await killOrphan(entry.pid);
log?.(`[lifecycle] reaped orphaned OpenCode pid ${entry.pid} (reparented/owner gone)`);
return true;
};
/**
* Kill any genuinely-orphaned OpenCode processes WE previously spawned, and
* prune their registry files. Safe to call at startup before spawning a new
* server. Returns { inspected, reaped }.
*/
export const reapOrphanedProcesses = async ({ log } = {}) => {
const records = readAllEntries();
if (records.length === 0) return { inspected: 0, reaped: 0 };
let reaped = 0;
for (const { entry, filePath } of records) {
let drop = false;
try {
const wasReaped = await processEntry(entry, { log });
if (wasReaped) reaped += 1;
// Drop the file when the process is gone (reaped now, or already dead);
// keep it only while the process is still alive and owned by a live owner.
drop = wasReaped || !isPidAlive(entry.pid);
} catch (error) {
log?.(`[lifecycle] reap check failed for pid ${entry.pid}: ${error?.message ?? error}`);
}
if (drop) {
try { fs.rmSync(filePath, { force: true }); } catch {}
}
}
return { inspected: records.length, reaped };
};