@linxin666/dsh-client-ui-task-board 0.2.7 → 0.2.8

package/lib/index.js

@@ -852,8 +852,48 @@ function cloneTasks(tasks) {
 function hasOpenExecution(task) {
 	return task.executions.some((execution) => execution.endedAt === void 0);
 }
+/**
+* Process states that are dead but still occupy the PID table: `Z` (zombie)
+* and `X` (dead, being reaped). `process.kill(pid, 0)` reports such PIDs as
+* alive, so a crash leftover whose child was never reaped would otherwise be
+* mistaken for a live owner and block ledger startup forever.
+*/
+const DEAD_STATES = /* @__PURE__ */ new Set(["Z", "X"]);
+/**
+* Best-effort single-letter process state ('R','S','D','Z',...) or undefined
+* when no probe is available on this platform. Linux reads /proc/<pid>/stat
+* directly (no subprocess); other POSIX shells out to `ps -o stat=`; Windows
+* has no zombie state, so it returns undefined and the kill(0) probe alone
+* is authoritative there.
+*/
+function processState(pid) {
+	if (process.platform === "linux") try {
+		const stat = readFileSync(`/proc/${pid}/stat`, "utf8");
+		const end = stat.lastIndexOf(")");
+		if (end === -1) return void 0;
+		return stat.slice(end + 2).split(" ")[0] || void 0;
+	} catch {
+		return;
+	}
+	if (process.platform === "win32") return void 0;
+	try {
+		const probe = spawnSync("ps", [
+			"-o",
+			"stat=",
+			"-p",
+			String(pid)
+		], { timeout: PROCESS_PROBE_TIMEOUT_MS });
+		if (probe.status !== 0 || probe.stdout.length === 0) return void 0;
+		const state = probe.stdout.toString("utf8").trim();
+		return state.length > 0 ? state[0] : void 0;
+	} catch {
+		return;
+	}
+}
 function processIsAlive(pid) {
 	if (!Number.isSafeInteger(pid) || pid <= 0) return false;
+	const state = processState(pid);
+	if (state !== void 0 && DEAD_STATES.has(state)) return false;
 	try {
 		process.kill(pid, 0);
 		return true;
@@ -865,6 +905,28 @@ const PROCESS_PROBE_TIMEOUT_MS = 3e3;
 let ownStartTime;
 let ownStartTimeResolved = false;
 /**
+* Exact process start time (Unix epoch ms) on Linux, read straight from
+* /proc (field 22 = start ticks since boot, btime = boot epoch seconds).
+* No subprocess and no rounding, so the recorded `startedAt` from a previous
+* boot compares exactly against the live process identity.
+*/
+function linuxStartTimeMs(pid) {
+	try {
+		const stat = readFileSync(`/proc/${pid}/stat`, "utf8");
+		const end = stat.lastIndexOf(")");
+		if (end === -1) return void 0;
+		const ticks = Number(stat.slice(end + 2).split(" ")[19]);
+		if (!Number.isFinite(ticks)) return void 0;
+		const bootMatch = /^btime\s+(\d+)/m.exec(readFileSync("/proc/stat", "utf8"));
+		if (bootMatch === null) return void 0;
+		const btime = Number(bootMatch[1]);
+		if (!Number.isFinite(btime)) return void 0;
+		return btime * 1e3 + ticks * 1e3 / 100;
+	} catch {
+		return;
+	}
+}
+/**
 * Best-effort start time (Unix epoch ms) of a live process. Used to prove
 * whether the ledger lock really belongs to the PID recorded in it, so a
 * crash leftover whose PID was reused by an unrelated process (issue #786)
@@ -872,6 +934,7 @@ let ownStartTimeResolved = false;
 * undefined when the platform probe is unavailable; callers fail closed.
 */
 function processStartTimeMs(pid) {
+	if (process.platform === "linux") return linuxStartTimeMs(pid);
 	if (process.platform === "win32") {
 		const probe = spawnSync("powershell", [
 			"-NoProfile",
@@ -924,6 +987,21 @@ function ownProcessStartTimeMs() {
 	}
 	return ownStartTime;
 }
+/**
+* Bounded tolerance for legacy lock records. Locks written before the
+* ms-precise probe recorded `startedAt` from `ps -o lstart=` at whole-second
+* resolution; probing the SAME live process exactly (via /proc) then differs
+* in the sub-second remainder. Treating that as PID reuse would steal a live
+* owner's lock during a rolling upgrade and start a second ledger writer.
+* Records written by the ms-precise probe carry `probe: 'exact'` and are
+* compared strictly; anything else (older locks, second-granularity probes)
+* falls back to this bounded tolerance.
+*/
+const LEGACY_START_TOLERANCE_MS = 2e3;
+/** Whether the recorded start time proves the recorded PID is another process. */
+function startTimeMismatch(recorded, actual, exact) {
+	return exact ? recorded !== actual : Math.abs(recorded - actual) > LEGACY_START_TOLERANCE_MS;
+}
 function betterExecution(a, b) {
 	if (a.endedAt === void 0 && b.endedAt !== void 0) return b;
 	if (b.endedAt === void 0 && a.endedAt !== void 0) return a;
@@ -1369,10 +1447,13 @@ var HostTaskLedger = class {
 	acquireLock() {
 		for (let attempt = 0; attempt < 2; attempt += 1) try {
 			const fd = openSync(this.lockFile, "wx", 384);
+			const startedAt = ownProcessStartTimeMs();
+			const probe = process.platform === "linux" || process.platform === "win32" ? "exact" : "legacy";
 			writeFileSync(fd, JSON.stringify({
 				pid: process.pid,
 				token: this.lockToken,
-				startedAt: ownProcessStartTimeMs()
+				startedAt,
+				probe
 			}), { encoding: "utf8" });
 			fsyncSync(fd);
 			try {
@@ -1383,23 +1464,25 @@ var HostTaskLedger = class {
 			if (error.code !== "EEXIST") throw error;
 			let pid;
 			let ownerStartedAt;
+			let ownerExact = false;
 			try {
 				const owner = JSON.parse(readFileSync(this.lockFile, "utf8"));
 				if (typeof owner.pid === "number") pid = owner.pid;
 				if (typeof owner.startedAt === "number") ownerStartedAt = owner.startedAt;
+				ownerExact = owner.probe === "exact";
 			} catch {
-				throw new Error(`task-board ledger lock is unreadable: ${this.lockFile}`);
+				throw new Error(`task-board ledger lock is unreadable: ${this.lockFile}; if this is a leftover from an unclean shutdown and no other DSH host is running, remove it manually and retry`);
 			}
 			if (pid !== void 0 && processIsAlive(pid)) {
 				const actualStartedAt = pid === process.pid ? ownProcessStartTimeMs() : processStartTimeMs(pid);
-				if (!(actualStartedAt !== void 0 && (ownerStartedAt !== void 0 ? actualStartedAt !== ownerStartedAt : (() => {
+				if (!(actualStartedAt !== void 0 && (ownerStartedAt !== void 0 ? startTimeMismatch(ownerStartedAt, actualStartedAt, ownerExact) : (() => {
 					try {
 						return statSync(this.lockFile).mtimeMs < actualStartedAt;
 					} catch {
 						return true;
 					}
 				})()))) {
-					const hint = ownerStartedAt !== void 0 && actualStartedAt === ownerStartedAt ? "" : `; if this PID was reused after a crash and no other DSH host is running, remove ${this.lockFile} manually and retry`;
+					const hint = ownerStartedAt !== void 0 && actualStartedAt !== void 0 && !startTimeMismatch(ownerStartedAt, actualStartedAt, ownerExact) ? "" : `; if this PID was reused after a crash and no other DSH host is running, remove ${this.lockFile} manually and retry`;
 					throw new Error(`task-board ledger is already owned by process ${pid}${hint}`);
 				}
 			}

package/lib/types/host-ledger.d.ts

@@ -9,6 +9,15 @@ export interface OpenedRun {
     task: TaskRecord;
     execution: ExecutionRecord;
 }
+/**
+ * Best-effort single-letter process state ('R','S','D','Z',...) or undefined
+ * when no probe is available on this platform. Linux reads /proc/<pid>/stat
+ * directly (no subprocess); other POSIX shells out to `ps -o stat=`; Windows
+ * has no zombie state, so it returns undefined and the kill(0) probe alone
+ * is authoritative there.
+ */
+export declare function processState(pid: number): string | undefined;
+export declare function processIsAlive(pid: number): boolean;
 export declare class HostTaskLedger {
     private readonly now;
     private document;

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@linxin666/dsh-client-ui-task-board",
   "description": "Host-authoritative task board for the DSH Web GUI with real session execution, Host cron scheduling, and optional cross-platform idle-sleep protection; mounted without DSH source changes.",
-  "version": "0.2.7",
+  "version": "0.2.8",
   "type": "module",
   "main": "lib/index.js",
   "types": "lib/types/index.d.ts",
@@ -22,6 +22,9 @@
     "./package.json": "./package.json"
   },
   "dsh": {
+    "engines": {
+      "dsh": ">=0.1.1-rc.1"
+    },
     "bundle": {
       "patch": "./cordis.patch.yml"
     },
@@ -35,8 +38,8 @@
     }
   },
   "peerDependencies": {
-    "@deepseek-ai/dsh-agent": "^0.1.1-rc.1",
-    "@deepseek-ai/dsh-commands": "^0.1.1-rc.1",
+    "@deepseek-ai/dsh-agent": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-commands": "^0.1.1-rc.2",
     "react": "^18.2.0"
   },
   "dependencies": {
@@ -44,17 +47,17 @@
   },
   "devDependencies": {
     "@deepseek-ai/cordis": "^4.0.1",
-    "@deepseek-ai/dsh-agent": "^0.1.1-rc.1",
-    "@deepseek-ai/dsh-client-connection": "^0.1.1-rc.1",
-    "@deepseek-ai/dsh-client-locale": "^0.1.1-rc.1",
-    "@deepseek-ai/dsh-client-runtime": "^0.1.1-rc.1",
-    "@deepseek-ai/dsh-client-ui-settings": "^0.1.1-rc.1",
-    "@deepseek-ai/dsh-client-ui-slots": "^0.1.1-rc.1",
-    "@deepseek-ai/dsh-commands": "^0.1.1-rc.1",
-    "@deepseek-ai/dsh-host-apiproxy": "^0.1.1-rc.1",
-    "@deepseek-ai/dsh-host-webserver": "^0.1.1-rc.1",
-    "@deepseek-ai/dsh-settings": "^0.1.1-rc.1",
-    "@deepseek-ai/dsh-system-prompt": "^0.1.1-rc.1",
+    "@deepseek-ai/dsh-agent": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-client-connection": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-client-locale": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-client-runtime": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-client-ui-settings": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-client-ui-slots": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-commands": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-host-apiproxy": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-host-webserver": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-settings": "^0.1.1-rc.2",
+    "@deepseek-ai/dsh-system-prompt": "^0.1.1-rc.2",
     "@types/react": "~18.3.1",
     "@types/react-dom": "^18.3.5",
     "@types/node": "^22.20.0",

package/src/host-ledger.ts

@@ -59,8 +59,47 @@ function hasOpenExecution(task: TaskRecord): boolean {
   return task.executions.some(execution => execution.endedAt === undefined)
 }
 
-function processIsAlive(pid: number): boolean {
+/**
+ * Process states that are dead but still occupy the PID table: `Z` (zombie)
+ * and `X` (dead, being reaped). `process.kill(pid, 0)` reports such PIDs as
+ * alive, so a crash leftover whose child was never reaped would otherwise be
+ * mistaken for a live owner and block ledger startup forever.
+ */
+const DEAD_STATES = new Set(['Z', 'X'])
+
+/**
+ * Best-effort single-letter process state ('R','S','D','Z',...) or undefined
+ * when no probe is available on this platform. Linux reads /proc/<pid>/stat
+ * directly (no subprocess); other POSIX shells out to `ps -o stat=`; Windows
+ * has no zombie state, so it returns undefined and the kill(0) probe alone
+ * is authoritative there.
+ */
+export function processState(pid: number): string | undefined {
+  if (process.platform === 'linux') {
+    try {
+      const stat = readFileSync(`/proc/${pid}/stat`, 'utf8')
+      const end = stat.lastIndexOf(')')
+      if (end === -1) return undefined
+      return stat.slice(end + 2).split(' ')[0] || undefined
+    } catch {
+      return undefined // no such process (or unreadable)
+    }
+  }
+  if (process.platform === 'win32') return undefined
+  try {
+    const probe = spawnSync('ps', ['-o', 'stat=', '-p', String(pid)], { timeout: PROCESS_PROBE_TIMEOUT_MS })
+    if (probe.status !== 0 || probe.stdout.length === 0) return undefined
+    const state = probe.stdout.toString('utf8').trim()
+    return state.length > 0 ? state[0] : undefined
+  } catch {
+    return undefined
+  }
+}
+
+export function processIsAlive(pid: number): boolean {
   if (!Number.isSafeInteger(pid) || pid <= 0) return false
+  const state = processState(pid)
+  if (state !== undefined && DEAD_STATES.has(state)) return false
   try {
     process.kill(pid, 0)
     return true
@@ -74,6 +113,29 @@ const PROCESS_PROBE_TIMEOUT_MS = 3000
 let ownStartTime: number | undefined
 let ownStartTimeResolved = false
 
+/**
+ * Exact process start time (Unix epoch ms) on Linux, read straight from
+ * /proc (field 22 = start ticks since boot, btime = boot epoch seconds).
+ * No subprocess and no rounding, so the recorded `startedAt` from a previous
+ * boot compares exactly against the live process identity.
+ */
+function linuxStartTimeMs(pid: number): number | undefined {
+  try {
+    const stat = readFileSync(`/proc/${pid}/stat`, 'utf8')
+    const end = stat.lastIndexOf(')')
+    if (end === -1) return undefined
+    const ticks = Number(stat.slice(end + 2).split(' ')[19])
+    if (!Number.isFinite(ticks)) return undefined
+    const bootMatch = /^btime\s+(\d+)/m.exec(readFileSync('/proc/stat', 'utf8'))
+    if (bootMatch === null) return undefined
+    const btime = Number(bootMatch[1])
+    if (!Number.isFinite(btime)) return undefined
+    return btime * 1000 + (ticks * 1000) / 100 // USER_HZ is 100 on Linux
+  } catch {
+    return undefined
+  }
+}
+
 /**
  * Best-effort start time (Unix epoch ms) of a live process. Used to prove
  * whether the ledger lock really belongs to the PID recorded in it, so a
@@ -82,6 +144,7 @@ let ownStartTimeResolved = false
  * undefined when the platform probe is unavailable; callers fail closed.
  */
 function processStartTimeMs(pid: number): number | undefined {
+  if (process.platform === 'linux') return linuxStartTimeMs(pid)
   if (process.platform === 'win32') {
     const probe = spawnSync(
       'powershell',
@@ -93,8 +156,8 @@ function processStartTimeMs(pid: number): number | undefined {
     const started = Number(probe.stdout.toString('utf8').trim())
     return Number.isFinite(started) ? started : undefined
   }
-  // POSIX: ps lstart with a forced English locale, falling back to the
-  // elapsed-seconds column when lstart cannot be parsed.
+  // Other POSIX (macOS...): ps lstart with a forced English locale, falling
+  // back to the elapsed-seconds column when lstart cannot be parsed.
   const env = { ...process.env, LC_ALL: 'C' }
   const probe = spawnSync('ps', ['-o', 'lstart=', '-p', String(pid)], { timeout: PROCESS_PROBE_TIMEOUT_MS, env })
   if (probe.status === 0 && probe.stdout.length > 0) {
@@ -116,6 +179,23 @@ function ownProcessStartTimeMs(): number | undefined {
   return ownStartTime
 }
 
+/**
+ * Bounded tolerance for legacy lock records. Locks written before the
+ * ms-precise probe recorded `startedAt` from `ps -o lstart=` at whole-second
+ * resolution; probing the SAME live process exactly (via /proc) then differs
+ * in the sub-second remainder. Treating that as PID reuse would steal a live
+ * owner's lock during a rolling upgrade and start a second ledger writer.
+ * Records written by the ms-precise probe carry `probe: 'exact'` and are
+ * compared strictly; anything else (older locks, second-granularity probes)
+ * falls back to this bounded tolerance.
+ */
+const LEGACY_START_TOLERANCE_MS = 2000
+
+/** Whether the recorded start time proves the recorded PID is another process. */
+function startTimeMismatch(recorded: number, actual: number, exact: boolean): boolean {
+  return exact ? recorded !== actual : Math.abs(recorded - actual) > LEGACY_START_TOLERANCE_MS
+}
+
 function betterExecution(a: ExecutionRecord, b: ExecutionRecord): ExecutionRecord {
   if (a.endedAt === undefined && b.endedAt !== undefined) return b
   if (b.endedAt === undefined && a.endedAt !== undefined) return a
@@ -560,7 +640,13 @@ export class HostTaskLedger {
     for (let attempt = 0; attempt < 2; attempt += 1) {
       try {
         const fd = openSync(this.lockFile, 'wx', 0o600)
-        writeFileSync(fd, JSON.stringify({ pid: process.pid, token: this.lockToken, startedAt: ownProcessStartTimeMs() }), { encoding: 'utf8' })
+        const startedAt = ownProcessStartTimeMs()
+        // Linux /proc and Windows PowerShell probes are ms-precise; locks they
+        // write are compared strictly. Other POSIX probes (ps) stay
+        // second-granularity, so their records are compared with the bounded
+        // legacy tolerance.
+        const probe = process.platform === 'linux' || process.platform === 'win32' ? 'exact' : 'legacy'
+        writeFileSync(fd, JSON.stringify({ pid: process.pid, token: this.lockToken, startedAt, probe }), { encoding: 'utf8' })
         fsyncSync(fd)
         try { chmodSync(this.lockFile, 0o600) } catch { /* Windows ACLs own access */ }
         return fd
@@ -569,29 +655,36 @@ export class HostTaskLedger {
         if (code !== 'EEXIST') throw error
         let pid: number | undefined
         let ownerStartedAt: number | undefined
+        let ownerExact = false
         try {
-          const owner = JSON.parse(readFileSync(this.lockFile, 'utf8')) as { pid?: unknown; startedAt?: unknown }
+          const owner = JSON.parse(readFileSync(this.lockFile, 'utf8')) as { pid?: unknown; startedAt?: unknown; probe?: unknown }
           if (typeof owner.pid === 'number') pid = owner.pid
           if (typeof owner.startedAt === 'number') ownerStartedAt = owner.startedAt
+          ownerExact = owner.probe === 'exact'
         } catch {
-          throw new Error(`task-board ledger lock is unreadable: ${this.lockFile}`)
+          // A power-loss mid-write can leave a truncated lock; the same event
+          // killed the writer, so fail closed but explain the recovery.
+          throw new Error(`task-board ledger lock is unreadable: ${this.lockFile}; if this is a leftover from an unclean shutdown and no other DSH host is running, remove it manually and retry`)
         }
         if (pid !== undefined && processIsAlive(pid)) {
           const actualStartedAt = pid === process.pid ? ownProcessStartTimeMs() : processStartTimeMs(pid)
           // A reused PID is exposed when the live process identity no longer
-          // matches the recorded one: either the recorded start time differs,
-          // or (legacy locks without one) the lock file predates the live
-          // process and therefore cannot have been written by it. Takeover is
-          // safe in both cases — the original owner is gone.
+          // matches the recorded one: either the recorded start time differs
+          // beyond the probe's resolution (strict for ms-precise 'exact'
+          // records, a bounded legacy tolerance for old second-granularity
+          // records written by ps), or (legacy locks without a start time)
+          // the lock file predates the live process and therefore cannot
+          // have been written by it. Takeover is safe in both cases — the
+          // original owner is gone.
           const staleReuse = actualStartedAt !== undefined && (
             ownerStartedAt !== undefined
-              ? actualStartedAt !== ownerStartedAt
+              ? startTimeMismatch(ownerStartedAt, actualStartedAt, ownerExact)
               : (() => {
                 try { return statSync(this.lockFile).mtimeMs < actualStartedAt } catch { return true }
               })()
           )
           if (!staleReuse) {
-            const confirmedOwner = ownerStartedAt !== undefined && actualStartedAt === ownerStartedAt
+            const confirmedOwner = ownerStartedAt !== undefined && actualStartedAt !== undefined && !startTimeMismatch(ownerStartedAt, actualStartedAt, ownerExact)
             const hint = confirmedOwner
               ? ''
               : `; if this PID was reused after a crash and no other DSH host is running, remove ${this.lockFile} manually and retry`