Update hf-sync-manager.mjs

hf-sync-manager.mjs

@@ -1,16 +1,19 @@
 // ============================================================
-// HF Sync Manager - Full Mirror Edition
+// HF Sync Manager - Full Mirror + Emergency Upload
 //
-// Mirrors ~/.openclaw/ to HF Dataset every 30 min.
+// CRITICAL FIX for HF Spaces:
+// On HF free tier, OpenClaw “full process restart” = complete
+// container restart = /tmp cleared. /tmp-based markers don’t work.
 //
-// Critical rules:
-// 1. NEVER restore openclaw.json from Dataset.
-//    It is owned by setup-hf-config.mjs only.
-// 2. RUNNING_FLAG is never deleted.
-//    /tmp is only cleared on container restart, not gateway restart.
-//    This correctly distinguishes the two restart types.
+// Solution:
+// 1. Include openclaw.json in BOTH upload and restore.
+//    setup.mjs will patch (not overwrite) it, so no restart loop.
+// 2. SIGTERM handler: when container is dying, immediately upload
+//    openclaw.json to Dataset before process exits. This preserves
+//    user’s just-saved settings for the next container start.
+// 3. Regular 30-min sync for workspace files.
 //
-// Excluded: openclaw.json, sessions, qmd, canvas, *.bak
+// Excluded: sessions (transcripts), qmd (vector index), canvas
 // ============================================================
 
 import fs   from "node:fs";
@@ -26,17 +29,16 @@ var HF_TOKEN     = (process.env.HF_TOKEN      || "").trim();
 var DATASET_ID   = (process.env.HF_DATASET_ID  || "").trim();
 var REPO_DIR     = "/tmp/oc-dataset";
 var INTERVAL     = 30 * 60 * 1000;
-var RUNNING_FLAG = "/tmp/.hf-sync-running";
 
+// Excluded from sync (runtime-only, auto-rebuilt by OpenClaw)
 var EXCLUDE_NAMES = [
-"openclaw.json",
-"openclaw.json.bak",
-"sessions",
-"qmd",
-"canvas",
-".git"
+"sessions",           // agents/*/sessions/ - session transcripts
+"qmd",               // agents/*/qmd/      - vector search sqlite index
+"canvas",            // UI runtime files
+".git",
+"openclaw.json.bak"
 ];
-var EXCLUDE_EXT = [".bak", ".tmp", ".log"];
+var EXCLUDE_EXT = [".tmp", ".log"];
 
 // –– helpers ———————————————–
 
@@ -51,11 +53,11 @@ return "https://user:" + HF_TOKEN +
 "@huggingface.co/datasets/" + DATASET_ID;
 }
 
-function git(args) {
+function git(args, timeoutMs) {
 return execSync("git " + args, {
 cwd:     REPO_DIR,
 stdio:   "pipe",
-timeout: 120000,
+timeout: timeoutMs || 120000,
 env:     Object.assign({}, process.env, { GIT_TERMINAL_PROMPT: "0" })
 }).toString().trim();
 }
@@ -71,19 +73,27 @@ if (name.indexOf(EXCLUDE_EXT[i]) >= 0) return true;
 return false;
 }
 
-// –– git ––––––––––––––––––––––––––
+function isValidJson(p) {
+try {
+if (!fs.existsSync(p)) return false;
+JSON.parse(fs.readFileSync(p, "utf-8").trim());
+return true;
+} catch (e) { return false; }
+}
+
+// –– git repo setup ––––––––––––––––––––
 
 function ensureRepo() {
 if (fs.existsSync(path.join(REPO_DIR, ".git"))) return;
 execSync("rm -rf " + REPO_DIR, { stdio: "pipe" });
 try {
 execSync("git clone --depth 1 " + repoUrl() + " " + REPO_DIR, {
-stdio: "pipe", timeout: 120000,
+stdio: "pipe", timeout: 60000,
 env: Object.assign({}, process.env, { GIT_TERMINAL_PROMPT: "0" })
 });
 log("Repo cloned");
 } catch (e) {
-log("Clone failed, init empty: " + e.message);
+log("Clone failed, init: " + e.message);
 fs.mkdirSync(REPO_DIR, { recursive: true });
 execSync("git init " + REPO_DIR, { stdio: "pipe" });
 git("remote add origin " + repoUrl());
@@ -96,24 +106,24 @@ git("config pull.rebase false");
 }
 
 function pull() {
-try { git("fetch --quiet origin"); } catch (e) { /* ignore */ }
-try { git("pull --quiet --no-rebase origin main"); }
+try { git("fetch --quiet origin", 30000); } catch (e) { /* ignore */ }
+try { git("pull --quiet --no-rebase origin main", 30000); }
 catch (e) {
-try { git("pull --quiet --no-rebase origin master"); }
+try { git("pull --quiet --no-rebase origin master", 30000); }
 catch (e2) { /* empty repo */ }
 }
 }
 
-function push() {
+function push(label, timeoutMs) {
 try {
 git("add -A");
 var changed = git("diff --cached --name-only");
 if (!changed) return 0;
 var n  = changed.split("\n").filter(Boolean).length;
 var ts = new Date().toISOString().replace("T", " ").substring(0, 19);
-git('commit -m "sync ' + ts + '"');
-try { git("push --quiet origin HEAD:main"); }
-catch (e) { git("push --quiet origin HEAD:master"); }
+git('commit -m "' + (label || "sync") + " " + ts + '"');
+try { git("push --quiet origin HEAD:main", timeoutMs || 30000); }
+catch (e) { git("push --quiet origin HEAD:master", timeoutMs || 30000); }
 return n;
 } catch (e) {
 log("Push failed: " + e.message);
@@ -121,7 +131,7 @@ return 0;
 }
 }
 
-// –– mirror ————————————————
+// –– mirror: local -> repo ———————————
 
 function mirrorToRepo(srcDir, dstDir) {
 if (!fs.existsSync(srcDir)) return 0;
@@ -145,6 +155,8 @@ log("Copy error [" + name + "]: " + e.message);
 return count;
 }
 
+// –– mirror: repo -> local ———————��———
+
 function mirrorFromRepo(srcDir, dstDir) {
 if (!fs.existsSync(srcDir)) return 0;
 fs.mkdirSync(dstDir, { recursive: true });
@@ -176,57 +188,82 @@ fs.mkdirSync(path.join(WORKSPACE, "memory"), { recursive: true });
 var soul = path.join(WORKSPACE, "SOUL.md");
 if (!fs.existsSync(soul)) {
 fs.writeFileSync(soul, [
-"# Soul",
-"",
-"You are a helpful, warm, concise AI assistant.",
-"",
-"## Language",
-"",
+"# Soul", "",
+"You are a helpful, warm, concise AI assistant.", "",
+"## Language", "",
 "Default language: Simplified Chinese.",
-"Always reply in Chinese unless the user writes in another language first.",
-"",
-"## Tone",
-"",
+"Always reply in Chinese unless the user writes in another language first.", "",
+"## Tone", "",
 "- Natural and friendly, not overly formal",
 "- Concise and to the point",
 "- Ask one clarifying question at a time when needed"
 ].join("\n") + "\n", "utf-8");
-log("Seeded workspace/SOUL.md");
+log("Seeded SOUL.md");
 }
 
 var agents = path.join(WORKSPACE, "AGENTS.md");
 if (!fs.existsSync(agents)) {
 fs.writeFileSync(agents, [
-"# Agent Instructions",
-"",
-"## Boot sequence",
-"",
+"# Agent Instructions", "",
+"## Boot sequence", "",
 "1. Read SOUL.md - language and persona",
 "2. Read USER.md if present - user profile",
 "3. Read MEMORY.md - long-term facts and rules",
-"4. Read today and yesterday memory/YYYY-MM-DD.md if present",
-"",
-"## Memory rules",
-"",
+"4. Read today and yesterday memory/YYYY-MM-DD.md if present", "",
+"## Memory rules", "",
 "- Write important facts to MEMORY.md when asked",
 "- Log daily context to memory/YYYY-MM-DD.md",
 "- Do not ask for information already provided"
 ].join("\n") + "\n", "utf-8");
-log("Seeded workspace/AGENTS.md");
+log("Seeded AGENTS.md");
 }
 
 var mem = path.join(WORKSPACE, "MEMORY.md");
 if (!fs.existsSync(mem)) {
 fs.writeFileSync(mem, [
-"# Long-term Memory",
-"",
+"# Long-term Memory", "",
 "<!-- OpenClaw writes important facts here. -->",
 "<!-- Loaded at the start of every session. -->"
 ].join("\n") + "\n", "utf-8");
-log("Seeded workspace/MEMORY.md");
+log("Seeded MEMORY.md");
 }
 }
 
+// –– EMERGENCY upload (SIGTERM handler) ––––––––––
+// When OpenClaw triggers “full process restart”, the container dies.
+// We have ~10s before SIGKILL to upload openclaw.json to Dataset.
+// This preserves the user’s just-saved settings for the next boot.
+
+function emergencyUploadConfig() {
+log("SIGTERM: emergency config upload starting…");
+var localCfg = path.join(OPENCLAW_DIR, "openclaw.json");
+if (!isValidJson(localCfg)) {
+log("SIGTERM: config invalid, skip");
+return;
+}
+try {
+ensureRepo();
+pull();
+var repoOC = path.join(REPO_DIR, "openclaw");
+fs.mkdirSync(repoOC, { recursive: true });
+fs.copyFileSync(localCfg, path.join(repoOC, "openclaw.json"));
+var n = push("emergency-config", 8000);
+log("SIGTERM: emergency upload " + (n > 0 ? "done (" + n + " files)" : "no changes"));
+} catch (e) {
+log("SIGTERM: emergency upload failed: " + e.message);
+}
+}
+
+process.on("SIGTERM", function() {
+emergencyUploadConfig();
+process.exit(0);
+});
+
+// SIGINT for local testing
+process.on("SIGINT", function() {
+process.exit(0);
+});
+
 // –– boot restore ——————————————
 
 function bootRestore() {
@@ -249,7 +286,10 @@ return;
 var repoOC = path.join(REPO_DIR, "openclaw");
 if (fs.existsSync(repoOC)) {
 var n = mirrorFromRepo(repoOC, OPENCLAW_DIR);
-log("Boot restore done: " + n + " file(s) restored (openclaw.json excluded)");
+log("Boot restore done: " + n + " file(s) restored to " + OPENCLAW_DIR);
+if (fs.existsSync(path.join(OPENCLAW_DIR, "openclaw.json"))) {
+log("Config restored from Dataset (setup will patch env.vars only)");
+}
 } else {
 log("No data in Dataset yet - starting fresh");
 }
@@ -277,21 +317,9 @@ if (fs.existsSync(repoOC)) {
 fs.mkdirSync(repoOC, { recursive: true });
 
 var n = mirrorToRepo(OPENCLAW_DIR, repoOC);
+log("Mirrored " + n + " file(s)");
 
-// Upload openclaw.json with JSON validation (excluded from restore, not from upload)
-var localCfg = path.join(OPENCLAW_DIR, "openclaw.json");
-if (fs.existsSync(localCfg)) {
-  try {
-    JSON.parse(fs.readFileSync(localCfg, "utf-8").trim());
-    fs.copyFileSync(localCfg, path.join(repoOC, "openclaw.json"));
-    n++;
-    log("openclaw.json uploaded (JSON valid)");
-  } catch (e) {
-    log("openclaw.json skipped (invalid JSON)");
-  }
-}
-
-var pushed = push();
+var pushed = push("sync");
 if (pushed > 0) {
   log("Sync done: " + pushed + " file(s) pushed");
 } else {
@@ -308,7 +336,6 @@ log("Sync failed (retry in 30 min): " + e.message);
 log("Starting…");
 log("OPENCLAW_DIR : " + OPENCLAW_DIR);
 log("DATASET      : " + (DATASET_ID || "NOT SET"));
-log("RUNNING_FLAG : " + RUNNING_FLAG);
 
 fs.mkdirSync(OPENCLAW_DIR, { recursive: true });
 fs.mkdirSync(WORKSPACE,    { recursive: true });
@@ -320,19 +347,7 @@ execSync("git config --global http.postBuffer 52428800",     { stdio: "pipe" });
 execSync("git config --global pull.rebase false",            { stdio: "pipe" });
 } catch (e) { /* non-fatal */ }
 
-var isGatewayRestart = fs.existsSync(RUNNING_FLAG);
-
-if (isGatewayRestart) {
-log("Gateway restart - skipping boot restore (data already on disk)");
-} else {
-log("Container start - running full boot restore");
 bootRestore();
-}
-
-// Write flag. NEVER delete it on exit.
-// Only container restart clears /tmp.
-fs.writeFileSync(RUNNING_FLAG, String(process.pid), "utf-8");
-log("Running flag written");
 
 if (!HF_TOKEN || !DATASET_ID) {
 log("Sync disabled: set HF_TOKEN and HF_DATASET_ID in Secrets");