feat: full-disk persistence — tar entire / filesystem every 60s

Sync everything on disk except virtual (/proc, /sys, /dev),
our persist path (/data), and temp dirs (/tmp, /run).
Interval changed from 120s to 60s, first sync after 30s.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

entrypoint.py

@@ -24,7 +24,7 @@ from datetime import datetime, timezone
 PERSIST_PATH = os.environ.get("PERSIST_PATH", "/data")
 HF_TOKEN = os.environ.get("HF_TOKEN", "")
 HF_DATASET_REPO = os.environ.get("HF_DATASET_REPO", "")
-SYNC_INTERVAL = int(os.environ.get("SYNC_INTERVAL", "120"))
+SYNC_INTERVAL = int(os.environ.get("SYNC_INTERVAL", "60"))
 SSH_PORT = os.environ.get("SSH_PORT", "2222")
 TTYD_PORT = os.environ.get("TTYD_PORT", "7681")
 LOGFILE = "/var/log/huggingrun.log"
@@ -32,36 +32,28 @@ STATE_FILE = os.path.join(PERSIST_PATH, "state.tar.zst")
 PKG_FILE = os.path.join(PERSIST_PATH, "user-packages.list")
 BASE_PKG_FILE = "/etc/base-packages.list"
 
-# Full filesystem persistence — these dirs are synced to/from the dataset.
-# tar -p preserves all permissions, so no corruption.
-PERSIST_DIRS = [
-    "home",
-    "root",
-    "etc",
-    "opt",
-    "var",
-    "usr/local",
-]
-
-# Exclude from tar: Docker-managed, virtual, transient, or regenerable files
+# Full-disk persistence: tar everything from / except virtual/transient dirs.
+# Excludes only things that CANNOT or SHOULD NOT be persisted.
 TAR_EXCLUDES = [
+    # Virtual filesystems (kernel-provided, not real files)
+    "proc",
+    "sys",
+    "dev",
+    # Our own persist path (avoid recursive tar of state.tar.zst)
+    "data",
+    # Temporary
+    "tmp",
+    "run",
     # Docker-managed (overwritten each container start)
     "etc/hostname",
     "etc/hosts",
     "etc/resolv.conf",
     "etc/mtab",
     # Transient runtime
-    "var/run",
     "var/lock",
-    "var/cache/apt",
-    "var/lib/apt/lists",
-    "var/log",          # logs are ephemeral
-    # Sockets, pids, caches
+    # Sockets, pids
     "*.sock",
     "*.pid",
-    "__pycache__",
-    "*.pyc",
-    ".cache",
 ]
 
 
@@ -237,23 +229,22 @@ def save_and_upload():
     except Exception:
         pass
 
-    # Create tar.zst of all persist dirs
+    # Create tar.zst of entire filesystem (full-disk persistence)
     t0 = time.time()
-    dirs_to_persist = [d for d in PERSIST_DIRS if os.path.isdir(f"/{d}")]
-    if not dirs_to_persist:
-        log("  nothing to persist")
-        return
-
     excludes = " ".join(f"--exclude='{e}'" for e in TAR_EXCLUDES)
-    cmd = f"tar --zstd {excludes} -cpf {STATE_FILE} -C / {' '.join(dirs_to_persist)}"
+    cmd = f"tar --zstd {excludes} -cpf {STATE_FILE} -C / ."
     rc, out = run(cmd)
     elapsed = time.time() - t0
     if rc != 0:
-        log(f"  tar failed ({elapsed:.1f}s)")
-        return
+        # tar returns 1 for "file changed as we read it" — that's OK
+        if rc == 1:
+            log(f"  tar: warnings (files changed during archive) ({elapsed:.1f}s)")
+        else:
+            log(f"  tar failed rc={rc} ({elapsed:.1f}s)")
+            return
 
     size_mb = os.path.getsize(STATE_FILE) / 1024 / 1024
-    log(f"  tar: {size_mb:.1f}MB ({elapsed:.1f}s) [{', '.join(dirs_to_persist)}]")
+    log(f"  tar: {size_mb:.1f}MB ({elapsed:.1f}s) [full disk]")
 
     # Upload both files
     api = HfApi(token=HF_TOKEN)
@@ -292,8 +283,8 @@ def save_and_upload():
 
 # ── Sync Thread ───────────────────────────────────────────────────────
 def sync_loop():
-    log("sync thread: waiting 60s before first sync")
-    time.sleep(60)
+    log("sync thread: waiting 30s before first sync")
+    time.sleep(30)
     cycle = 0
     while True:
         cycle += 1