fix(update): stop a timed-out git from leaving .git/index.lock behind

subprocess.run's timeout SIGKILLs only the direct git child, so the lock
is left behind and git's helpers are orphaned. On a slow SD card the same
pull then times out on every retry and recreates the lock, so removing it
by hand does not help.

- run_git starts git in its own session, stops the group with SIGTERM
  before SIGKILL, and clears the lock if nothing owns it. Update Code's
  pull, checkout, reset and stash go through it, with a 600s pull timeout.
- Read-only calls (describe --dirty, status, the update-check fetch) set
  GIT_OPTIONAL_LOCKS=0 so a timed-out poll cannot take the lock.
- A timed-out pull returns a clear error; a rebase left by an interrupted
  update is aborted before pulling.
- The manual-fix hint says to stop the services and kill running git.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

test/test_auto_update.py

@@ -823,6 +823,7 @@ def test_core_update_names_the_requirement_files_that_failed():
     pkg.api_v3.plugin_store_manager = None
     failed_install = subprocess.CompletedProcess([], 1, stdout='', stderr='boom')
     with patch.object(pkg.subprocess, 'run', run), \
+            patch.object(system.git_index_lock, 'run_git', run), \
             patch.object(system, '_pip_install_requirements', return_value=failed_install):
         result = system.perform_core_update()
     assert result['status'] == 'success'

test/test_git_index_lock.py

@@ -11,6 +11,7 @@ import subprocess
 import sys
 import time
 from pathlib import Path
+import contextlib
 from unittest.mock import patch
 
 import pytest
@@ -27,6 +28,13 @@ needs_git = pytest.mark.skipif(shutil.which('git') is None, reason='git not inst
 OLD = 1_000_000.0
 
 
+@contextlib.contextmanager
+def _patched(fake):
+    """Fake git for both the plain subprocess calls and the timeout-safe runner."""
+    with patch.object(mod.subprocess, 'run', fake), patch.object(mod.git_index_lock, 'run_git', fake):
+        yield
+
+
 def _checkout(tmp_path, lock_age=None):
     repo = tmp_path / 'LEDMatrix'
     (repo / '.git').mkdir(parents=True)
@@ -173,7 +181,7 @@ def test_update_code_clears_the_lock_before_running_git(client):
         return gil.REMOVED
     calls = []
     with patch.object(gil, 'clear_stale_index_lock', clear), \
-            patch.object(mod.subprocess, 'run', _fake_git(calls)):
+            _patched(_fake_git(calls)):
         data = _update(client)
     assert seen == [str(mod.PROJECT_ROOT)]
     assert data['status'] == 'success'
@@ -182,7 +190,7 @@ def test_update_code_clears_the_lock_before_running_git(client):
 def test_update_code_waits_for_a_git_that_is_still_running(client):
     calls = []
     with patch.object(gil, 'clear_stale_index_lock', lambda d: gil.IN_USE), \
-            patch.object(mod.subprocess, 'run', _fake_git(calls)):
+            _patched(_fake_git(calls)):
         data = _update(client)
     assert data['status'] == 'error'
     assert 'Another git command is running' in data['message']
@@ -193,8 +201,45 @@ def test_a_lock_it_could_not_clear_gets_the_manual_fix(client):
     err = "fatal: Unable to create '/home/ledpi/LEDMatrix/.git/index.lock': File exists.\n"
     calls = []
     with patch.object(gil, 'clear_stale_index_lock', lambda d: gil.UNKNOWN), \
-            patch.object(mod.subprocess, 'run', _fake_git(calls, pull_rc=128, pull_err=err)):
+            _patched(_fake_git(calls, pull_rc=128, pull_err=err)):
         data = _update(client)
     assert data['status'] == 'error'
     assert data['message'].startswith('Update failed: fatal: Unable to create')
     assert 'rm -f ' in data['message'] and 'index.lock' in data['message']
+
+
+# -- run_git: a timed-out git must not leave a lock or an orphaned helper ------
+
+def _lock_holder_script(tmp_path, lock):
+    """A stand-in for a slow git: takes the lock, spawns a helper, then hangs."""
+    script = tmp_path / 'slow.sh'
+    script.write_text(f'#!/bin/sh\n: > "{lock}"\nsleep 60 &\necho $! > "{tmp_path}/helper.pid"\nwait\n')
+    script.chmod(0o755)
+    return script
+
+
+def test_run_git_returns_a_completed_process(tmp_path):
+    result = gil.run_git([sys.executable, '-c', 'print("hi")'], cwd=str(tmp_path), timeout=30)
+    assert (result.returncode, result.stdout.strip()) == (0, 'hi')
+
+
+@pytest.mark.skipif(not os.path.isdir('/proc'), reason='needs /proc')
+def test_run_git_timeout_stops_the_group_and_clears_the_lock(tmp_path, monkeypatch):
+    (tmp_path / '.git').mkdir()
+    lock = tmp_path / '.git' / 'index.lock'
+    script = _lock_holder_script(tmp_path, lock)
+    monkeypatch.setattr(gil, 'MIN_STALE_AGE_SECONDS', 0)
+    monkeypatch.setattr(gil, 'TERM_GRACE_SECONDS', 1)
+    # The scan only sees processes named git, so the stand-in must not match;
+    # the lock is then provably unowned once the group is gone.
+    with pytest.raises(subprocess.TimeoutExpired):
+        gil.run_git([str(script)], cwd=str(tmp_path), timeout=1)
+    assert not lock.exists()
+    helper = int((tmp_path / 'helper.pid').read_text())
+    time.sleep(0.2)
+    with pytest.raises(ProcessLookupError):
+        os.kill(helper, 0)
+
+
+def test_read_only_env_disables_optional_locks():
+    assert gil.read_only_env()['GIT_OPTIONAL_LOCKS'] == '0'

test/test_update_prompts_restart.py

@@ -9,6 +9,7 @@ restart is owed, and the UI raises the existing restart-pending banner.
 import subprocess
 import sys
 from pathlib import Path
+import contextlib
 from unittest.mock import patch
 
 import pytest
@@ -20,6 +21,13 @@ from web_interface.blueprints import api_v3 as mod  # noqa: E402
 from web_interface.blueprints.api_v3 import api_v3  # noqa: E402
 
 
+@contextlib.contextmanager
+def _patched(fake):
+    """Fake git for both the plain subprocess calls and the timeout-safe runner."""
+    with patch.object(mod.subprocess, 'run', fake), patch.object(mod.git_index_lock, 'run_git', fake):
+        yield
+
+
 @pytest.fixture
 def client():
     app = Flask(__name__)
@@ -62,7 +70,7 @@ def _pull(client):
 
 class TestRestartIsRequestedWhenCodeChanged:
     def test_a_pull_that_moved_head_asks_for_a_restart(self, client):
-        with patch.object(mod.subprocess, 'run', _git(['aaa111', 'bbb222'])):
+        with _patched(_git(['aaa111', 'bbb222'])):
             data = _pull(client)
         assert data['status'] == 'success'
         assert data['restart_required'] is True, (
@@ -70,15 +78,14 @@ class TestRestartIsRequestedWhenCodeChanged:
             "nothing told the user to restart")
 
     def test_already_up_to_date_does_not(self, client):
-        with patch.object(mod.subprocess, 'run',
+        with _patched(_git(['aaa111', 'aaa111'], pull_out='Already up to date.\n')):
-                          _git(['aaa111', 'aaa111'], pull_out='Already up to date.\n')):
             data = _pull(client)
         assert data['status'] == 'success'
         assert data['restart_required'] is False, (
             "prompting after a no-op update trains users to ignore the prompt")
 
     def test_a_failed_pull_does_not(self, client):
-        with patch.object(mod.subprocess, 'run', _git(['aaa111'], pull_rc=1)):
+        with _patched(_git(['aaa111'], pull_rc=1)):
             data = _pull(client)
         assert data['status'] == 'error'
         assert data['restart_required'] is False
@@ -100,19 +107,18 @@ class TestUnitsAreRefreshedWithTheCode:
         return calls
 
     def test_an_update_that_moved_head_refreshes_the_units(self, client, refresh):
-        with patch.object(mod.subprocess, 'run', _git(['aaa111', 'bbb222'])):
+        with _patched(_git(['aaa111', 'bbb222'])):
             data = _pull(client)
         assert refresh == [True]
         assert data['unit_refresh']['status'] == 'refreshed'
         assert 'Service settings updated' in data['message']
 
     def test_nothing_new_touches_no_units(self, client, refresh):
-        with patch.object(mod.subprocess, 'run',
+        with _patched(_git(['aaa111', 'aaa111'], pull_out='Already up to date.\n')):
-                          _git(['aaa111', 'aaa111'], pull_out='Already up to date.\n')):
             data = _pull(client)
         assert refresh == [] and data['unit_refresh'] is None
 
     def test_a_failed_pull_touches_no_units(self, client, refresh):
-        with patch.object(mod.subprocess, 'run', _git(['aaa111'], pull_rc=1)):
+        with _patched(_git(['aaa111'], pull_rc=1)):
             data = _pull(client)
         assert refresh == [] and data['unit_refresh'] is None

web_interface/auto_update.py

@@ -302,10 +302,12 @@ def local_changes(project_dir, run=None, timeout=30):
     Returns the changed paths, or None when git could not tell. Raises what
     ``run`` raises (e.g. ``subprocess.TimeoutExpired``).
     """
-    run = run or subprocess.run
+    # A status that times out must not leave index.lock: don't refresh the index.
+    extra = {} if run else {'env': git_index_lock.read_only_env()}
+    run = run or git_index_lock.run_git
     result = run(['git', '-c', 'core.fileMode=false', 'status', '--porcelain',
                   '--untracked-files=no', '-z'],
-                 cwd=str(project_dir), capture_output=True, text=True, timeout=timeout)
+                 cwd=str(project_dir), capture_output=True, text=True, timeout=timeout, **extra)
     if result.returncode != 0:
         return None
     changed = []

web_interface/blueprints/api_v3/__init__.py

@@ -35,6 +35,7 @@ from typing import Dict, Any, Optional, Tuple, Type
 from urllib.parse import urlparse, urlunparse
 logger = logging.getLogger(__name__)
 # Import new infrastructure
+from web_interface import git_index_lock
 from src.web_interface.api_helpers import (success_response, error_response,
                                            exception_error_response, validate_request_json)
 from src.web_interface.errors import ErrorCode
@@ -558,7 +559,8 @@ def get_git_version(project_dir=None):
             capture_output=True,
             text=True,
             timeout=5,
-            cwd=str(project_dir)
+            cwd=str(project_dir),
+            env=git_index_lock.read_only_env()
         )
 
         if result.returncode == 0:

web_interface/blueprints/api_v3/system.py

@@ -14,6 +14,7 @@ from web_interface.blueprints.api_v3 import (
 )
 import threading
 
+from web_interface import git_index_lock
 from web_interface.cache import get_cached, set_cached
 from web_interface.system_metrics import collect_system_metrics, format_uptime
 import web_interface.blueprints.api_v3 as _pkg
@@ -163,6 +164,7 @@ def check_for_update():
             ['git', 'fetch', '--quiet', '--tags', '--force', update_channel.REMOTE,
              update_channel.BETA_BRANCH],
             capture_output=True, timeout=10, cwd=cwd,
+            env=git_index_lock.read_only_env(),
         )
         if fetch_result.returncode != 0:
             stderr = fetch_result.stderr.decode(errors='replace').strip()
@@ -332,7 +334,7 @@ def channel_status(project_dir=None, fetch=True):
 
 def _perform_core_update_locked(stash_local_changes=True):
     project_dir = str(PROJECT_ROOT)
-    from web_interface import update_channel, git_index_lock
+    from web_interface import update_channel
 
     # A git killed mid-command (a timed-out pull, a power cut) leaves
     # .git/index.lock behind, and every later update then fails with
@@ -343,6 +345,14 @@ def _perform_core_update_locked(stash_local_changes=True):
                 'message': 'Another git command is running in the LEDMatrix folder; '
                            'try again in a minute.'}
 
+    # A pull killed mid-rebase leaves .git/rebase-merge behind, and every
+    # later `pull --rebase` then fails. Nothing is running (checked above).
+    git_dir = Path(project_dir) / '.git'
+    if any((git_dir / name).exists() for name in ('rebase-merge', 'rebase-apply')):
+        logger.warning("aborting a rebase left by an interrupted update")
+        git_index_lock.run_git(['git', 'rebase', '--abort'], capture_output=True, text=True,
+                               timeout=60, cwd=project_dir)
+
     # Which code to move to: the newest release (stable) or main (beta).
     # See web_interface/update_channel.py; it never picks an older commit.
     channel, fetch_error = channel_status(project_dir)
@@ -434,25 +444,33 @@ def _perform_core_update_locked(stash_local_changes=True):
     # Stable checks out the release tag, carrying edits across the way the
     # pull's --autostash does.
     channel_note = ''
-    if action == update_channel.ACTION_CHECKOUT_TAG:
+    try:
-        result, autostash_note = update_channel.checkout_release(project_dir, channel.newest_release)
+        if action == update_channel.ACTION_CHECKOUT_TAG:
-        channel_note = f"Now on release {channel.newest_release} (stable channel). {autostash_note}".strip()
+            result, autostash_note = update_channel.checkout_release(project_dir, channel.newest_release)
-    elif action == update_channel.ACTION_SWITCH_TO_BETA:
+            channel_note = f"Now on release {channel.newest_release} (stable channel). {autostash_note}".strip()
-        result, autostash_note = update_channel.checkout_beta_branch(project_dir)
+        elif action == update_channel.ACTION_SWITCH_TO_BETA:
-        if result.returncode == 0:
+            result, autostash_note = update_channel.checkout_beta_branch(project_dir)
-            result = subprocess.run(['git', 'pull', '--rebase', '--autostash'],
+            if result.returncode == 0:
-                                    capture_output=True, text=True, timeout=60, cwd=project_dir)
+                result = git_index_lock.run_git(['git', 'pull', '--rebase', '--autostash'],
-        channel_note = f"Now following {update_channel.BETA_BRANCH} (beta channel). {autostash_note}".strip()
+                                               capture_output=True, text=True,
-    else:
+                                               timeout=git_index_lock.MUTATING_TIMEOUT_SECONDS, cwd=project_dir)
-        result = subprocess.run(
+            channel_note = f"Now following {update_channel.BETA_BRANCH} (beta channel). {autostash_note}".strip()
-            pull_args,
+        else:
-            capture_output=True,
+            result = git_index_lock.run_git(
-            text=True,
+                pull_args,
-            timeout=60,
+                capture_output=True,
-            cwd=project_dir
+                text=True,
-        )
+                timeout=git_index_lock.MUTATING_TIMEOUT_SECONDS,
-        if channel.waiting and channel.newest_release:
+                cwd=project_dir
-            channel_note = channel.message
+            )
+            if channel.waiting and channel.newest_release:
+                channel_note = channel.message
+    except subprocess.TimeoutExpired:
+        # run_git has stopped git and removed the lock it left.
+        logger.warning("git timed out during Update Code")
+        return {'status': 'error', 'restart_required': False, 'dependency_failures': [],
+                'message': 'Update failed: git took too long (a slow network or SD card). '
+                           'The partial update was cleaned up; try again.'}
 
     # Give the branch tracking information so the next pull is a plain
     # `git pull` — otherwise every update repeats the fallback.

web_interface/git_index_lock.py

@@ -18,6 +18,8 @@ the file alone. Standard library only.
 """
 import logging
 import os
+import signal
+import subprocess
 import time
 from pathlib import Path
 
@@ -33,6 +35,15 @@ REMOVED = 'removed'    # it was stale and has been removed
 IN_USE = 'in_use'      # a git process is running here (or the lock is brand new)
 UNKNOWN = 'unknown'    # could not tell whether git is running; left alone
 
+#: How long a timed-out git gets to exit after SIGTERM (which lets it remove
+#: its own lock) before the whole process group is SIGKILLed.
+TERM_GRACE_SECONDS = 5
+
+#: Timeout for commands that rewrite the checkout (pull, checkout, reset,
+#: stash). A Pi on an SD card can take minutes; killing git is what leaves the
+#: lock behind, so this is generous.
+MUTATING_TIMEOUT_SECONDS = 600
+
 _PROC = Path('/proc')
 
 
@@ -137,7 +148,59 @@ def clear_stale_index_lock(project_dir, proc_root=_PROC, now=None):
     return REMOVED
 
 
+def _stop_group(proc):
+    """SIGTERM the process group, then SIGKILL whatever is left of it."""
+    try:
+        os.killpg(proc.pid, signal.SIGTERM)
+    except (ProcessLookupError, PermissionError):
+        pass
+    try:
+        proc.wait(timeout=TERM_GRACE_SECONDS)
+    except subprocess.TimeoutExpired:
+        pass
+    # Also reaps git's children (git-remote-https, rebase helpers) that
+    # outlive a killed parent.
+    try:
+        os.killpg(proc.pid, signal.SIGKILL)
+    except (ProcessLookupError, PermissionError):
+        pass
+
+
+def run_git(cmd, *, cwd=None, timeout=None, capture_output=True, text=True, **kwargs):
+    """``subprocess.run`` for git, without leaving ``.git/index.lock`` behind.
+
+    ``subprocess.run`` SIGKILLs only the direct child when ``timeout`` expires,
+    which skips git's cleanup and orphans its helpers. This starts git in its
+    own session, stops the whole group with SIGTERM before SIGKILL, removes a
+    lock nothing owns any more, then raises ``TimeoutExpired`` as
+    ``subprocess.run`` would.
+    """
+    if capture_output:
+        kwargs['stdout'] = kwargs['stderr'] = subprocess.PIPE
+    with subprocess.Popen(cmd, cwd=cwd, text=text, start_new_session=True, **kwargs) as proc:
+        try:
+            stdout, stderr = proc.communicate(timeout=timeout)
+        except subprocess.TimeoutExpired as exc:
+            _stop_group(proc)
+            exc.stdout, exc.stderr = proc.communicate()
+            if cwd is not None:
+                clear_stale_index_lock(cwd)
+            raise
+        except BaseException:
+            _stop_group(proc)
+            raise
+    return subprocess.CompletedProcess(proc.args, proc.returncode, stdout, stderr)
+
+
+def read_only_env():
+    """Environment for git calls that only read: ``GIT_OPTIONAL_LOCKS=0``
+    stops ``status`` and ``describe --dirty`` refreshing (and so locking) the
+    index, so a poll that times out cannot leave a lock."""
+    return {**os.environ, 'GIT_OPTIONAL_LOCKS': '0'}
+
+
 def manual_fix_hint(project_dir):
     """The sentence shown when a lock blocks the update and was not removed."""
-    return (f"If no other git command is running, remove the stale lock with "
+    lock = Path(project_dir) / '.git' / 'index.lock'
-            f"rm -f {Path(project_dir) / '.git' / 'index.lock'} and try again.")
+    return (f"If it keeps coming back, stop the LEDMatrix services, kill any running git "
+            f"(ps aux | grep git), then remove it with rm -f {lock} and try again.")

web_interface/update_channel.py

@@ -24,6 +24,8 @@ import re
 import subprocess  # nosec B404 - list-form argv only, no shell  # nosemgrep
 from pathlib import Path
 
+from web_interface import git_index_lock
+
 PROJECT_ROOT = Path(__file__).resolve().parent.parent
 CHANNELS = ('stable', 'beta')
 DEFAULT_CHANNEL = 'stable'
@@ -105,7 +107,7 @@ class ChannelStatus(dict):
 
 
 def _git(project_dir, run, *args, timeout=30):
-    run = run or subprocess.run
+    run = run or git_index_lock.run_git
     return run(['git', *args], cwd=str(project_dir), capture_output=True, text=True, timeout=timeout)