Compare commits

..
Author SHA1 Message Date
ChuckBuilds 73fff8d2d5 test(systemd): pin the arena value instead of accepting a range
Review follow-up. The range check accepted 1, 3 and 4, so a change to 4 --
which hands most of the resident saving back -- passed a test whose whole
purpose is to notice that.

Pinned to the value the unit ships, in one named constant. Raising it is still
a legitimate response to a frame-time regression, but it should be a visible
edit here rather than silent drift, and the failure message says so.

Mutation-checked: changing the unit to 4 now fails.
2026-08-20 01:55:14 -04:00
ChuckBuilds 446207ffbc perf(systemd): cap glibc malloc arenas on the display service
Measured on a live rig 2.5 hours after start:

    RSS                          1030 MB
    Private_Dirty                 988 MB
    anonymous mappings > 10 MB       23
    largest        104, 79, 66, 63, 63 MB, on 64 MB-aligned addresses
    threads                           9
    cores                             3   -> glibc ceiling = 8 x 3 = 24 arenas

23 against a ceiling of 24, all 64 MB-aligned: these are glibc's per-thread
malloc arenas, not live objects. The data the process was actually holding
accounts for perhaps 15 MB -- the widest scroll strip observed was 35,746 x 64,
about 7 MB as RGB and the same again for its numpy mirror.

It is bloat rather than a leak: sampled four times over 135 seconds, RSS sat
between 990 and 1030 MB rather than climbing. glibc gives each allocating
thread its own arena, grows them to hold peak demand, and never gives them
back. A process that builds and drops large images across several threads is
exactly the shape that produces this.

The device had 59 MB free at the time, on 1845 MB total.

MALLOC_ARENA_MAX=2 trades a little allocator concurrency for that resident
memory. It is a tuning knob rather than a fix for a defect, so the rationale
and the measurements sit next to it in the unit file, and a test asserts they
stay there -- a bare environment variable invites removal by whoever meets it
next.

Two things this is NOT, both checked rather than assumed:

- Not an OOM problem today. A grep for "oom" in the service journal returned
  24 matches, all of which were the radar logging zoom=9 and zoom=7. The kernel
  OOM killer has not fired: dmesg has zero matches.
- Not currently capped by the unit's MemoryMax=85% either. That directive is in
  this file but absent from the unit actually installed on the rig, which
  reports MemoryMax=infinity, so nothing is enforcing a ceiling there.

The saving is unmeasured on hardware: applying it needs a service restart,
which blanks the panel, so that is the user's call rather than something to do
mid-audit. If p99 frame time regresses -- it sits at 18.4 ms against a 16.7 ms
budget for 60 FPS, so there is not much headroom -- raise the value rather than
remove it.
2026-08-20 00:17:27 -04:00
4 changed files with 108 additions and 179 deletions
+1 -75
View File
@@ -111,12 +111,6 @@ class RestoreOptions:
"""Which sections of a backup should be restored.""" """Which sections of a backup should be restored."""
restore_config: bool = True restore_config: bool = True
#: Whether to take the backup's display.hardware block as well.
#: Off by default: that block describes the panel physically wired to
#: *this* device -- its size, chain length, mapping, multiplexing and
#: refresh cap. A backup carries the panel of the machine it was taken
#: on, and restoring one onto a different rig drives the wrong geometry.
restore_hardware: bool = False
restore_secrets: bool = True restore_secrets: bool = True
restore_wifi: bool = True restore_wifi: bool = True
restore_fonts: bool = True restore_fonts: bool = True
@@ -555,72 +549,6 @@ def _copy_file(src: Path, dst: Path) -> None:
raise raise
_HARDWARE_PATH = ("display", "hardware")
def _restore_config_preserving_hardware(src: Path, dst: Path, keep_hardware: bool) -> None:
"""Copy a backed-up config.json, optionally keeping the local panel block.
display.hardware describes the panel physically attached to this device:
cols, rows, chain_length, hardware_mapping, panel_type, multiplexing and
the refresh-rate cap. None of that travels with a configuration -- it is a
property of the machine. Restoring a backup taken on a 512x64 rig onto a
128x32 one used to overwrite the smaller panel's geometry with the larger
one's, which is not a setting the user can see going wrong; the display
simply stops being right.
Falls back to a plain copy when either file cannot be parsed, so a restore
never fails because of this.
"""
if not keep_hardware:
_copy_file(src, dst)
return
try:
incoming = json.loads(src.read_text(encoding="utf-8"))
local = json.loads(dst.read_text(encoding="utf-8")) if dst.exists() else {}
except (OSError, ValueError) as exc:
logger.warning(
"[Backup] Could not merge local panel config (%s); restoring the "
"backup's config.json as-is", exc)
_copy_file(src, dst)
return
section, key = _HARDWARE_PATH
local_hw = (local.get(section) or {}).get(key)
if not isinstance(local_hw, dict) or not local_hw:
_copy_file(src, dst)
return
if not isinstance(incoming.get(section), dict):
incoming[section] = {}
incoming_hw = incoming[section].get(key)
incoming[section][key] = local_hw
if isinstance(incoming_hw, dict) and incoming_hw != local_hw:
logger.info(
"[Backup] Kept this device's display.hardware; the backup's panel "
"was %sx%s chain %s, this one is %sx%s chain %s",
incoming_hw.get("cols"), incoming_hw.get("rows"),
incoming_hw.get("chain_length"),
local_hw.get("cols"), local_hw.get("rows"),
local_hw.get("chain_length"))
# Write the merged result to a scratch file and hand it to _copy_file
# rather than renaming it into place here. _copy_file preserves the
# destination's mode and owner on purpose -- these config files are
# installed root-owned while the web interface that runs a restore is not
# root -- and a bare write would have replaced a root-owned config.json
# with one owned by the web user at whatever the umask allows.
scratch = dst.with_suffix(dst.suffix + ".merged-tmp")
try:
scratch.write_text(json.dumps(incoming, indent=2) + "\n", encoding="utf-8")
_copy_file(scratch, dst)
finally:
try:
scratch.unlink()
except OSError:
pass
def restore_backup( def restore_backup(
zip_path: Path, zip_path: Path,
project_root: Path, project_root: Path,
@@ -656,9 +584,7 @@ def restore_backup(
# Main config. # Main config.
if options.restore_config and (tmp_dir / _CONFIG_REL).exists(): if options.restore_config and (tmp_dir / _CONFIG_REL).exists():
try: try:
_restore_config_preserving_hardware( _copy_file(tmp_dir / _CONFIG_REL, project_root / _CONFIG_REL)
tmp_dir / _CONFIG_REL, project_root / _CONFIG_REL,
keep_hardware=not options.restore_hardware)
result.restored.append("config") result.restored.append("config")
except OSError as e: except OSError as e:
logger.error("[Backup] Failed to restore config.json: %s", e, exc_info=True) logger.error("[Backup] Failed to restore config.json: %s", e, exc_info=True)
+12
View File
@@ -8,6 +8,18 @@ Type=simple
User=root User=root
WorkingDirectory=__PROJECT_ROOT_DIR__ WorkingDirectory=__PROJECT_ROOT_DIR__
Environment=PYTHONDONTWRITEBYTECODE=1 Environment=PYTHONDONTWRITEBYTECODE=1
# glibc gives each allocating thread its own malloc arena, up to 8 x CPU count,
# and an arena that has grown is never handed back to the OS. This process runs
# 9 threads on a 3-core Pi, so the ceiling is 24 arenas -- and a rig measured at
# 1030 MB resident held 23 large anonymous mappings on 64 MB-aligned addresses,
# 920 MB of them, while the live data it was actually holding (widest scroll
# strip seen: 35,746 x 64) accounts for roughly 15 MB. That gap is arena bloat,
# not leaked objects: RSS was flat across repeated sampling, not climbing.
#
# Capping the arenas trades a little allocator concurrency for a large amount of
# resident memory on a device that has neither to spare. 2 is the usual value;
# raise it if frame times regress.
Environment=MALLOC_ARENA_MAX=2
ExecStart=/usr/bin/python3 __PROJECT_ROOT_DIR__/run.py ExecStart=/usr/bin/python3 __PROJECT_ROOT_DIR__/run.py
# Restart=always, not on-failure: run.py exiting 0 (a clean shutdown path taken # Restart=always, not on-failure: run.py exiting 0 (a clean shutdown path taken
# for a reason that no longer applies, e.g. a config reload) would otherwise leave # for a reason that no longer applies, e.g. a config reload) would otherwise leave
-104
View File
@@ -1,104 +0,0 @@
#!/usr/bin/env python3
"""A restore must not repoint this device at another machine's panel.
display.hardware describes the panel physically wired to this device -- cols,
rows, chain_length, hardware_mapping, panel_type, multiplexing, the refresh
cap. A backup carries the panel of the machine it was taken on. Restoring a
512x64 rig's backup onto a 128x32 one used to overwrite the smaller panel's
geometry with the larger one's, and nothing on screen explains why: the
display just stops being right.
That is not hypothetical. It happened, and the rig it happened to had to be
reflashed.
"""
import json
import sys
from pathlib import Path
sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
from src.backup_manager import _restore_config_preserving_hardware # noqa: E402
BIG = {"display": {"hardware": {"cols": 128, "rows": 64, "chain_length": 4,
"hardware_mapping": "adafruit-hat-pwm"},
"runtime": {"gpio_slowdown": 4}},
"timezone": "America/New_York", "some-plugin": {"enabled": True}}
SMALL = {"display": {"hardware": {"cols": 64, "rows": 32, "chain_length": 2,
"hardware_mapping": "regular"},
"runtime": {"gpio_slowdown": 2}},
"timezone": "UTC"}
def _run(tmp, keep):
src = tmp / "backup_config.json"; src.write_text(json.dumps(BIG))
dst = tmp / "config.json"; dst.write_text(json.dumps(SMALL))
_restore_config_preserving_hardware(src, dst, keep_hardware=keep)
return json.loads(dst.read_text())
def test_local_panel_survives(tmp_path):
out = _run(tmp_path, keep=True)
hw = out["display"]["hardware"]
assert (hw["cols"], hw["rows"], hw["chain_length"]) == (64, 32, 2), (
"the restore repointed this device at the backup's panel")
assert hw["hardware_mapping"] == "regular", "panel wiring came from the backup"
def test_everything_else_is_restored(tmp_path):
out = _run(tmp_path, keep=True)
assert out["timezone"] == "America/New_York", "config was not restored"
assert out["some-plugin"] == {"enabled": True}, "plugin config was not restored"
assert out["display"]["runtime"] == {"gpio_slowdown": 4}, (
"only display.hardware should be held back")
def test_opting_in_takes_the_backups_panel(tmp_path):
out = _run(tmp_path, keep=False)
hw = out["display"]["hardware"]
assert (hw["cols"], hw["rows"], hw["chain_length"]) == (128, 64, 4)
def test_a_device_with_no_local_hardware_takes_the_backups(tmp_path):
src = tmp_path / "b.json"; src.write_text(json.dumps(BIG))
dst = tmp_path / "c.json"; dst.write_text(json.dumps({"timezone": "UTC"}))
_restore_config_preserving_hardware(src, dst, keep_hardware=True)
out = json.loads(dst.read_text())
assert out["display"]["hardware"]["cols"] == 128, (
"nothing local to preserve, so the backup's panel should be used")
def test_unparseable_local_config_still_restores(tmp_path):
src = tmp_path / "b.json"; src.write_text(json.dumps(BIG))
dst = tmp_path / "c.json"; dst.write_text("{ not json")
_restore_config_preserving_hardware(src, dst, keep_hardware=True)
assert json.loads(dst.read_text())["timezone"] == "America/New_York", (
"a restore must never fail because of this merge")
def test_the_destination_mode_is_preserved(tmp_path):
"""config.json is installed with a deliberate mode; a merge must not widen it.
_copy_file preserves the destination's mode and owner on purpose -- these
files are root-owned while the web interface running the restore is not.
Writing the merged result directly would have replaced that with whatever
the umask allowed.
"""
import os
import stat as statmod
src = tmp_path / "b.json"; src.write_text(json.dumps(BIG))
dst = tmp_path / "c.json"; dst.write_text(json.dumps(SMALL))
os.chmod(dst, 0o600)
_restore_config_preserving_hardware(src, dst, keep_hardware=True)
mode = statmod.S_IMODE(os.stat(dst).st_mode)
assert mode == 0o600, f"restore widened config.json from 0600 to {oct(mode)}"
assert json.loads(dst.read_text())["display"]["hardware"]["cols"] == 64
def test_no_scratch_file_is_left_behind(tmp_path):
src = tmp_path / "b.json"; src.write_text(json.dumps(BIG))
dst = tmp_path / "c.json"; dst.write_text(json.dumps(SMALL))
_restore_config_preserving_hardware(src, dst, keep_hardware=True)
leftovers = [p.name for p in tmp_path.iterdir() if "tmp" in p.name]
assert not leftovers, f"scratch files left in place: {leftovers}"
+95
View File
@@ -0,0 +1,95 @@
"""The display unit must cap glibc's malloc arenas.
glibc hands each allocating thread its own malloc arena, up to 8 x CPU count,
and an arena that has grown is never returned to the OS. This process runs
threads for the render loop, the update workers and the background fetchers, so
on a 3-core Pi the ceiling is 24 arenas.
Measured on a live rig, 2.5 hours in:
RSS 1030 MB
Private_Dirty 988 MB
anonymous mappings > 10 MB 23 (ceiling is 8 x 3 = 24)
largest few 104, 79, 66, 63, 63 MB, on 64 MB-aligned addresses
against live data that accounts for perhaps 15 MB -- the widest scroll strip
observed was 35,746 x 64, about 7 MB as RGB and the same again for its numpy
mirror. Repeated sampling showed RSS flat between 990 and 1030 MB rather than
climbing, so this is arena bloat rather than a leak: memory Python has freed
but glibc is holding per-arena.
The device had 59 MB free at the time.
Capping the arena count trades a little allocator concurrency for that resident
memory. The render loop is latency-sensitive, so if p99 frame time regresses the
right response is to raise this rather than remove it.
"""
import re
from pathlib import Path
import pytest
UNIT = (Path(__file__).resolve().parent.parent / "systemd" / "ledmatrix.service")
#: The value the unit is expected to carry. 2 is the usual choice for a
#: threaded Python process; 1-4 all keep some of the saving, but only one of
#: them is what this project ships.
EXPECTED_ARENA_MAX = 2
def _environment(unit_text):
return dict(
line.split("=", 2)[1:3] if line.count("=") >= 2 else (line.split("=", 1)[1], "")
for line in unit_text.splitlines()
if line.startswith("Environment=")
)
def test_the_unit_exists():
assert UNIT.is_file(), f"{UNIT} is missing"
def test_malloc_arena_max_is_capped():
env = _environment(UNIT.read_text(encoding="utf-8"))
assert "MALLOC_ARENA_MAX" in env, (
"the display unit does not cap glibc arenas; on a 3-core Pi the default "
"ceiling is 24 and a measured rig held 23 of them, 920 MB"
)
value = int(env["MALLOC_ARENA_MAX"])
# Pinned, not a range. A range let a change to 4 -- which hands most of the
# saving back -- pass unnoticed, which was the point of the finding that
# prompted this. Raising it is a legitimate response to a frame-time
# regression, but it should be a visible edit here rather than a silent
# drift, so the number lives in one place and changing it shows up in
# review.
assert value == EXPECTED_ARENA_MAX, (
f"MALLOC_ARENA_MAX={value}, expected {EXPECTED_ARENA_MAX}. If this was "
"raised deliberately because frame times regressed, update "
"EXPECTED_ARENA_MAX here and say so in the commit."
)
def test_the_reason_is_recorded_next_to_it():
"""A bare tuning knob invites removal by whoever meets it next."""
text = UNIT.read_text(encoding="utf-8")
index = text.index("Environment=MALLOC_ARENA_MAX")
preamble = text[:index].splitlines()[-12:]
comment = "\n".join(line for line in preamble if line.startswith("#"))
assert "arena" in comment.lower(), "no explanation precedes the setting"
assert re.search(r"\d", comment), (
"the explanation cites no measurement, so a reader cannot tell whether "
"it still applies to their hardware"
)
@pytest.mark.parametrize("unit", ["ledmatrix.service"])
def test_the_unit_still_parses_as_ini(unit):
"""systemd will refuse a malformed unit, and the panel stays dark."""
import configparser
path = UNIT.parent / unit
parser = configparser.ConfigParser(strict=False)
# systemd allows repeated keys; ConfigParser needs them merged, not rejected.
parser.read_string(path.read_text(encoding="utf-8"))
assert parser.has_section("Service")
assert parser.has_option("Service", "ExecStart")