personal memory agent
Add session readiness pre-flight to observer startup
Observer was crash-looping after reboot before desktop session was ready
(no PulseAudio, no DBus, no display). Now:
- check_session_ready() verifies DISPLAY, DBUS, and pactl before setup
- Exit code 75 (EX_TEMPFAIL) signals "not ready" to supervisor
- Supervisor uses 15s retry delay for tempfail (vs 0/1/5s backoff),
skips noisy error-level logging and desktop notifications
- Audio detection retries 3x with 5s delay for device initialization
- Hardened detect.py: wrapped sc.default_speaker() and
sc.all_microphones() in try/except
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
+92
-16
+14
-6
observe/detect.py
+14
-6
observe/detect.py
···
1
1
# SPDX-License-Identifier: AGPL-3.0-only
2
2
# Copyright (c) 2026 sol pbc
3
3
4
+
import logging
4
5
import threading
5
6
6
7
import numpy as np
7
8
import soundcard as sc
9
+
10
+
logger = logging.getLogger(__name__)
8
11
9
12
10
13
def input_detect(duration=0.4, sample_rate=44100):
11
14
t = np.linspace(0, duration, int(sample_rate * duration), endpoint=False)
12
15
tone = 0.5 * np.sin(2 * np.pi * 18000 * t) # ultrasonic
13
16
14
-
devices = sc.all_microphones(include_loopback=True)
17
+
try:
18
+
devices = sc.all_microphones(include_loopback=True)
19
+
except Exception:
20
+
logger.warning("Failed to enumerate audio devices")
21
+
return None, None
15
22
if not devices:
16
-
print("No matching devices found:")
17
-
for mic in devices:
18
-
print(mic)
23
+
logger.warning("No audio devices found")
19
24
return None, None
20
25
21
26
results = {}
···
33
38
34
39
def play_tone():
35
40
barrier.wait()
36
-
sp = sc.default_speaker()
37
-
sp.play(tone, samplerate=sample_rate)
41
+
try:
42
+
sp = sc.default_speaker()
43
+
sp.play(tone, samplerate=sample_rate)
44
+
except Exception:
45
+
logger.warning("No default speaker available for tone detection")
38
46
39
47
threads = []
40
48
for mic in devices:
+57
-2
observe/linux/observer.py
+57
-2
observe/linux/observer.py
···
21
21
import logging
22
22
import os
23
23
import platform
24
+
import shutil
24
25
import signal
25
26
import socket
27
+
import subprocess
26
28
import sys
27
29
import time
28
30
from pathlib import Path
···
59
61
TMUX_ACTIVITY_THRESHOLD = 5 # seconds - fixed window for activity detection
60
62
61
63
64
+
# Exit codes
65
+
EXIT_TEMPFAIL = 75 # EX_TEMPFAIL: session not ready, retry later
66
+
62
67
# Capture modes
63
68
MODE_IDLE = "idle"
64
69
MODE_SCREENCAST = "screencast"
65
70
MODE_TMUX = "tmux"
71
+
72
+
# Audio detection retry
73
+
DETECT_RETRIES = 3
74
+
DETECT_RETRY_DELAY = 5 # seconds
66
75
67
76
68
77
class Observer:
···
140
149
141
150
async def setup(self):
142
151
"""Initialize audio devices and DBus connection."""
143
-
# Detect and start audio recorder
144
-
if not self.audio_recorder.detect():
152
+
# Detect audio devices with retry (devices may still be initializing)
153
+
detected = False
154
+
for attempt in range(DETECT_RETRIES):
155
+
if self.audio_recorder.detect():
156
+
detected = True
157
+
break
158
+
if attempt < DETECT_RETRIES - 1:
159
+
logger.info(
160
+
"Audio detection attempt %d/%d failed, retrying in %ds",
161
+
attempt + 1,
162
+
DETECT_RETRIES,
163
+
DETECT_RETRY_DELAY,
164
+
)
165
+
await asyncio.sleep(DETECT_RETRY_DELAY)
166
+
if not detected:
145
167
logger.error("Failed to detect audio devices")
146
168
return False
147
169
···
807
829
logger.info("Callosum connection stopped")
808
830
809
831
832
+
def check_session_ready() -> str | None:
833
+
"""Check if the desktop session is ready for observation.
834
+
835
+
Returns None if ready, or a description of what's missing.
836
+
"""
837
+
# Display server
838
+
if not os.environ.get("DISPLAY") and not os.environ.get("WAYLAND_DISPLAY"):
839
+
return "no display server (DISPLAY/WAYLAND_DISPLAY not set)"
840
+
841
+
# DBus session bus
842
+
if not os.environ.get("DBUS_SESSION_BUS_ADDRESS"):
843
+
return "no DBus session bus (DBUS_SESSION_BUS_ADDRESS not set)"
844
+
845
+
# PulseAudio / PipeWire audio
846
+
pactl = shutil.which("pactl")
847
+
if pactl:
848
+
try:
849
+
subprocess.run(
850
+
[pactl, "info"],
851
+
capture_output=True,
852
+
timeout=5,
853
+
).check_returncode()
854
+
except (subprocess.CalledProcessError, subprocess.TimeoutExpired):
855
+
return "audio server not responding (pactl info failed)"
856
+
return None
857
+
858
+
810
859
async def async_main(args):
811
860
"""Async entry point."""
861
+
# Pre-flight: check session prerequisites before attempting setup
862
+
not_ready = check_session_ready()
863
+
if not_ready:
864
+
logger.warning("Session not ready: %s", not_ready)
865
+
return EXIT_TEMPFAIL
866
+
812
867
observer = Observer(
813
868
interval=args.interval,
814
869
)
+21
-8
think/supervisor.py
+21
-8
think/supervisor.py
···
36
36
DEFAULT_THRESHOLD = 60
37
37
CHECK_INTERVAL = 30
38
38
MAX_UPDATED_CATCHUP = 4
39
+
EXIT_TEMPFAIL = 75 # EX_TEMPFAIL: service prerequisites not ready
40
+
TEMPFAIL_DELAY = 15 # seconds to wait before retrying a tempfail exit
39
41
40
42
# Global shutdown flag
41
43
shutdown_requested = False
···
933
935
return
934
936
935
937
exited_names = [managed.name for managed in exited]
936
-
msg = f"Runner process exited: {', '.join(sorted(exited_names))}"
937
-
logging.error(msg)
938
938
exit_key = ("runner_exit", tuple(sorted(exited_names)))
939
939
940
-
await alert_mgr.alert_if_ready(exit_key, msg)
940
+
# Check if all exits are tempfail (session not ready)
941
+
all_tempfail = all(m.process.returncode == EXIT_TEMPFAIL for m in exited)
942
+
943
+
if all_tempfail:
944
+
logging.info("Runner waiting for session: %s", ", ".join(sorted(exited_names)))
945
+
else:
946
+
msg = f"Runner process exited: {', '.join(sorted(exited_names))}"
947
+
logging.error(msg)
948
+
await alert_mgr.alert_if_ready(exit_key, msg)
941
949
942
950
for managed in exited:
943
951
# Clear any pending restart request for this service
944
952
_restart_requests.pop(managed.name, None)
945
953
946
954
returncode = managed.process.returncode
955
+
is_tempfail = returncode == EXIT_TEMPFAIL
947
956
logging.info("%s exited with code %s", managed.name, returncode)
948
957
949
958
# Emit stopped event
···
967
976
968
977
# Handle restart if needed
969
978
if managed.restart and not shutdown_requested:
970
-
policy = _get_restart_policy(managed.name)
971
-
uptime = time.time() - policy.last_start if policy.last_start else 0
972
-
if uptime >= 60:
973
-
policy.reset_attempts()
974
-
delay = policy.next_delay()
979
+
# Tempfail: use fixed longer delay, don't burn through backoff
980
+
if is_tempfail:
981
+
delay = TEMPFAIL_DELAY
982
+
else:
983
+
policy = _get_restart_policy(managed.name)
984
+
uptime = time.time() - policy.last_start if policy.last_start else 0
985
+
if uptime >= 60:
986
+
policy.reset_attempts()
987
+
delay = policy.next_delay()
975
988
if delay:
976
989
logging.info("Waiting %ss before restarting %s", delay, managed.name)
977
990
for _ in range(delay):