personal memory agent
Merge branch 'hopper-dvw2lfip-sol-health-command'
+274
-9
+1
-1
muse/help.md
+1
-1
muse/help.md
···
27
27
- `sol supervisor` - Run supervisor services.
28
28
- `sol detect-created` - Detect newly created content artifacts.
29
29
- `sol top` - Show runtime/service activity status.
30
-
- `sol logs` - View service health logs.
30
+
- `sol health` - Show service health status. Use `sol health logs` to view service logs.
31
31
- `sol callosum` - Interact with Callosum message bus tooling.
32
32
- `sol streams` - Manage or inspect stream-related state.
33
33
- `sol journal-stats` - Show journal statistics.
+2
-2
sol.py
+2
-2
sol.py
···
46
46
"schedule": "think.scheduler",
47
47
"detect-created": "think.detect_created",
48
48
"top": "think.top",
49
-
"logs": "think.logs_cli",
49
+
"health": "think.health_cli",
50
50
"callosum": "think.callosum",
51
51
"notify": "think.notify_cli",
52
52
"streams": "think.streams",
···
99
99
"supervisor",
100
100
"schedule",
101
101
"top",
102
-
"logs",
102
+
"health",
103
103
"callosum",
104
104
"notify",
105
105
],
+112
tests/test_health_cli.py
+112
tests/test_health_cli.py
···
1
+
# SPDX-License-Identifier: AGPL-3.0-only
2
+
# Copyright (c) 2026 sol pbc
3
+
4
+
from __future__ import annotations
5
+
6
+
import sys
7
+
from unittest.mock import patch
8
+
9
+
import pytest
10
+
11
+
from think.health_cli import health_check, main, print_status
12
+
13
+
14
+
def test_health_check_no_socket(tmp_path, monkeypatch, capsys):
15
+
monkeypatch.setenv("JOURNAL_PATH", str(tmp_path))
16
+
17
+
result = health_check()
18
+
19
+
captured = capsys.readouterr()
20
+
assert result == 1
21
+
assert "callosum socket not found" in captured.err
22
+
23
+
24
+
def test_health_check_prints_status(capsys):
25
+
status = {
26
+
"services": [
27
+
{"name": "supervisor", "pid": 1001, "uptime_seconds": 65},
28
+
{"name": "observer", "pid": 2002, "uptime_seconds": 5},
29
+
],
30
+
"crashed": [{"name": "sync", "restart_attempts": 2}],
31
+
"tasks": [{"name": "dream", "duration_seconds": 12}],
32
+
"queues": {"indexer": 3, "planner": 0},
33
+
"stale_heartbeats": [],
34
+
}
35
+
36
+
print_status(status)
37
+
38
+
output = capsys.readouterr().out
39
+
assert "Services:" in output
40
+
assert "supervisor" in output
41
+
assert "pid 1001" in output
42
+
assert "observer" in output
43
+
assert "Crashed:" in output
44
+
assert "sync" in output
45
+
assert "Tasks:" in output
46
+
assert "dream" in output
47
+
assert "queued indexer" in output
48
+
assert "Heartbeat: ok" in output
49
+
50
+
51
+
def test_health_check_timeout(tmp_path, monkeypatch, capsys):
52
+
monkeypatch.setenv("JOURNAL_PATH", str(tmp_path))
53
+
sock = tmp_path / "health" / "callosum.sock"
54
+
sock.parent.mkdir(parents=True)
55
+
sock.touch()
56
+
monkeypatch.setattr("think.health_cli.STATUS_TIMEOUT", 0.1)
57
+
58
+
with patch("think.health_cli.CallosumConnection") as mock_conn_cls:
59
+
mock_conn = mock_conn_cls.return_value
60
+
mock_conn.start.return_value = None
61
+
mock_conn.stop.return_value = None
62
+
63
+
result = health_check()
64
+
65
+
captured = capsys.readouterr()
66
+
assert result == 1
67
+
assert "Timed out waiting for supervisor status" in captured.err
68
+
mock_conn.stop.assert_called_once()
69
+
70
+
71
+
def test_health_check_receives_status(tmp_path, monkeypatch, capsys):
72
+
monkeypatch.setenv("JOURNAL_PATH", str(tmp_path))
73
+
sock = tmp_path / "health" / "callosum.sock"
74
+
sock.parent.mkdir(parents=True)
75
+
sock.touch()
76
+
77
+
with patch("think.health_cli.CallosumConnection") as mock_conn_cls:
78
+
mock_conn = mock_conn_cls.return_value
79
+
80
+
def _start(*, callback):
81
+
callback(
82
+
{
83
+
"tract": "supervisor",
84
+
"event": "status",
85
+
"services": [
86
+
{"name": "supervisor", "pid": 111, "uptime_seconds": 120}
87
+
],
88
+
"tasks": [],
89
+
"queues": {},
90
+
"stale_heartbeats": [],
91
+
}
92
+
)
93
+
94
+
mock_conn.start.side_effect = _start
95
+
mock_conn.stop.return_value = None
96
+
97
+
result = health_check()
98
+
99
+
captured = capsys.readouterr()
100
+
assert result == 0
101
+
assert "Services:" in captured.out
102
+
assert "supervisor" in captured.out
103
+
104
+
105
+
def test_main_routes_to_logs(monkeypatch):
106
+
monkeypatch.setattr(sys, "argv", ["sol health", "logs", "--help"])
107
+
108
+
with patch("think.logs_cli.main", side_effect=SystemExit(0)) as mock_logs_main:
109
+
with pytest.raises(SystemExit):
110
+
main()
111
+
112
+
mock_logs_main.assert_called_once()
+153
think/health_cli.py
+153
think/health_cli.py
···
1
+
# SPDX-License-Identifier: AGPL-3.0-only
2
+
# Copyright (c) 2026 sol pbc
3
+
4
+
"""CLI for service health status and logs.
5
+
6
+
Usage:
7
+
sol health Show current service health status
8
+
sol health logs View service health logs
9
+
"""
10
+
11
+
from __future__ import annotations
12
+
13
+
import argparse
14
+
import sys
15
+
import threading
16
+
from datetime import timedelta
17
+
from pathlib import Path
18
+
from typing import Any
19
+
20
+
from think.callosum import CallosumConnection
21
+
from think.utils import get_journal, setup_cli
22
+
23
+
STATUS_TIMEOUT = 10
24
+
25
+
26
+
def format_uptime(seconds: int) -> str:
27
+
"""Format uptime in human-readable format."""
28
+
if seconds < 60:
29
+
return f"{seconds}s"
30
+
31
+
delta = timedelta(seconds=seconds)
32
+
parts = []
33
+
if delta.days:
34
+
parts.append(f"{delta.days}d")
35
+
36
+
hours = delta.seconds // 3600
37
+
if hours:
38
+
parts.append(f"{hours}h")
39
+
40
+
mins = (delta.seconds % 3600) // 60
41
+
if mins:
42
+
parts.append(f"{mins}m")
43
+
44
+
return " ".join(parts)
45
+
46
+
47
+
def print_status(status: dict[str, Any]) -> None:
48
+
"""Print supervisor status in a human-readable format."""
49
+
print("Services:")
50
+
for service in status.get("services", []):
51
+
name = service.get("name", "?")
52
+
pid = service.get("pid", "?")
53
+
uptime_seconds = int(service.get("uptime_seconds", 0) or 0)
54
+
print(f" {name:16} pid {pid} uptime {format_uptime(uptime_seconds)}")
55
+
56
+
crashed = status.get("crashed") or []
57
+
if crashed:
58
+
print()
59
+
print("Crashed:")
60
+
for service in crashed:
61
+
name = service.get("name", "?")
62
+
attempts = service.get("restart_attempts", 0)
63
+
print(f" {name:16} {attempts} restart attempts")
64
+
65
+
print()
66
+
tasks = status.get("tasks") or []
67
+
queues = status.get("queues") or {}
68
+
non_zero_queues = [(name, count) for name, count in sorted(queues.items()) if count]
69
+
70
+
if tasks:
71
+
print("Tasks:")
72
+
for task in tasks:
73
+
name = task.get("name", "?")
74
+
duration = task.get("duration_seconds", 0)
75
+
print(f" {name:16} {duration}s")
76
+
for name, count in non_zero_queues:
77
+
print(f" queued {name:9} {count}")
78
+
elif non_zero_queues:
79
+
print("Tasks:")
80
+
for name, count in non_zero_queues:
81
+
print(f" queued {name:9} {count}")
82
+
else:
83
+
print("Tasks: none")
84
+
85
+
stale = status.get("stale_heartbeats") or []
86
+
if stale:
87
+
print()
88
+
print(f"Heartbeat: STALE ({', '.join(stale)})")
89
+
else:
90
+
print("Heartbeat: ok")
91
+
92
+
93
+
def health_check() -> int:
94
+
"""Request and print one-shot supervisor status."""
95
+
sock_path = Path(get_journal()) / "health" / "callosum.sock"
96
+
if not sock_path.exists():
97
+
print(
98
+
f"Cannot connect: callosum socket not found at {sock_path}",
99
+
file=sys.stderr,
100
+
)
101
+
return 1
102
+
103
+
status_event = threading.Event()
104
+
status_holder: dict[str, dict[str, Any]] = {}
105
+
106
+
def callback(msg: dict[str, Any]) -> None:
107
+
if msg.get("tract") == "supervisor" and msg.get("event") == "status":
108
+
status_holder["data"] = msg
109
+
status_event.set()
110
+
111
+
conn = CallosumConnection(socket_path=sock_path)
112
+
conn.start(callback=callback)
113
+
try:
114
+
got_status = status_event.wait(timeout=STATUS_TIMEOUT)
115
+
finally:
116
+
conn.stop()
117
+
118
+
if not got_status:
119
+
print(
120
+
f"Timed out waiting for supervisor status ({STATUS_TIMEOUT:g}s)",
121
+
file=sys.stderr,
122
+
)
123
+
return 1
124
+
125
+
print_status(status_holder["data"])
126
+
return 0
127
+
128
+
129
+
def main() -> None:
130
+
"""Entry point for ``sol health``."""
131
+
args = sys.argv[1:]
132
+
if args and args[0] == "logs":
133
+
sys.argv = ["sol health logs"] + args[1:]
134
+
from think.logs_cli import main as logs_main
135
+
136
+
logs_main()
137
+
return
138
+
139
+
parser = argparse.ArgumentParser(
140
+
prog="sol health",
141
+
description=(
142
+
"Show service health status.\n\n"
143
+
"Subcommands:\n"
144
+
" logs View service health logs (sol health logs -h for details)"
145
+
),
146
+
formatter_class=argparse.RawDescriptionHelpFormatter,
147
+
)
148
+
setup_cli(parser)
149
+
sys.exit(health_check())
150
+
151
+
152
+
if __name__ == "__main__":
153
+
main()
+6
-6
think/logs_cli.py
+6
-6
think/logs_cli.py
···
4
4
"""CLI for viewing service health logs.
5
5
6
6
Usage:
7
-
sol logs Show last 5 lines from each service
8
-
sol logs -c 20 Show last 20 lines from each service
9
-
sol logs -f Follow all logs for new output
10
-
sol logs --since 30m Lines from last 30 minutes
11
-
sol logs --service observer Only show observer logs
12
-
sol logs --grep "error" Lines matching regex "error"
7
+
sol health logs Show last 5 lines from each service
8
+
sol health logs -c 20 Show last 20 lines from each service
9
+
sol health logs -f Follow all logs for new output
10
+
sol health logs --since 30m Lines from last 30 minutes
11
+
sol health logs --service observer Only show observer logs
12
+
sol health logs --grep "error" Lines matching regex "error"
13
13
"""
14
14
15
15
from __future__ import annotations