feat: add pipeline health checks module and CLI runner

- Create health/ package with HealthStatus, HealthCheck, HealthReport
- Add check_motion_freshness, check_embedding_coverage, check_llm_coverage
- Add scripts/health_check.py CLI with text/JSON output and exit codes
- Add comprehensive tests for core, checks, and CLI

P4-005: Pipeline health checks
This commit is contained in:
2026-05-01 00:02:45 +02:00
parent 04cc62ea06
commit e352d7c7bc
6 changed files with 495 additions and 0 deletions
+69
View File
@@ -0,0 +1,69 @@
import sys
from unittest.mock import MagicMock, patch
import pytest
from scripts.health_check import main
class TestHealthCheckCLI:
@patch("scripts.health_check.duckdb.connect")
@patch("scripts.health_check.config")
def test_all_ok_exits_0(self, mock_config, mock_connect):
mock_config.DATABASE_PATH = "/fake/db"
mock_conn = MagicMock()
mock_conn.execute.return_value.fetchone.side_effect = [
[150], # motion count
[100], # total motions
[100], # embeddings count
[100], # total motions
[0], # missing explanations
]
mock_connect.return_value = mock_conn
with patch.object(sys, "argv", ["health_check"]):
exit_code = main()
assert exit_code == 0
@patch("scripts.health_check.duckdb.connect")
@patch("scripts.health_check.config")
def test_critical_exits_2(self, mock_config, mock_connect):
mock_config.DATABASE_PATH = "/fake/db"
mock_conn = MagicMock()
mock_conn.execute.return_value.fetchone.side_effect = [
[0], # no recent motions
[100], # total motions
[100], # embeddings count
[100], # total motions
[0], # missing explanations
]
mock_connect.return_value = mock_conn
with patch.object(sys, "argv", ["health_check"]):
exit_code = main()
assert exit_code == 2
@patch("scripts.health_check.duckdb.connect")
@patch("scripts.health_check.config")
def test_json_format(self, mock_config, mock_connect, capsys):
mock_config.DATABASE_PATH = "/fake/db"
mock_conn = MagicMock()
mock_conn.execute.return_value.fetchone.side_effect = [
[150], [100], [100], [100], [0]
]
mock_connect.return_value = mock_conn
with patch.object(sys, "argv", ["health_check", "--format", "json"]):
main()
captured = capsys.readouterr()
assert '"status": "ok"' in captured.out
assert '"exit_code": 0' in captured.out
@patch("scripts.health_check.duckdb.connect")
def test_db_connect_failure_exits_2(self, mock_connect):
mock_connect.side_effect = Exception("cannot open")
with patch.object(sys, "argv", ["health_check"]):
exit_code = main()
assert exit_code == 2
+92
View File
@@ -0,0 +1,92 @@
from datetime import datetime, timedelta
from unittest.mock import MagicMock
import pytest
from health import HealthStatus
from health.checks import check_embedding_coverage, check_llm_coverage, check_motion_freshness
class TestCheckMotionFreshness:
def test_recent_motions_ok(self):
conn = MagicMock()
conn.execute.return_value.fetchone.return_value = [150]
result = check_motion_freshness(conn, max_age_days=7, min_motions=100)
assert result.status == HealthStatus.OK
assert result.details["count"] == 150
def test_no_motions_critical(self):
conn = MagicMock()
conn.execute.return_value.fetchone.return_value = [0]
result = check_motion_freshness(conn, max_age_days=7, min_motions=100)
assert result.status == HealthStatus.CRITICAL
def test_low_count_warning(self):
conn = MagicMock()
conn.execute.return_value.fetchone.return_value = [50]
result = check_motion_freshness(conn, max_age_days=7, min_motions=100)
assert result.status == HealthStatus.WARNING
def test_query_error_critical(self):
conn = MagicMock()
conn.execute.side_effect = Exception("db down")
result = check_motion_freshness(conn)
assert result.status == HealthStatus.CRITICAL
assert "db down" in result.message
class TestCheckEmbeddingCoverage:
def test_full_coverage_ok(self):
conn = MagicMock()
conn.execute.return_value.fetchone.side_effect = [[100], [100]]
result = check_embedding_coverage(conn, min_coverage=0.95)
assert result.status == HealthStatus.OK
assert result.details["coverage"] == 1.0
def test_low_coverage_warning(self):
conn = MagicMock()
conn.execute.return_value.fetchone.side_effect = [[100], [80]]
result = check_embedding_coverage(conn, min_coverage=0.95)
assert result.status == HealthStatus.WARNING
assert result.details["coverage"] == 0.8
def test_empty_db_critical(self):
conn = MagicMock()
conn.execute.return_value.fetchone.side_effect = [[0], [0]]
result = check_embedding_coverage(conn)
assert result.status == HealthStatus.CRITICAL
def test_query_error_critical(self):
conn = MagicMock()
conn.execute.side_effect = Exception("db down")
result = check_embedding_coverage(conn)
assert result.status == HealthStatus.CRITICAL
class TestCheckLLMCoverage:
def test_full_coverage_ok(self):
conn = MagicMock()
conn.execute.return_value.fetchone.side_effect = [[100], [0]]
result = check_llm_coverage(conn)
assert result.status == HealthStatus.OK
assert result.details["missing_ratio"] == 0.0
def test_some_missing_warning(self):
conn = MagicMock()
conn.execute.return_value.fetchone.side_effect = [[100], [10]]
result = check_llm_coverage(conn)
assert result.status == HealthStatus.WARNING
assert result.details["missing_ratio"] == 0.1
def test_too_many_missing_critical(self):
conn = MagicMock()
conn.execute.return_value.fetchone.side_effect = [[100], [20]]
result = check_llm_coverage(conn, max_missing_ratio=0.15)
assert result.status == HealthStatus.CRITICAL
assert result.details["missing_ratio"] == 0.2
def test_query_error_critical(self):
conn = MagicMock()
conn.execute.side_effect = Exception("db down")
result = check_llm_coverage(conn)
assert result.status == HealthStatus.CRITICAL
+54
View File
@@ -0,0 +1,54 @@
import pytest
from health import HealthCheck, HealthReport, HealthStatus, run_checks
class TestHealthStatus:
def test_ok_less_than_warning(self):
assert HealthStatus.OK.value == "ok"
def test_warning_less_than_critical(self):
assert HealthStatus.WARNING.value == "warning"
class TestHealthReport:
def test_all_ok_returns_ok(self):
checks = [
HealthCheck("a", HealthStatus.OK, "fine", {}),
HealthCheck("b", HealthStatus.OK, "fine", {}),
]
report = run_checks(checks)
assert report.status == HealthStatus.OK
assert report.exit_code == 0
def test_one_warning_returns_warning(self):
checks = [
HealthCheck("a", HealthStatus.OK, "fine", {}),
HealthCheck("b", HealthStatus.WARNING, "hmm", {}),
]
report = run_checks(checks)
assert report.status == HealthStatus.WARNING
assert report.exit_code == 1
def test_one_critical_returns_critical(self):
checks = [
HealthCheck("a", HealthStatus.OK, "fine", {}),
HealthCheck("b", HealthStatus.CRITICAL, "bad", {}),
]
report = run_checks(checks)
assert report.status == HealthStatus.CRITICAL
assert report.exit_code == 2
def test_critical_trumps_warning(self):
checks = [
HealthCheck("a", HealthStatus.WARNING, "hmm", {}),
HealthCheck("b", HealthStatus.CRITICAL, "bad", {}),
]
report = run_checks(checks)
assert report.status == HealthStatus.CRITICAL
assert report.exit_code == 2
def test_empty_checks_returns_ok(self):
report = run_checks([])
assert report.status == HealthStatus.OK
assert report.exit_code == 0