bstein/titan-iac
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

monitoring: switch power telemetry to ananke metrics

Browse Source
This commit is contained in:
Brad Stein 2026-04-08 23:33:17 -03:00
parent cf6252c55a
commit e0b124ca4e
8 changed files with 274 additions and 230 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

240
scripts/dashboards_render_atlas.py
View File

@ -442,70 +442,83 @@ PLATFORM_TEST_ACTIVITY_30D = (
'or label_replace(sum by (status) (increase(metis_builds_total[30d])), "source", "metis-build", "__name__", ".*") ' 'or label_replace(sum by (status) (increase(metis_builds_total[30d])), "source", "metis-build", "__name__", ".*") '
'or label_replace(sum by (status) (increase(metis_flashes_total[30d])), "source", "metis-flash", "__name__", ".*")' 'or label_replace(sum by (status) (increase(metis_flashes_total[30d])), "source", "metis-flash", "__name__", ".*")'
) )
HECATE_SELECTOR = 'job="hecate-power"' PLATFORM_TEST_SUCCESS_RATE_BY_SUITE_SERIES = (
HECATE_UPS_DB_NAME = "Pyrphoros" 'label_replace(100 * (sum(increase(ariadne_task_runs_total{status="ok"}[$__interval])) or on() vector(0)) '
HECATE_UPS_DB_NODE = "titan-db" '/ clamp_min((sum(increase(ariadne_task_runs_total[$__interval])) or on() vector(0)), 1), '
HECATE_UPS_TETHYS_NAME = "Statera" '"suite", "ariadne", "__name__", ".*") '
HECATE_UPS_TETHYS_NODE = "titan-24" 'or label_replace(100 * (sum(increase(metis_builds_total{status="ok"}[$__interval])) or on() vector(0)) '
HECATE_UPS_ON_BATTERY = f"sum(hecate_ups_on_battery{{{HECATE_SELECTOR}}}) or on() vector(0)" '/ clamp_min((sum(increase(metis_builds_total[$__interval])) or on() vector(0)), 1), '
HECATE_UPS_LOW_BATTERY = f"sum(hecate_ups_low_battery{{{HECATE_SELECTOR}}}) or on() vector(0)" '"suite", "metis-build", "__name__", ".*") '
HECATE_UPS_RUNTIME_MIN = f"min(hecate_ups_runtime_seconds{{{HECATE_SELECTOR}}}) or on() vector(0)" 'or label_replace(100 * (sum(increase(metis_flashes_total{status="ok"}[$__interval])) or on() vector(0)) '
HECATE_UPS_RUNTIME_HEADROOM_PERCENT = ( '/ clamp_min((sum(increase(metis_flashes_total[$__interval])) or on() vector(0)), 1), '
f"100 * min(hecate_ups_runtime_seconds{{{HECATE_SELECTOR}}}) / " '"suite", "metis-flash", "__name__", ".*")'
f"clamp_min(max(hecate_ups_threshold_seconds{{{HECATE_SELECTOR}}}), 1)"
) )
HECATE_UPS_TRIGGER_COUNT_1D = f"increase(hecate_shutdown_triggers_total{{{HECATE_SELECTOR}}}[1d]) or on() vector(0)" ANANKE_SELECTOR = 'job="ananke-power"'
HECATE_UPS_RUNTIME_DB = ( ANANKE_UPS_DB_NAME = "Pyrphoros"
f'max(hecate_ups_runtime_seconds{{{HECATE_SELECTOR},instance="titan-db"}}) or on() vector(0)' ANANKE_UPS_DB_NODE = "titan-db"
ANANKE_UPS_TETHYS_NAME = "Statera"
ANANKE_UPS_TETHYS_NODE = "titan-24"
ANANKE_UPS_DB_SELECTOR = f'{ANANKE_SELECTOR},source="{ANANKE_UPS_DB_NAME}"'
ANANKE_UPS_TETHYS_SELECTOR = f'{ANANKE_SELECTOR},source="{ANANKE_UPS_TETHYS_NAME}"'
ANANKE_UPS_ON_BATTERY = f"sum(ananke_ups_on_battery{{{ANANKE_SELECTOR}}}) or on() vector(0)"
ANANKE_UPS_LOW_BATTERY = f"sum(ananke_ups_low_battery{{{ANANKE_SELECTOR}}}) or on() vector(0)"
ANANKE_UPS_RUNTIME_MIN = f"min(ananke_ups_runtime_seconds{{{ANANKE_SELECTOR}}}) or on() vector(0)"
ANANKE_UPS_RUNTIME_HEADROOM_PERCENT = (
f"100 * min(ananke_ups_runtime_seconds{{{ANANKE_SELECTOR}}}) / "
f"clamp_min(max(ananke_ups_threshold_seconds{{{ANANKE_SELECTOR}}}), 1)"
) )
HECATE_UPS_RUNTIME_TETHYS = ( ANANKE_UPS_TRIGGER_COUNT_1D = f"increase(ananke_shutdown_triggers_total{{{ANANKE_SELECTOR}}}[1d]) or on() vector(0)"
f'max(hecate_ups_runtime_seconds{{{HECATE_SELECTOR},instance="titan-24"}}) or on() vector(0)' ANANKE_UPS_RUNTIME_DB = (
f'max(ananke_ups_runtime_seconds{{{ANANKE_UPS_DB_SELECTOR}}}) or on() vector(0)'
) )
HECATE_UPS_ON_BATTERY_DB = ( ANANKE_UPS_RUNTIME_TETHYS = (
f'max(hecate_ups_on_battery{{{HECATE_SELECTOR},instance="titan-db"}}) or on() vector(0)' f'max(ananke_ups_runtime_seconds{{{ANANKE_UPS_TETHYS_SELECTOR}}}) or on() vector(0)'
) )
HECATE_UPS_ON_BATTERY_TETHYS = ( ANANKE_UPS_ON_BATTERY_DB = (
f'max(hecate_ups_on_battery{{{HECATE_SELECTOR},instance="titan-24"}}) or on() vector(0)' f'max(ananke_ups_on_battery{{{ANANKE_UPS_DB_SELECTOR}}}) or on() vector(0)'
) )
HECATE_UPS_BATTERY_CHARGE_DB = ( ANANKE_UPS_ON_BATTERY_TETHYS = (
f'max(hecate_ups_battery_charge_percent{{{HECATE_SELECTOR},instance="titan-db"}}) or on() vector(0)' f'max(ananke_ups_on_battery{{{ANANKE_UPS_TETHYS_SELECTOR}}}) or on() vector(0)'
) )
HECATE_UPS_BATTERY_CHARGE_TETHYS = ( ANANKE_UPS_BATTERY_CHARGE_DB = (
f'max(hecate_ups_battery_charge_percent{{{HECATE_SELECTOR},instance="titan-24"}}) or on() vector(0)' f'max(ananke_ups_battery_charge_percent{{{ANANKE_UPS_DB_SELECTOR}}}) or on() vector(0)'
) )
HECATE_UPS_LOAD_DB = ( ANANKE_UPS_BATTERY_CHARGE_TETHYS = (
f'max(hecate_ups_load_percent{{{HECATE_SELECTOR},instance="titan-db"}}) or on() vector(0)' f'max(ananke_ups_battery_charge_percent{{{ANANKE_UPS_TETHYS_SELECTOR}}}) or on() vector(0)'
) )
HECATE_UPS_LOAD_TETHYS = ( ANANKE_UPS_LOAD_DB = (
f'max(hecate_ups_load_percent{{{HECATE_SELECTOR},instance="titan-24"}}) or on() vector(0)' f'max(ananke_ups_load_percent{{{ANANKE_UPS_DB_SELECTOR}}}) or on() vector(0)'
) )
HECATE_UPS_DRAW_WATTS_DB = ( ANANKE_UPS_LOAD_TETHYS = (
f'max((hecate_ups_load_percent{{{HECATE_SELECTOR},instance="titan-db"}} ' f'max(ananke_ups_load_percent{{{ANANKE_UPS_TETHYS_SELECTOR}}}) or on() vector(0)'
f'* hecate_ups_power_nominal_watts{{{HECATE_SELECTOR},instance="titan-db"}}) / 100) or on() vector(0)'
) )
HECATE_UPS_DRAW_WATTS_TETHYS = ( ANANKE_UPS_DRAW_WATTS_DB = (
f'max((hecate_ups_load_percent{{{HECATE_SELECTOR},instance="titan-24"}} ' f'max((ananke_ups_load_percent{{{ANANKE_UPS_DB_SELECTOR}}} '
f'* hecate_ups_power_nominal_watts{{{HECATE_SELECTOR},instance="titan-24"}}) / 100) or on() vector(0)' f'* ananke_ups_power_nominal_watts{{{ANANKE_UPS_DB_SELECTOR}}}) / 100) or on() vector(0)'
) )
HECATE_UPS_DRAW_WATTS_TOTAL = ( ANANKE_UPS_DRAW_WATTS_TETHYS = (
f'sum((hecate_ups_load_percent{{{HECATE_SELECTOR}}} * hecate_ups_power_nominal_watts{{{HECATE_SELECTOR}}}) / 100) ' f'max((ananke_ups_load_percent{{{ANANKE_UPS_TETHYS_SELECTOR}}} '
f'* ananke_ups_power_nominal_watts{{{ANANKE_UPS_TETHYS_SELECTOR}}}) / 100) or on() vector(0)'
)
ANANKE_UPS_DRAW_WATTS_TOTAL = (
f'sum((ananke_ups_load_percent{{{ANANKE_SELECTOR}}} * ananke_ups_power_nominal_watts{{{ANANKE_SELECTOR}}}) / 100) '
"or on() vector(0)" "or on() vector(0)"
) )
HECATE_UPS_DRAW_WATTS_DB_SERIES = ( ANANKE_UPS_DRAW_WATTS_DB_SERIES = (
f'((hecate_ups_load_percent{{{HECATE_SELECTOR},instance="titan-db"}} ' f'((ananke_ups_load_percent{{{ANANKE_UPS_DB_SELECTOR}}} '
f'* hecate_ups_power_nominal_watts{{{HECATE_SELECTOR},instance="titan-db"}}) / 100)' f'* ananke_ups_power_nominal_watts{{{ANANKE_UPS_DB_SELECTOR}}}) / 100)'
) )
HECATE_UPS_DRAW_WATTS_TETHYS_SERIES = ( ANANKE_UPS_DRAW_WATTS_TETHYS_SERIES = (
f'((hecate_ups_load_percent{{{HECATE_SELECTOR},instance="titan-24"}} ' f'((ananke_ups_load_percent{{{ANANKE_UPS_TETHYS_SELECTOR}}} '
f'* hecate_ups_power_nominal_watts{{{HECATE_SELECTOR},instance="titan-24"}}) / 100)' f'* ananke_ups_power_nominal_watts{{{ANANKE_UPS_TETHYS_SELECTOR}}}) / 100)'
) )
HECATE_UPS_DRAW_WATTS_TOTAL_SERIES = ( ANANKE_UPS_DRAW_WATTS_TOTAL_SERIES = (
f'sum((hecate_ups_load_percent{{{HECATE_SELECTOR}}} * hecate_ups_power_nominal_watts{{{HECATE_SELECTOR}}}) / 100)' f'sum((ananke_ups_load_percent{{{ANANKE_SELECTOR}}} * ananke_ups_power_nominal_watts{{{ANANKE_SELECTOR}}}) / 100)'
) )
HECATE_UPS_RUNTIME_BY_SOURCE = f"hecate_ups_runtime_seconds{{{HECATE_SELECTOR}}}" ANANKE_UPS_RUNTIME_BY_SOURCE = f"ananke_ups_runtime_seconds{{{ANANKE_SELECTOR}}}"
HECATE_UPS_LOAD_BY_SOURCE = f"hecate_ups_load_percent{{{HECATE_SELECTOR}}}" ANANKE_UPS_LOAD_BY_SOURCE = f"ananke_ups_load_percent{{{ANANKE_SELECTOR}}}"
HECATE_UPS_CHARGE_BY_SOURCE = f"hecate_ups_battery_charge_percent{{{HECATE_SELECTOR}}}" ANANKE_UPS_CHARGE_BY_SOURCE = f"ananke_ups_battery_charge_percent{{{ANANKE_SELECTOR}}}"
HECATE_UPS_TRIGGER_BY_SOURCE = f"hecate_ups_trigger_active{{{HECATE_SELECTOR}}}" ANANKE_UPS_TRIGGER_BY_SOURCE = f"ananke_ups_trigger_active{{{ANANKE_SELECTOR}}}"
CLIMATE_SENSOR_COUNT = "count(atlas_climate_temperature_celsius) or on() vector(0)" CLIMATE_SENSOR_COUNT = "count(atlas_climate_temperature_celsius) or on() vector(0)"
CLIMATE_TEMP_MAX = "max(atlas_climate_tent_temperature_celsius) or max(atlas_climate_temperature_celsius) or on() vector(0)" CLIMATE_TEMP_MAX = "max(atlas_climate_tent_temperature_celsius) or max(atlas_climate_temperature_celsius) or on() vector(0)"
CLIMATE_PRESSURE_CURRENT = "max(atlas_climate_tent_pressure_kpa) or max(atlas_climate_pressure_kpa) or on() vector(0)" CLIMATE_PRESSURE_CURRENT = "max(atlas_climate_tent_pressure_kpa) or max(atlas_climate_pressure_kpa) or on() vector(0)"
@ -1293,37 +1306,37 @@ def build_overview():
decimals=1, decimals=1,
text_mode="name_and_value", text_mode="name_and_value",
targets=[ targets=[
{"refId": "A", "expr": HECATE_UPS_DRAW_WATTS_DB, "legendFormat": f"{HECATE_UPS_DB_NAME} Draw (W)", "instant": True}, {"refId": "A", "expr": ANANKE_UPS_DRAW_WATTS_DB, "legendFormat": f"{ANANKE_UPS_DB_NAME} Draw (W)", "instant": True},
{"refId": "B", "expr": HECATE_UPS_RUNTIME_DB, "legendFormat": f"{HECATE_UPS_DB_NAME} Discharge ETA", "instant": True}, {"refId": "B", "expr": ANANKE_UPS_RUNTIME_DB, "legendFormat": f"{ANANKE_UPS_DB_NAME} Discharge ETA", "instant": True},
{"refId": "C", "expr": HECATE_UPS_ON_BATTERY_DB, "legendFormat": f"{HECATE_UPS_DB_NAME} Status", "instant": True}, {"refId": "C", "expr": ANANKE_UPS_ON_BATTERY_DB, "legendFormat": f"{ANANKE_UPS_DB_NAME} Status", "instant": True},
{"refId": "D", "expr": HECATE_UPS_DRAW_WATTS_TETHYS, "legendFormat": f"{HECATE_UPS_TETHYS_NAME} Draw (W)", "instant": True}, {"refId": "D", "expr": ANANKE_UPS_DRAW_WATTS_TETHYS, "legendFormat": f"{ANANKE_UPS_TETHYS_NAME} Draw (W)", "instant": True},
{"refId": "E", "expr": HECATE_UPS_RUNTIME_TETHYS, "legendFormat": f"{HECATE_UPS_TETHYS_NAME} Discharge ETA", "instant": True}, {"refId": "E", "expr": ANANKE_UPS_RUNTIME_TETHYS, "legendFormat": f"{ANANKE_UPS_TETHYS_NAME} Discharge ETA", "instant": True},
{"refId": "F", "expr": HECATE_UPS_ON_BATTERY_TETHYS, "legendFormat": f"{HECATE_UPS_TETHYS_NAME} Status", "instant": True}, {"refId": "F", "expr": ANANKE_UPS_ON_BATTERY_TETHYS, "legendFormat": f"{ANANKE_UPS_TETHYS_NAME} Status", "instant": True},
], ],
field_overrides=[ field_overrides=[
{ {
"matcher": {"id": "byName", "options": f"{HECATE_UPS_DB_NAME} Draw (W)"}, "matcher": {"id": "byName", "options": f"{ANANKE_UPS_DB_NAME} Draw (W)"},
"properties": [{"id": "unit", "value": "watt"}, {"id": "description", "value": f"Attached node: {HECATE_UPS_DB_NODE}"}], "properties": [{"id": "unit", "value": "watt"}, {"id": "description", "value": f"Attached node: {ANANKE_UPS_DB_NODE}"}],
}, },
{ {
"matcher": {"id": "byName", "options": f"{HECATE_UPS_TETHYS_NAME} Draw (W)"}, "matcher": {"id": "byName", "options": f"{ANANKE_UPS_TETHYS_NAME} Draw (W)"},
"properties": [{"id": "unit", "value": "watt"}, {"id": "description", "value": f"Attached node: {HECATE_UPS_TETHYS_NODE}"}], "properties": [{"id": "unit", "value": "watt"}, {"id": "description", "value": f"Attached node: {ANANKE_UPS_TETHYS_NODE}"}],
}, },
{ {
"matcher": {"id": "byName", "options": f"{HECATE_UPS_DB_NAME} Discharge ETA"}, "matcher": {"id": "byName", "options": f"{ANANKE_UPS_DB_NAME} Discharge ETA"},
"properties": [{"id": "unit", "value": "s"}, {"id": "description", "value": f"Attached node: {HECATE_UPS_DB_NODE}"}], "properties": [{"id": "unit", "value": "s"}, {"id": "description", "value": f"Attached node: {ANANKE_UPS_DB_NODE}"}],
}, },
{ {
"matcher": {"id": "byName", "options": f"{HECATE_UPS_TETHYS_NAME} Discharge ETA"}, "matcher": {"id": "byName", "options": f"{ANANKE_UPS_TETHYS_NAME} Discharge ETA"},
"properties": [{"id": "unit", "value": "s"}, {"id": "description", "value": f"Attached node: {HECATE_UPS_TETHYS_NODE}"}], "properties": [{"id": "unit", "value": "s"}, {"id": "description", "value": f"Attached node: {ANANKE_UPS_TETHYS_NODE}"}],
}, },
{ {
"matcher": {"id": "byName", "options": f"{HECATE_UPS_DB_NAME} Status"}, "matcher": {"id": "byName", "options": f"{ANANKE_UPS_DB_NAME} Status"},
"properties": [{"id": "mappings", "value": status_mapping}, {"id": "description", "value": f"Attached node: {HECATE_UPS_DB_NODE}"}], "properties": [{"id": "mappings", "value": status_mapping}, {"id": "description", "value": f"Attached node: {ANANKE_UPS_DB_NODE}"}],
}, },
{ {
"matcher": {"id": "byName", "options": f"{HECATE_UPS_TETHYS_NAME} Status"}, "matcher": {"id": "byName", "options": f"{ANANKE_UPS_TETHYS_NAME} Status"},
"properties": [{"id": "mappings", "value": status_mapping}, {"id": "description", "value": f"Attached node: {HECATE_UPS_TETHYS_NODE}"}], "properties": [{"id": "mappings", "value": status_mapping}, {"id": "description", "value": f"Attached node: {ANANKE_UPS_TETHYS_NODE}"}],
}, },
], ],
links=link_to("atlas-power"), links=link_to("atlas-power"),
@ -1338,9 +1351,9 @@ def build_overview():
{"h": 6, "w": 4, "x": 4, "y": 14}, {"h": 6, "w": 4, "x": 4, "y": 14},
unit="watt", unit="watt",
targets=[ targets=[
{"refId": "A", "expr": HECATE_UPS_DRAW_WATTS_DB_SERIES, "legendFormat": HECATE_UPS_DB_NAME}, {"refId": "A", "expr": ANANKE_UPS_DRAW_WATTS_DB_SERIES, "legendFormat": ANANKE_UPS_DB_NAME},
{"refId": "B", "expr": HECATE_UPS_DRAW_WATTS_TETHYS_SERIES, "legendFormat": HECATE_UPS_TETHYS_NAME}, {"refId": "B", "expr": ANANKE_UPS_DRAW_WATTS_TETHYS_SERIES, "legendFormat": ANANKE_UPS_TETHYS_NAME},
{"refId": "C", "expr": HECATE_UPS_DRAW_WATTS_TOTAL_SERIES, "legendFormat": "combined"}, {"refId": "C", "expr": ANANKE_UPS_DRAW_WATTS_TOTAL_SERIES, "legendFormat": "combined"},
], ],
legend_display="list", legend_display="list",
legend_placement="bottom", legend_placement="bottom",
@ -1470,27 +1483,26 @@ def build_overview():
links=link_to("atlas-jobs"), links=link_to("atlas-jobs"),
) )
) )
test_success = stat_panel( test_success = timeseries_panel(
46, 46,
"Platform Test Success Rate", "Platform Test Success Rate",
TEST_SUCCESS_RATE, None,
{"h": 3, "w": 6, "x": 12, "y": 11}, {"h": 3, "w": 6, "x": 12, "y": 11},
unit="percent", unit="percent",
decimals=2, targets=[
thresholds={ {
"mode": "absolute", "refId": "A",
"steps": [ "expr": PLATFORM_TEST_SUCCESS_RATE_BY_SUITE_SERIES,
{"color": "red", "value": None}, "legendFormat": "{{suite}}",
{"color": "orange", "value": 90}, }
{"color": "yellow", "value": 97}, ],
{"color": "green", "value": 99}, legend_display="list",
], legend_placement="bottom",
},
links=link_to("atlas-jobs"), links=link_to("atlas-jobs"),
) )
test_success["description"] = ( test_success["description"] = (
"Atlas Overview mirrors the Atlas Jobs internal dashboard for automation test health. " "Rolling pass rate per platform suite (Ariadne task automation + Metis build/flash pipelines). "
"This panel rolls up the shared Ariadne and Metis CI metrics from that internal dashboard." "Each line tracks pass percentage over time for its suite."
) )
panels.append(test_success) panels.append(test_success)
test_failures = stat_panel( test_failures = stat_panel(
@ -2956,6 +2968,26 @@ def build_jobs_dashboard():
"Atlas Overview test panels depend on this internal activity table sourced from Ariadne and Metis counters." "Atlas Overview test panels depend on this internal activity table sourced from Ariadne and Metis counters."
) )
panels.append(tests_panel) panels.append(tests_panel)
suite_panel = timeseries_panel(
19,
"Platform Test Success Rate by Suite",
None,
{"h": 6, "w": 16, "x": 8, "y": 17},
unit="percent",
targets=[
{
"refId": "A",
"expr": PLATFORM_TEST_SUCCESS_RATE_BY_SUITE_SERIES,
"legendFormat": "{{suite}}",
}
],
legend_display="list",
legend_placement="bottom",
)
suite_panel["description"] = (
"Per-suite pass percentage over time. Used by Atlas Overview and kept here for detailed triage."
)
panels.append(suite_panel)
return { return {
"uid": "atlas-jobs", "uid": "atlas-jobs",
@ -2993,37 +3025,37 @@ def build_power_dashboard():
decimals=1, decimals=1,
text_mode="name_and_value", text_mode="name_and_value",
targets=[ targets=[
{"refId": "A", "expr": HECATE_UPS_DRAW_WATTS_DB, "legendFormat": f"{HECATE_UPS_DB_NAME} Draw (W)", "instant": True}, {"refId": "A", "expr": ANANKE_UPS_DRAW_WATTS_DB, "legendFormat": f"{ANANKE_UPS_DB_NAME} Draw (W)", "instant": True},
{"refId": "B", "expr": HECATE_UPS_RUNTIME_DB, "legendFormat": f"{HECATE_UPS_DB_NAME} Discharge ETA", "instant": True}, {"refId": "B", "expr": ANANKE_UPS_RUNTIME_DB, "legendFormat": f"{ANANKE_UPS_DB_NAME} Discharge ETA", "instant": True},
{"refId": "C", "expr": HECATE_UPS_ON_BATTERY_DB, "legendFormat": f"{HECATE_UPS_DB_NAME} Status", "instant": True}, {"refId": "C", "expr": ANANKE_UPS_ON_BATTERY_DB, "legendFormat": f"{ANANKE_UPS_DB_NAME} Status", "instant": True},
{"refId": "D", "expr": HECATE_UPS_DRAW_WATTS_TETHYS, "legendFormat": f"{HECATE_UPS_TETHYS_NAME} Draw (W)", "instant": True}, {"refId": "D", "expr": ANANKE_UPS_DRAW_WATTS_TETHYS, "legendFormat": f"{ANANKE_UPS_TETHYS_NAME} Draw (W)", "instant": True},
{"refId": "E", "expr": HECATE_UPS_RUNTIME_TETHYS, "legendFormat": f"{HECATE_UPS_TETHYS_NAME} Discharge ETA", "instant": True}, {"refId": "E", "expr": ANANKE_UPS_RUNTIME_TETHYS, "legendFormat": f"{ANANKE_UPS_TETHYS_NAME} Discharge ETA", "instant": True},
{"refId": "F", "expr": HECATE_UPS_ON_BATTERY_TETHYS, "legendFormat": f"{HECATE_UPS_TETHYS_NAME} Status", "instant": True}, {"refId": "F", "expr": ANANKE_UPS_ON_BATTERY_TETHYS, "legendFormat": f"{ANANKE_UPS_TETHYS_NAME} Status", "instant": True},
], ],
field_overrides=[ field_overrides=[
{ {
"matcher": {"id": "byName", "options": f"{HECATE_UPS_DB_NAME} Draw (W)"}, "matcher": {"id": "byName", "options": f"{ANANKE_UPS_DB_NAME} Draw (W)"},
"properties": [{"id": "unit", "value": "watt"}, {"id": "description", "value": f"Attached node: {HECATE_UPS_DB_NODE}"}], "properties": [{"id": "unit", "value": "watt"}, {"id": "description", "value": f"Attached node: {ANANKE_UPS_DB_NODE}"}],
}, },
{ {
"matcher": {"id": "byName", "options": f"{HECATE_UPS_TETHYS_NAME} Draw (W)"}, "matcher": {"id": "byName", "options": f"{ANANKE_UPS_TETHYS_NAME} Draw (W)"},
"properties": [{"id": "unit", "value": "watt"}, {"id": "description", "value": f"Attached node: {HECATE_UPS_TETHYS_NODE}"}], "properties": [{"id": "unit", "value": "watt"}, {"id": "description", "value": f"Attached node: {ANANKE_UPS_TETHYS_NODE}"}],
}, },
{ {
"matcher": {"id": "byName", "options": f"{HECATE_UPS_DB_NAME} Discharge ETA"}, "matcher": {"id": "byName", "options": f"{ANANKE_UPS_DB_NAME} Discharge ETA"},
"properties": [{"id": "unit", "value": "s"}, {"id": "description", "value": f"Attached node: {HECATE_UPS_DB_NODE}"}], "properties": [{"id": "unit", "value": "s"}, {"id": "description", "value": f"Attached node: {ANANKE_UPS_DB_NODE}"}],
}, },
{ {
"matcher": {"id": "byName", "options": f"{HECATE_UPS_TETHYS_NAME} Discharge ETA"}, "matcher": {"id": "byName", "options": f"{ANANKE_UPS_TETHYS_NAME} Discharge ETA"},
"properties": [{"id": "unit", "value": "s"}, {"id": "description", "value": f"Attached node: {HECATE_UPS_TETHYS_NODE}"}], "properties": [{"id": "unit", "value": "s"}, {"id": "description", "value": f"Attached node: {ANANKE_UPS_TETHYS_NODE}"}],
}, },
{ {
"matcher": {"id": "byName", "options": f"{HECATE_UPS_DB_NAME} Status"}, "matcher": {"id": "byName", "options": f"{ANANKE_UPS_DB_NAME} Status"},
"properties": [{"id": "mappings", "value": status_mapping}, {"id": "description", "value": f"Attached node: {HECATE_UPS_DB_NODE}"}], "properties": [{"id": "mappings", "value": status_mapping}, {"id": "description", "value": f"Attached node: {ANANKE_UPS_DB_NODE}"}],
}, },
{ {
"matcher": {"id": "byName", "options": f"{HECATE_UPS_TETHYS_NAME} Status"}, "matcher": {"id": "byName", "options": f"{ANANKE_UPS_TETHYS_NAME} Status"},
"properties": [{"id": "mappings", "value": status_mapping}, {"id": "description", "value": f"Attached node: {HECATE_UPS_TETHYS_NODE}"}], "properties": [{"id": "mappings", "value": status_mapping}, {"id": "description", "value": f"Attached node: {ANANKE_UPS_TETHYS_NODE}"}],
}, },
], ],
description=( description=(
@ -3039,9 +3071,9 @@ def build_power_dashboard():
{"h": 8, "w": 12, "x": 12, "y": 0}, {"h": 8, "w": 12, "x": 12, "y": 0},
unit="watt", unit="watt",
targets=[ targets=[
{"refId": "A", "expr": HECATE_UPS_DRAW_WATTS_DB_SERIES, "legendFormat": HECATE_UPS_DB_NAME}, {"refId": "A", "expr": ANANKE_UPS_DRAW_WATTS_DB_SERIES, "legendFormat": ANANKE_UPS_DB_NAME},
{"refId": "B", "expr": HECATE_UPS_DRAW_WATTS_TETHYS_SERIES, "legendFormat": HECATE_UPS_TETHYS_NAME}, {"refId": "B", "expr": ANANKE_UPS_DRAW_WATTS_TETHYS_SERIES, "legendFormat": ANANKE_UPS_TETHYS_NAME},
{"refId": "C", "expr": HECATE_UPS_DRAW_WATTS_TOTAL_SERIES, "legendFormat": "combined"}, {"refId": "C", "expr": ANANKE_UPS_DRAW_WATTS_TOTAL_SERIES, "legendFormat": "combined"},
], ],
legend_display="table", legend_display="table",
legend_placement="right", legend_placement="right",

38
services/monitoring/dashboards/atlas-jobs.json
View File

@ -1235,6 +1235,44 @@
} }
], ],
"description": "Atlas Overview test panels depend on this internal activity table sourced from Ariadne and Metis counters." "description": "Atlas Overview test panels depend on this internal activity table sourced from Ariadne and Metis counters."
},
{
"id": 19,
"type": "timeseries",
"title": "Platform Test Success Rate by Suite",
"datasource": {
"type": "prometheus",
"uid": "atlas-vm"
},
"gridPos": {
"h": 6,
"w": 16,
"x": 8,
"y": 17
},
"targets": [
{
"refId": "A",
"expr": "label_replace(100 * (sum(increase(ariadne_task_runs_total{status=\"ok\"}[$__interval])) or on() vector(0)) / clamp_min((sum(increase(ariadne_task_runs_total[$__interval])) or on() vector(0)), 1), \"suite\", \"ariadne\", \"__name__\", \".*\") or label_replace(100 * (sum(increase(metis_builds_total{status=\"ok\"}[$__interval])) or on() vector(0)) / clamp_min((sum(increase(metis_builds_total[$__interval])) or on() vector(0)), 1), \"suite\", \"metis-build\", \"__name__\", \".*\") or label_replace(100 * (sum(increase(metis_flashes_total{status=\"ok\"}[$__interval])) or on() vector(0)) / clamp_min((sum(increase(metis_flashes_total[$__interval])) or on() vector(0)), 1), \"suite\", \"metis-flash\", \"__name__\", \".*\")",
"legendFormat": "{{suite}}"
}
],
"fieldConfig": {
"defaults": {
"unit": "percent"
},
"overrides": []
},
"options": {
"legend": {
"displayMode": "list",
"placement": "bottom"
},
"tooltip": {
"mode": "multi"
}
},
"description": "Per-suite pass percentage over time. Used by Atlas Overview and kept here for detailed triage."
} }
], ],
"time": { "time": {

74
services/monitoring/dashboards/atlas-overview.json
View File

@ -1087,37 +1087,37 @@
"targets": [ "targets": [
{ {
"refId": "A", "refId": "A",
"expr": "max((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-db\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-db\"}) / 100) or on() vector(0)", "expr": "max((ananke_ups_load_percent{job=\"ananke-power\",source=\"Pyrphoros\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Pyrphoros\"}) / 100) or on() vector(0)",
"legendFormat": "Pyrphoros Draw (W)", "legendFormat": "Pyrphoros Draw (W)",
"instant": true "instant": true
}, },
{ {
"refId": "B", "refId": "B",
"expr": "max(hecate_ups_runtime_seconds{job=\"hecate-power\",instance=\"titan-db\"}) or on() vector(0)", "expr": "max(ananke_ups_runtime_seconds{job=\"ananke-power\",source=\"Pyrphoros\"}) or on() vector(0)",
"legendFormat": "Pyrphoros Discharge ETA", "legendFormat": "Pyrphoros Discharge ETA",
"instant": true "instant": true
}, },
{ {
"refId": "C", "refId": "C",
"expr": "max(hecate_ups_on_battery{job=\"hecate-power\",instance=\"titan-db\"}) or on() vector(0)", "expr": "max(ananke_ups_on_battery{job=\"ananke-power\",source=\"Pyrphoros\"}) or on() vector(0)",
"legendFormat": "Pyrphoros Status", "legendFormat": "Pyrphoros Status",
"instant": true "instant": true
}, },
{ {
"refId": "D", "refId": "D",
"expr": "max((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-24\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-24\"}) / 100) or on() vector(0)", "expr": "max((ananke_ups_load_percent{job=\"ananke-power\",source=\"Statera\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Statera\"}) / 100) or on() vector(0)",
"legendFormat": "Statera Draw (W)", "legendFormat": "Statera Draw (W)",
"instant": true "instant": true
}, },
{ {
"refId": "E", "refId": "E",
"expr": "max(hecate_ups_runtime_seconds{job=\"hecate-power\",instance=\"titan-24\"}) or on() vector(0)", "expr": "max(ananke_ups_runtime_seconds{job=\"ananke-power\",source=\"Statera\"}) or on() vector(0)",
"legendFormat": "Statera Discharge ETA", "legendFormat": "Statera Discharge ETA",
"instant": true "instant": true
}, },
{ {
"refId": "F", "refId": "F",
"expr": "max(hecate_ups_on_battery{job=\"hecate-power\",instance=\"titan-24\"}) or on() vector(0)", "expr": "max(ananke_ups_on_battery{job=\"ananke-power\",source=\"Statera\"}) or on() vector(0)",
"legendFormat": "Statera Status", "legendFormat": "Statera Status",
"instant": true "instant": true
} }
@ -1309,17 +1309,17 @@
"targets": [ "targets": [
{ {
"refId": "A", "refId": "A",
"expr": "((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-db\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-db\"}) / 100)", "expr": "((ananke_ups_load_percent{job=\"ananke-power\",source=\"Pyrphoros\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Pyrphoros\"}) / 100)",
"legendFormat": "Pyrphoros" "legendFormat": "Pyrphoros"
}, },
{ {
"refId": "B", "refId": "B",
"expr": "((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-24\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-24\"}) / 100)", "expr": "((ananke_ups_load_percent{job=\"ananke-power\",source=\"Statera\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Statera\"}) / 100)",
"legendFormat": "Statera" "legendFormat": "Statera"
}, },
{ {
"refId": "C", "refId": "C",
"expr": "sum((hecate_ups_load_percent{job=\"hecate-power\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\"}) / 100)", "expr": "sum((ananke_ups_load_percent{job=\"ananke-power\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\"}) / 100)",
"legendFormat": "combined" "legendFormat": "combined"
} }
], ],
@ -1805,7 +1805,7 @@
}, },
{ {
"id": 46, "id": 46,
"type": "stat", "type": "timeseries",
"title": "Platform Test Success Rate", "title": "Platform Test Success Rate",
"datasource": { "datasource": {
"type": "prometheus", "type": "prometheus",
@ -1819,57 +1819,25 @@
}, },
"targets": [ "targets": [
{ {
"expr": "100 * ((sum(increase(ariadne_task_runs_total{status=\"ok\"}[30d])) or on() vector(0)) + (sum(increase(metis_builds_total{status=\"ok\"}[30d])) or on() vector(0)) + (sum(increase(metis_flashes_total{status=\"ok\"}[30d])) or on() vector(0))) / clamp_min(((sum(increase(ariadne_task_runs_total[30d])) or on() vector(0)) + (sum(increase(metis_builds_total[30d])) or on() vector(0)) + (sum(increase(metis_flashes_total[30d])) or on() vector(0))), 1)", "refId": "A",
"refId": "A" "expr": "label_replace(100 * (sum(increase(ariadne_task_runs_total{status=\"ok\"}[$__interval])) or on() vector(0)) / clamp_min((sum(increase(ariadne_task_runs_total[$__interval])) or on() vector(0)), 1), \"suite\", \"ariadne\", \"__name__\", \".*\") or label_replace(100 * (sum(increase(metis_builds_total{status=\"ok\"}[$__interval])) or on() vector(0)) / clamp_min((sum(increase(metis_builds_total[$__interval])) or on() vector(0)), 1), \"suite\", \"metis-build\", \"__name__\", \".*\") or label_replace(100 * (sum(increase(metis_flashes_total{status=\"ok\"}[$__interval])) or on() vector(0)) / clamp_min((sum(increase(metis_flashes_total[$__interval])) or on() vector(0)), 1), \"suite\", \"metis-flash\", \"__name__\", \".*\")",
"legendFormat": "{{suite}}"
} }
], ],
"fieldConfig": { "fieldConfig": {
"defaults": { "defaults": {
"color": { "unit": "percent"
"mode": "thresholds"
},
"mappings": [],
"thresholds": {
"mode": "absolute",
"steps": [
{
"color": "red",
"value": null
},
{
"color": "orange",
"value": 90
},
{
"color": "yellow",
"value": 97
},
{
"color": "green",
"value": 99
}
]
},
"unit": "percent",
"custom": {
"displayMode": "auto"
},
"decimals": 2
}, },
"overrides": [] "overrides": []
}, },
"options": { "options": {
"colorMode": "value", "legend": {
"graphMode": "area", "displayMode": "list",
"justifyMode": "center", "placement": "bottom"
"reduceOptions": {
"calcs": [
"lastNotNull"
],
"fields": "",
"values": false
}, },
"textMode": "value" "tooltip": {
"mode": "multi"
}
}, },
"links": [ "links": [
{ {
@ -1878,7 +1846,7 @@
"targetBlank": true "targetBlank": true
} }
], ],
"description": "Atlas Overview mirrors the Atlas Jobs internal dashboard for automation test health. This panel rolls up the shared Ariadne and Metis CI metrics from that internal dashboard." "description": "Rolling pass rate per platform suite (Ariadne task automation + Metis build/flash pipelines). Each line tracks pass percentage over time for its suite."
}, },
{ {
"id": 47, "id": 47,

18
services/monitoring/dashboards/atlas-power.json
View File

@ -21,37 +21,37 @@
"targets": [ "targets": [
{ {
"refId": "A", "refId": "A",
"expr": "max((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-db\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-db\"}) / 100) or on() vector(0)", "expr": "max((ananke_ups_load_percent{job=\"ananke-power\",source=\"Pyrphoros\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Pyrphoros\"}) / 100) or on() vector(0)",
"legendFormat": "Pyrphoros Draw (W)", "legendFormat": "Pyrphoros Draw (W)",
"instant": true "instant": true
}, },
{ {
"refId": "B", "refId": "B",
"expr": "max(hecate_ups_runtime_seconds{job=\"hecate-power\",instance=\"titan-db\"}) or on() vector(0)", "expr": "max(ananke_ups_runtime_seconds{job=\"ananke-power\",source=\"Pyrphoros\"}) or on() vector(0)",
"legendFormat": "Pyrphoros Discharge ETA", "legendFormat": "Pyrphoros Discharge ETA",
"instant": true "instant": true
}, },
{ {
"refId": "C", "refId": "C",
"expr": "max(hecate_ups_on_battery{job=\"hecate-power\",instance=\"titan-db\"}) or on() vector(0)", "expr": "max(ananke_ups_on_battery{job=\"ananke-power\",source=\"Pyrphoros\"}) or on() vector(0)",
"legendFormat": "Pyrphoros Status", "legendFormat": "Pyrphoros Status",
"instant": true "instant": true
}, },
{ {
"refId": "D", "refId": "D",
"expr": "max((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-24\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-24\"}) / 100) or on() vector(0)", "expr": "max((ananke_ups_load_percent{job=\"ananke-power\",source=\"Statera\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Statera\"}) / 100) or on() vector(0)",
"legendFormat": "Statera Draw (W)", "legendFormat": "Statera Draw (W)",
"instant": true "instant": true
}, },
{ {
"refId": "E", "refId": "E",
"expr": "max(hecate_ups_runtime_seconds{job=\"hecate-power\",instance=\"titan-24\"}) or on() vector(0)", "expr": "max(ananke_ups_runtime_seconds{job=\"ananke-power\",source=\"Statera\"}) or on() vector(0)",
"legendFormat": "Statera Discharge ETA", "legendFormat": "Statera Discharge ETA",
"instant": true "instant": true
}, },
{ {
"refId": "F", "refId": "F",
"expr": "max(hecate_ups_on_battery{job=\"hecate-power\",instance=\"titan-24\"}) or on() vector(0)", "expr": "max(ananke_ups_on_battery{job=\"ananke-power\",source=\"Statera\"}) or on() vector(0)",
"legendFormat": "Statera Status", "legendFormat": "Statera Status",
"instant": true "instant": true
} }
@ -236,17 +236,17 @@
"targets": [ "targets": [
{ {
"refId": "A", "refId": "A",
"expr": "((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-db\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-db\"}) / 100)", "expr": "((ananke_ups_load_percent{job=\"ananke-power\",source=\"Pyrphoros\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Pyrphoros\"}) / 100)",
"legendFormat": "Pyrphoros" "legendFormat": "Pyrphoros"
}, },
{ {
"refId": "B", "refId": "B",
"expr": "((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-24\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-24\"}) / 100)", "expr": "((ananke_ups_load_percent{job=\"ananke-power\",source=\"Statera\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Statera\"}) / 100)",
"legendFormat": "Statera" "legendFormat": "Statera"
}, },
{ {
"refId": "C", "refId": "C",
"expr": "sum((hecate_ups_load_percent{job=\"hecate-power\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\"}) / 100)", "expr": "sum((ananke_ups_load_percent{job=\"ananke-power\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\"}) / 100)",
"legendFormat": "combined" "legendFormat": "combined"
} }
], ],

38
services/monitoring/grafana-dashboard-jobs.yaml
View File

@ -1244,6 +1244,44 @@ data:
} }
], ],
"description": "Atlas Overview test panels depend on this internal activity table sourced from Ariadne and Metis counters." "description": "Atlas Overview test panels depend on this internal activity table sourced from Ariadne and Metis counters."
},
{
"id": 19,
"type": "timeseries",
"title": "Platform Test Success Rate by Suite",
"datasource": {
"type": "prometheus",
"uid": "atlas-vm"
},
"gridPos": {
"h": 6,
"w": 16,
"x": 8,
"y": 17
},
"targets": [
{
"refId": "A",
"expr": "label_replace(100 * (sum(increase(ariadne_task_runs_total{status=\"ok\"}[$__interval])) or on() vector(0)) / clamp_min((sum(increase(ariadne_task_runs_total[$__interval])) or on() vector(0)), 1), \"suite\", \"ariadne\", \"__name__\", \".*\") or label_replace(100 * (sum(increase(metis_builds_total{status=\"ok\"}[$__interval])) or on() vector(0)) / clamp_min((sum(increase(metis_builds_total[$__interval])) or on() vector(0)), 1), \"suite\", \"metis-build\", \"__name__\", \".*\") or label_replace(100 * (sum(increase(metis_flashes_total{status=\"ok\"}[$__interval])) or on() vector(0)) / clamp_min((sum(increase(metis_flashes_total[$__interval])) or on() vector(0)), 1), \"suite\", \"metis-flash\", \"__name__\", \".*\")",
"legendFormat": "{{suite}}"
}
],
"fieldConfig": {
"defaults": {
"unit": "percent"
},
"overrides": []
},
"options": {
"legend": {
"displayMode": "list",
"placement": "bottom"
},
"tooltip": {
"mode": "multi"
}
},
"description": "Per-suite pass percentage over time. Used by Atlas Overview and kept here for detailed triage."
} }
], ],
"time": { "time": {

74
services/monitoring/grafana-dashboard-overview.yaml
View File

@ -1096,37 +1096,37 @@ data:
"targets": [ "targets": [
{ {
"refId": "A", "refId": "A",
"expr": "max((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-db\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-db\"}) / 100) or on() vector(0)", "expr": "max((ananke_ups_load_percent{job=\"ananke-power\",source=\"Pyrphoros\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Pyrphoros\"}) / 100) or on() vector(0)",
"legendFormat": "Pyrphoros Draw (W)", "legendFormat": "Pyrphoros Draw (W)",
"instant": true "instant": true
}, },
{ {
"refId": "B", "refId": "B",
"expr": "max(hecate_ups_runtime_seconds{job=\"hecate-power\",instance=\"titan-db\"}) or on() vector(0)", "expr": "max(ananke_ups_runtime_seconds{job=\"ananke-power\",source=\"Pyrphoros\"}) or on() vector(0)",
"legendFormat": "Pyrphoros Discharge ETA", "legendFormat": "Pyrphoros Discharge ETA",
"instant": true "instant": true
}, },
{ {
"refId": "C", "refId": "C",
"expr": "max(hecate_ups_on_battery{job=\"hecate-power\",instance=\"titan-db\"}) or on() vector(0)", "expr": "max(ananke_ups_on_battery{job=\"ananke-power\",source=\"Pyrphoros\"}) or on() vector(0)",
"legendFormat": "Pyrphoros Status", "legendFormat": "Pyrphoros Status",
"instant": true "instant": true
}, },
{ {
"refId": "D", "refId": "D",
"expr": "max((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-24\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-24\"}) / 100) or on() vector(0)", "expr": "max((ananke_ups_load_percent{job=\"ananke-power\",source=\"Statera\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Statera\"}) / 100) or on() vector(0)",
"legendFormat": "Statera Draw (W)", "legendFormat": "Statera Draw (W)",
"instant": true "instant": true
}, },
{ {
"refId": "E", "refId": "E",
"expr": "max(hecate_ups_runtime_seconds{job=\"hecate-power\",instance=\"titan-24\"}) or on() vector(0)", "expr": "max(ananke_ups_runtime_seconds{job=\"ananke-power\",source=\"Statera\"}) or on() vector(0)",
"legendFormat": "Statera Discharge ETA", "legendFormat": "Statera Discharge ETA",
"instant": true "instant": true
}, },
{ {
"refId": "F", "refId": "F",
"expr": "max(hecate_ups_on_battery{job=\"hecate-power\",instance=\"titan-24\"}) or on() vector(0)", "expr": "max(ananke_ups_on_battery{job=\"ananke-power\",source=\"Statera\"}) or on() vector(0)",
"legendFormat": "Statera Status", "legendFormat": "Statera Status",
"instant": true "instant": true
} }
@ -1318,17 +1318,17 @@ data:
"targets": [ "targets": [
{ {
"refId": "A", "refId": "A",
"expr": "((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-db\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-db\"}) / 100)", "expr": "((ananke_ups_load_percent{job=\"ananke-power\",source=\"Pyrphoros\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Pyrphoros\"}) / 100)",
"legendFormat": "Pyrphoros" "legendFormat": "Pyrphoros"
}, },
{ {
"refId": "B", "refId": "B",
"expr": "((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-24\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-24\"}) / 100)", "expr": "((ananke_ups_load_percent{job=\"ananke-power\",source=\"Statera\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Statera\"}) / 100)",
"legendFormat": "Statera" "legendFormat": "Statera"
}, },
{ {
"refId": "C", "refId": "C",
"expr": "sum((hecate_ups_load_percent{job=\"hecate-power\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\"}) / 100)", "expr": "sum((ananke_ups_load_percent{job=\"ananke-power\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\"}) / 100)",
"legendFormat": "combined" "legendFormat": "combined"
} }
], ],
@ -1814,7 +1814,7 @@ data:
}, },
{ {
"id": 46, "id": 46,
"type": "stat", "type": "timeseries",
"title": "Platform Test Success Rate", "title": "Platform Test Success Rate",
"datasource": { "datasource": {
"type": "prometheus", "type": "prometheus",
@ -1828,57 +1828,25 @@ data:
}, },
"targets": [ "targets": [
{ {
"expr": "100 * ((sum(increase(ariadne_task_runs_total{status=\"ok\"}[30d])) or on() vector(0)) + (sum(increase(metis_builds_total{status=\"ok\"}[30d])) or on() vector(0)) + (sum(increase(metis_flashes_total{status=\"ok\"}[30d])) or on() vector(0))) / clamp_min(((sum(increase(ariadne_task_runs_total[30d])) or on() vector(0)) + (sum(increase(metis_builds_total[30d])) or on() vector(0)) + (sum(increase(metis_flashes_total[30d])) or on() vector(0))), 1)", "refId": "A",
"refId": "A" "expr": "label_replace(100 * (sum(increase(ariadne_task_runs_total{status=\"ok\"}[$__interval])) or on() vector(0)) / clamp_min((sum(increase(ariadne_task_runs_total[$__interval])) or on() vector(0)), 1), \"suite\", \"ariadne\", \"__name__\", \".*\") or label_replace(100 * (sum(increase(metis_builds_total{status=\"ok\"}[$__interval])) or on() vector(0)) / clamp_min((sum(increase(metis_builds_total[$__interval])) or on() vector(0)), 1), \"suite\", \"metis-build\", \"__name__\", \".*\") or label_replace(100 * (sum(increase(metis_flashes_total{status=\"ok\"}[$__interval])) or on() vector(0)) / clamp_min((sum(increase(metis_flashes_total[$__interval])) or on() vector(0)), 1), \"suite\", \"metis-flash\", \"__name__\", \".*\")",
"legendFormat": "{{suite}}"
} }
], ],
"fieldConfig": { "fieldConfig": {
"defaults": { "defaults": {
"color": { "unit": "percent"
"mode": "thresholds"
},
"mappings": [],
"thresholds": {
"mode": "absolute",
"steps": [
{
"color": "red",
"value": null
},
{
"color": "orange",
"value": 90
},
{
"color": "yellow",
"value": 97
},
{
"color": "green",
"value": 99
}
]
},
"unit": "percent",
"custom": {
"displayMode": "auto"
},
"decimals": 2
}, },
"overrides": [] "overrides": []
}, },
"options": { "options": {
"colorMode": "value", "legend": {
"graphMode": "area", "displayMode": "list",
"justifyMode": "center", "placement": "bottom"
"reduceOptions": {
"calcs": [
"lastNotNull"
],
"fields": "",
"values": false
}, },
"textMode": "value" "tooltip": {
"mode": "multi"
}
}, },
"links": [ "links": [
{ {
@ -1887,7 +1855,7 @@ data:
"targetBlank": true "targetBlank": true
} }
], ],
"description": "Atlas Overview mirrors the Atlas Jobs internal dashboard for automation test health. This panel rolls up the shared Ariadne and Metis CI metrics from that internal dashboard." "description": "Rolling pass rate per platform suite (Ariadne task automation + Metis build/flash pipelines). Each line tracks pass percentage over time for its suite."
}, },
{ {
"id": 47, "id": 47,

18
services/monitoring/grafana-dashboard-power.yaml
View File

@ -30,37 +30,37 @@ data:
"targets": [ "targets": [
{ {
"refId": "A", "refId": "A",
"expr": "max((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-db\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-db\"}) / 100) or on() vector(0)", "expr": "max((ananke_ups_load_percent{job=\"ananke-power\",source=\"Pyrphoros\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Pyrphoros\"}) / 100) or on() vector(0)",
"legendFormat": "Pyrphoros Draw (W)", "legendFormat": "Pyrphoros Draw (W)",
"instant": true "instant": true
}, },
{ {
"refId": "B", "refId": "B",
"expr": "max(hecate_ups_runtime_seconds{job=\"hecate-power\",instance=\"titan-db\"}) or on() vector(0)", "expr": "max(ananke_ups_runtime_seconds{job=\"ananke-power\",source=\"Pyrphoros\"}) or on() vector(0)",
"legendFormat": "Pyrphoros Discharge ETA", "legendFormat": "Pyrphoros Discharge ETA",
"instant": true "instant": true
}, },
{ {
"refId": "C", "refId": "C",
"expr": "max(hecate_ups_on_battery{job=\"hecate-power\",instance=\"titan-db\"}) or on() vector(0)", "expr": "max(ananke_ups_on_battery{job=\"ananke-power\",source=\"Pyrphoros\"}) or on() vector(0)",
"legendFormat": "Pyrphoros Status", "legendFormat": "Pyrphoros Status",
"instant": true "instant": true
}, },
{ {
"refId": "D", "refId": "D",
"expr": "max((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-24\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-24\"}) / 100) or on() vector(0)", "expr": "max((ananke_ups_load_percent{job=\"ananke-power\",source=\"Statera\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Statera\"}) / 100) or on() vector(0)",
"legendFormat": "Statera Draw (W)", "legendFormat": "Statera Draw (W)",
"instant": true "instant": true
}, },
{ {
"refId": "E", "refId": "E",
"expr": "max(hecate_ups_runtime_seconds{job=\"hecate-power\",instance=\"titan-24\"}) or on() vector(0)", "expr": "max(ananke_ups_runtime_seconds{job=\"ananke-power\",source=\"Statera\"}) or on() vector(0)",
"legendFormat": "Statera Discharge ETA", "legendFormat": "Statera Discharge ETA",
"instant": true "instant": true
}, },
{ {
"refId": "F", "refId": "F",
"expr": "max(hecate_ups_on_battery{job=\"hecate-power\",instance=\"titan-24\"}) or on() vector(0)", "expr": "max(ananke_ups_on_battery{job=\"ananke-power\",source=\"Statera\"}) or on() vector(0)",
"legendFormat": "Statera Status", "legendFormat": "Statera Status",
"instant": true "instant": true
} }
@ -245,17 +245,17 @@ data:
"targets": [ "targets": [
{ {
"refId": "A", "refId": "A",
"expr": "((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-db\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-db\"}) / 100)", "expr": "((ananke_ups_load_percent{job=\"ananke-power\",source=\"Pyrphoros\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Pyrphoros\"}) / 100)",
"legendFormat": "Pyrphoros" "legendFormat": "Pyrphoros"
}, },
{ {
"refId": "B", "refId": "B",
"expr": "((hecate_ups_load_percent{job=\"hecate-power\",instance=\"titan-24\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\",instance=\"titan-24\"}) / 100)", "expr": "((ananke_ups_load_percent{job=\"ananke-power\",source=\"Statera\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\",source=\"Statera\"}) / 100)",
"legendFormat": "Statera" "legendFormat": "Statera"
}, },
{ {
"refId": "C", "refId": "C",
"expr": "sum((hecate_ups_load_percent{job=\"hecate-power\"} * hecate_ups_power_nominal_watts{job=\"hecate-power\"}) / 100)", "expr": "sum((ananke_ups_load_percent{job=\"ananke-power\"} * ananke_ups_power_nominal_watts{job=\"ananke-power\"}) / 100)",
"legendFormat": "combined" "legendFormat": "combined"
} }
], ],

4
services/monitoring/helmrelease.yaml
View File

@ -224,8 +224,8 @@ spec:
target_label: instance target_label: instance
replacement: titan-jh replacement: titan-jh
# --- Hecate power telemetry (host-level daemon on UPS hosts) --- # --- Ananke power telemetry (host-level daemon on UPS hosts) ---
- job_name: "hecate-power" - job_name: "ananke-power"
static_configs: static_configs:
- targets: ["192.168.22.10:9560"] - targets: ["192.168.22.10:9560"]
labels: labels: