Merge branch 'main' into enterprise

* main:
  Added temporary rescale guard clause
  (fix): energy unit was falsely omitted in badge
  Bump fastapi[standard] from 0.115.4 to 0.115.5 (#978)
  Made cluster accuracy warning more verbose
  Removing duplicate value check in cluster
  Bump tqdm from 4.66.6 to 4.67.0 (#975)

api/api_helpers.py

@@ -57,7 +57,12 @@ def store_artifact(artifact_type: Enum, key:str, data, ex=2592000):
 # The same unit for proper comparison!
 #
 def rescale_energy_value(value, unit):
-    # We only expect values to be mJ for energy!
+    if unit == 'mJ':
+        value = value * 1_000
+        unit = 'uJ'
+
+    # We only expect values to be uJ for energy in the future. Changing values now temporarily.
+    # TODO: Refactor this once all data in the DB is uJ
     if unit != 'uJ' and not unit.startswith('ugCO2e/'):
         raise RuntimeError('Unexpected unit occured for energy rescaling: ', unit)
 

api/main.py

@@ -512,7 +512,7 @@ async def get_badge_single(run_id: str, metric: str = 'ml-estimated'):
     if data is None or data == [] or data[1] is None: # special check for data[1] as this is aggregate query which always returns result
         badge_value = 'No energy data yet'
     else:
-        [energy_value, energy_unit] = rescale_energy_value(data[0], 'uJ')
+        [energy_value, energy_unit] = rescale_energy_value(data[0], data[1])
         badge_value= f"{energy_value:.2f} {energy_unit} {via}"
 
     badge = anybadge.Badge(

config.yml.example

@@ -58,9 +58,6 @@ cluster:
         cpu_energy_rapl_msr_component:
           threshold: 0.01 # 1%
           type: stddev_rel
-        cpu_energy_rapl_msr_component:
-          threshold: 0.01 # 1%
-          type: stddev_rel
         psu_carbon_ac_mcp_machine:
           threshold: 0.01 # 1%
           type: stddev_rel

docker/requirements.txt

@@ -1,7 +1,7 @@
 gunicorn==23.0.0
 psycopg[binary]==3.2.3
 psycopg_pool==3.2.3
-fastapi[standard]==0.115.4
+fastapi[standard]==0.115.5
 starlette>=0.35
 uvicorn[standard]==0.32.0
 pandas==2.2.3

lib/validate.py

@@ -100,11 +100,11 @@ def validate_workload_stddev(data, metrics):
 
         if metrics[el['metric']]['type'] == 'stddev_rel':
             if el['stddev_rel'] > metrics[el['metric']]['threshold']:
-                print(TerminalColors.FAIL, 'Warning. Threshold exceeded!', TerminalColors.ENDC)
+                info_string_acc.append(f"=> Warning! Threshold of {metrics[el['metric']]['threshold']} exceeded. Value is: {el['stddev_rel']}. Metric: {el['metric']}")
                 warning = True
         elif metrics[el['metric']]['type'] == 'stddev':
             if el['stddev'] > metrics[el['metric']]['threshold']:
-                print(TerminalColors.FAIL, 'Warning. Threshold exceeded!', TerminalColors.ENDC)
+                info_string_acc.append(f"=> Warning! Threshold of {metrics[el['metric']]['threshold']} exceeded. Value is: {el['stddev']}. Metric: {el['metric']}")
                 warning = True
         else:
             raise ValueError(f"{el['metric']} had unknown threshhold validation type: {metrics[el['metric']]['type']}")

requirements.txt

@@ -8,6 +8,6 @@ schema==0.7.7
 aiohttp==3.10.10
 
 # calibration script dep
-tqdm==4.66.6
+tqdm==4.67.0
 plotext==5.3.2
 docker==7.1.0