introduce a waste_total metric that normalizes CPU and Mem waste

slurm-eff-tool.py

@@ -84,6 +84,7 @@ ALL_COLUMNS = [
     "Walltime",
     "Walltime_max",
     "Time_Eff",
+    "waste_total",
     "jobname",
 ]
 
@@ -157,6 +158,21 @@ PRESET_COLUMNS = {
         "Mem_Eff",
         "waste_Mem",
         "jobname",],
+    "waste": [
+        "username",
+        "JobID",
+        "state",
+        "Count",
+        "NTasks",
+        "CPUs",
+        "Nodes",
+        "CPU_Eff",
+        "Mem_Eff",
+        "Time_Eff",
+        "waste_CPU",
+        "waste_Mem",
+        "waste_total",
+        "jobname",],
 }
 
 # One-character aliases for sorting and output format specifications.
@@ -176,6 +192,7 @@ ALIASES = {
     "j": "jobname",
     "X": "waste_CPU",
     "Y": "waste_Mem",
+    "T": "waste_total",
 }
 
 NUMERIC_COLUMNS = {
@@ -195,6 +212,7 @@ NUMERIC_COLUMNS = {
     "Walltime_max",
     "waste_CPU",
     "waste_Mem",
+    "waste_total",
 }
 
 state_mappings = {
@@ -419,6 +437,7 @@ class OutputRow:
     walltime_max: float | None   # in h
     waste_CPU: float | None   # CPU h
     waste_Mem: float | None
+    waste_total: float | None
     _cpu_eff_values: list[float] = field(default_factory=list, repr=False)
     _mem_eff_values: list[float] = field(default_factory=list, repr=False)
     _time_eff_values: list[float] = field(default_factory=list, repr=False)
@@ -446,6 +465,7 @@ class OutputRow:
             "jobname": self.jobname,
             "waste_CPU": self.waste_CPU,
             "waste_Mem": self.waste_Mem,
+            "waste_total": self.waste_total
         }
         if sdev:
             for col, vals in [
@@ -893,12 +913,12 @@ def make_single_row(rec: JobRecord, dflt_mpcpu: float) -> OutputRow:
     """Returns an OutputRow based on a single slurm job record."""
     walltime = rec.elapsed_sec / 3600
 
-    waste_mem = None
+    waste_mem = 0
     if rec.mem_eff is not None and rec.reqmem_gb is not None:
         waste_mem = max(0,walltime * (100-rec.mem_eff)/100 \
             * (rec.reqmem_gb - rec.cpus * dflt_mpcpu))
 
-    waste_cpu = None
+    waste_cpu = 0
     if rec.cpu_eff is not None:
         # We count failed jobs as wasted CPU!
         if rec.state != state_mappings['COMPLETED']:
@@ -906,6 +926,8 @@ def make_single_row(rec: JobRecord, dflt_mpcpu: float) -> OutputRow:
         else:
             waste_cpu = walltime * (100-rec.cpu_eff)/100 * rec.cpus
 
+    waste_total = waste_cpu + waste_mem/dflt_mpcpu
+
     jobid = rec.jobidraw
     if jobid == "":
         jobid = rec.jobid
@@ -932,6 +954,7 @@ def make_single_row(rec: JobRecord, dflt_mpcpu: float) -> OutputRow:
         walltime_max= walltime,
         waste_Mem=waste_mem,
         waste_CPU=waste_cpu,
+        waste_total=waste_total,
         _cpu_eff_values=[rec.cpu_eff] if rec.cpu_eff is not None else [],
         _mem_eff_values=[rec.mem_eff] if rec.mem_eff is not None else [],
         _time_eff_values=[rec.time_eff] if rec.time_eff is not None else [],
@@ -960,7 +983,7 @@ def make_aggregate_row(records: list[JobRecord], username: str,
     memory_efficiency=mean_or_none(mem_eff_vals)
     count=len(records)
 
-    waste_mem = None
+    waste_mem = 0
     if (memory_efficiency is not None) and (walltime is not None) \
        and first.reqmem_gb is not None:
         waste_mem = max(0,count * walltime * (100-memory_efficiency)/100 \
@@ -968,7 +991,7 @@ def make_aggregate_row(records: list[JobRecord], username: str,
 
     cpu_efficiency = mean_or_none(cpu_eff_vals)
 
-    waste_cpu=None
+    waste_cpu=0
     if cpu_efficiency is not None and walltime is not None:
         # We count failed jobs as wasted CPU!
         if first.state != state_mappings['COMPLETED']:
@@ -976,6 +999,8 @@ def make_aggregate_row(records: list[JobRecord], username: str,
         else:
             waste_cpu = count * walltime * (100-cpu_efficiency)/100 * first.cpus
 
+    waste_total = waste_cpu + waste_mem/dflt_mpcpu
+
     return OutputRow(
         username=username,
         JobID="",
@@ -998,6 +1023,7 @@ def make_aggregate_row(records: list[JobRecord], username: str,
         walltime_max=max([r.elapsed_sec for r in records if r.elapsed_sec is not None]) / 3600,
         waste_Mem=waste_mem,
         waste_CPU=waste_cpu,
+        waste_total=waste_total,
         _cpu_eff_values=cpu_eff_vals,
         _mem_eff_values=mem_eff_vals,
         _time_eff_values=time_eff_vals,