fix(ssh): change cpu_load_indicator calculation as it was wrong

antarest/launcher/service.py

@@ -737,18 +737,6 @@ def get_load(self) -> LauncherLoadDTO:
             launcher_status="SUCCESS",
         )
 
-    def get_versions(self) -> Dict[str, List[str]]:
-        version_dict = {}
-        if self.config.launcher.local:
-            version_dict["local"] = list(
-                self.config.launcher.local.binaries.keys()
-            )
-
-        if self.config.launcher.slurm:
-            version_dict[
-                "slurm"
-            ] = self.config.launcher.slurm.antares_versions_on_remote_server
-
     def get_solver_versions(self, solver: str) -> List[str]:
         """
             Fetch the list of solver versions from the configuration.

antarest/launcher/ssh_client.py

@@ -50,20 +50,23 @@ def execute_command(ssh_config: SSHConfigDTO, args: List[str]) -> Any:
 
 def parse_cpu_used(sinfo_output: str) -> float:
     cpu_info_splitted = sinfo_output.split("/")
-    amount_cpu_used = int(cpu_info_splitted[0])
-    total_amount_of_cpus_usable = amount_cpu_used + int(cpu_info_splitted[1])
-    return 100 * (amount_cpu_used / total_amount_of_cpus_usable)
+    cpu_used_count = int(cpu_info_splitted[0])
+    cpu_inactive_count = int(cpu_info_splitted[1])
+    return 100 * cpu_used_count / (cpu_used_count + cpu_inactive_count)
 
 
 def parse_cpu_load(sinfo_output: str) -> float:
-    cpus_load_splitted = sinfo_output.replace(" ", "").splitlines()
-    cpt = 0
-    total_load = 0.0
-    for load in cpus_load_splitted:
-        if load != "N/A":
-            cpt += 1
-            total_load += float(load)
-    return total_load / cpt
+    lines = sinfo_output.splitlines()
+    cpus_used = 0.0
+    cpus_available = 0.0
+    for line in lines:
+        values = line.split()
+        if "N/A" in values:
+            continue
+        cpus_used += float(values[0])
+        cpus_available += float(values[1])
+    ratio = cpus_used / max(cpus_available, 1)
+    return 100 * min(1.0, ratio)
 
 
 def calculates_slurm_load(
@@ -74,7 +77,7 @@ def calculates_slurm_load(
         ssh_config,
         ["sinfo", "--partition", partition, "-O", "NodeAIOT", "--noheader"],
     )
-    allocated_cpus = 100 * parse_cpu_used(sinfo_cpus_used)
+    allocated_cpus = parse_cpu_used(sinfo_cpus_used)
     # cluster load
     sinfo_cpus_load = execute_command(
         ssh_config,
@@ -84,11 +87,11 @@ def calculates_slurm_load(
             partition,
             "-N",
             "-O",
-            "CPUsLoad",
+            "CPUsLoad,CPUs",
             "--noheader",
         ],
     )
-    cluster_load = min(100.0, parse_cpu_load(sinfo_cpus_load))
+    cluster_load = parse_cpu_load(sinfo_cpus_load)
     # queued jobs
     queued_jobs = int(
         execute_command(

tests/launcher/test_ssh_client.py

@@ -1,3 +1,4 @@
+import math
 from unittest.mock import Mock
 
 import pytest
@@ -17,10 +18,10 @@ def test_parse_cpu_used():
 
 @pytest.mark.unit_test
 def test_parse_cpu_load():
-    sinfo_output = "0.01    \n0.01    \n2.88    \n20.13    \nN / A    \n80.17"
-    assert (
-        parse_cpu_load(sinfo_output)
-        == (0.01 + 0.01 + 2.88 + 20.13 + 80.17) / 5
+    sinfo_output = "0.01     24    \n0.01   24  \nN/A     24    \n9.94   24 "
+    assert math.isclose(
+        parse_cpu_load(sinfo_output),
+        100 * (0.01 + 0.01 + 9.94) / (24 + 24 + 24),
     )