chore: make validator stressor configuration on runtime and load

Signed-off-by: Huamin Chen <[email protected]>

e2e/tools/validator/scripts/stressor.sh

@@ -13,32 +13,80 @@ run() {
 	echo "      ‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾"
 }
 
+# stepwise load curve: each step is 20 seconds
+declare -a load_curve_stepwise=(
+	0:20
+	20:20
+	40:20
+	60:20
+	80:20
+	100:20
+	80:20
+	60:20
+	40:20
+	20:20
+	0:20
+)
+
+# default load curve: varying durations
+declare -a load_curve_default=(
+	0:5
+	10:20
+	25:20
+	50:20
+	75:20
+	100:30
+	75:20
+	50:20
+	25:20
+	10:20
+	0:5
+)
+
 main() {
+	local total_time=0
+	local repeats=5
+	local curve_type="default"
+
+	while getopts "t:r:c:" opt; do
+		case $opt in
+			t) total_time=$OPTARG ;;
+			c) curve_type=$OPTARG ;;
+			*) echo "Usage: $0 [-t total_time_in_seconds] [-c curve_type(default|stepwise)]" >&2; exit 1 ;;
+		esac
+	done
+
+	# Select load curve based on curve_type
+	local -a load_curve
+	case $curve_type in
+		"default") load_curve=("${load_curve_default[@]}") ;;
+		"stepwise") load_curve=("${load_curve_stepwise[@]}") ;;
+		*) echo "Invalid curve type. Use 'default' or 'stepwise'" >&2; exit 1 ;;
+	esac
 
 	local cpus
 	cpus=$(nproc)
 
-	# load and time
-	local -a load_curve=(
-		0:5
-		10:20
-		25:20
-		50:20
-		75:20
-		100:30
-		75:20
-		50:20
-		25:20
-		10:20
-		0:5
-	)
-
-	# sleep 5  so that first run and the second run look the same
+	# calculate the total duration of one cycle of the load curve
+	local total_cycle_time=0
+	for x in "${load_curve[@]}"; do
+		local time="${x##*:}"
+		total_cycle_time=$((total_cycle_time + time))
+	done
+
+	# calculate the repeats if total_time is provided
+	if [ "$total_time" -gt 0 ]; then
+		repeats=$((total_time / total_cycle_time))
+	fi
+
+	echo "Total time: $total_time seconds, Repeats: $repeats, Curve type: $curve_type"
+
+	# sleep 5 so that first run and the second run look the same
 	echo "Warmup .."
 	run stress-ng --cpu "$cpus" --cpu-method ackermann --cpu-load 0 --timeout 5
 
-	for i in $(seq 1 5); do
-		echo "Running: $i/5"
+	for i in $(seq 1 "$repeats"); do
+		echo "Running: $i/$repeats"
 		for x in "${load_curve[@]}"; do
 			local load="${x%%:*}"
 			local time="${x##*:}s"

e2e/tools/validator/src/validator/config/__init__.py

@@ -40,11 +40,17 @@ class Prometheus(NamedTuple):
     job: PrometheusJob
 
 
+class Stressor(NamedTuple):
+    total_runtime_seconds: int
+    curve_type: str
+
+
 class Validator(NamedTuple):
     log_level: str
     remote: Remote
     metal: Metal
     prometheus: Prometheus
+    stressor: Stressor
     validations_file: str
 
     def __repr__(self):
@@ -105,13 +111,23 @@ def load(config_file: str) -> Validator:
         job=job,
     )
 
+    stressor_config = config["stressor"]
+    if not stressor_config:
+        stressor = Stressor(total_runtime_seconds=1200, curve_type="default")
+    else:
+        stressor = Stressor(
+            total_runtime_seconds=stressor_config.get("total_runtime_seconds", 1200),
+            curve_type=stressor_config.get("curve_type", "default"),
+        )
+
     validations_file = config.get("validations_file", "validations.yaml")
     log_level = config.get("log_level", "warn")
 
     return Validator(
         remote=remote,
         metal=metal,
         prometheus=prometheus,
+        stressor=stressor,
         validations_file=validations_file,
         log_level=log_level,
     )

e2e/tools/validator/src/validator/stresser/__init__.py

@@ -27,6 +27,8 @@ def __init__(self, config: config.Remote):
         self.user = config.user
         self.port = config.port
         self.password = config.password
+        self.total_runtime_seconds = config.total_runtime_seconds
+        self.curve_type = config.curve_type
 
         self.ssh_client = paramiko.SSHClient()
         self.ssh_client.set_missing_host_key_policy(paramiko.AutoAddPolicy())
@@ -69,11 +71,13 @@ def run_script(self, script_path: str) -> ScriptResult:
 
         # ruff: noqa: S108 (Suppressed hard-coded path because we want to intentionally copy stress.sh inside `/tmp` dir)
         target_script = "/tmp/stress.sh"
+        cli_options = f"-t {self.total_runtime_seconds} -c {self.curve_type}"
+        command = f"{target_script} {cli_options}"
         self.copy(script_path, target_script)
 
         # ruff: noqa: DTZ005 (Suppressed non-time-zone aware object creation as it is not necessary for this use case)
         start_time = datetime.now()
-        _, stdout, stderr = self.ssh_client.exec_command(target_script)
+        _, stdout, stderr = self.ssh_client.exec_command(command)
 
         # ruff: noqa: T201 (Suppressed as printing is intentional and necessary in this context)
         print("stdout output:")

e2e/tools/validator/tests/validator/config/test_config.py

@@ -30,6 +30,10 @@ def minimal_config_file(config_file):
 
 prometheus:
   url: http://localhost:9090
+
+stressor:
+  total_runtime_seconds: 1200
+  curve_type: default
     """
     )
 
@@ -58,6 +62,34 @@ def test_minimal_config_file(minimal_config_file):
     assert prometheus.job.vm == "vm"
 
 
+@pytest.fixture
+def stressor_config_file(config_file):
+    return config_file(
+        """
+remote:
+  host: example.com
+
+metal:
+  vm:
+    pid: 1337
+
+prometheus:
+  url: http://localhost:9090
+
+stressor:
+  total_runtime_seconds: 1200
+  curve_type: default
+    """
+    )
+
+
+def test_stressor_config(stressor_config_file):
+    config = load(stressor_config_file)
+    stressor = config.stressor
+    assert stressor.total_runtime_seconds == 1200
+    assert stressor.curve_type == "default"
+
+
 @pytest.fixture
 def config_file_use_password(config_file):
     return config_file(
@@ -72,6 +104,10 @@ def config_file_use_password(config_file):
 
 prometheus:
   url: http://localhost:9090
+
+stressor:
+  total_runtime_seconds: 1200
+  curve_type: default
 """
     )
 
@@ -98,6 +134,10 @@ def config_file_job_override(config_file):
   vm:
     pid: 1337
 
+stressor:
+  total_runtime_seconds: 1200
+  curve_type: default
+
 prometheus:
   url: http://localhost:9090
 
@@ -130,6 +170,10 @@ def config_file_password_empty_pkey(config_file):
 
 prometheus:
   url: http://localhost:9090
+
+stressor:
+  total_runtime_seconds: 1200
+  curve_type: default
 """
     )
 

e2e/tools/validator/validations.yaml

@@ -2,6 +2,9 @@ config:
   mapping:
     actual: metal
     predicted: vm
+  stressor:
+    total_runtime_seconds: 1200
+    curve_type: default
 
 validations:
   - name: node-rapl - kepler-package