Add dev/load for distributed load testing

Author: deliahu

dev/load/cortex.yaml

@@ -0,0 +1,16 @@
+- name: load
+  kind: RealtimeAPI
+  predictor:
+    type: python
+    path: predictor.py
+    log_level: debug
+    config:
+      endpoint: http://a76d3b3343f1a43e8a5dd1f556689215-14bf49c6f95638fb.elb.us-west-2.amazonaws.com/iris-classifier
+      data: { "sepal_length": 2.2, "sepal_width": 3.6, "petal_length": 1.4, "petal_width": 3.3 }
+      sleep: 0.8
+      num_requests: 1000
+  autoscaling:
+    min_replicas: 50
+    max_replicas: 50
+  compute:
+    cpu: 1.3

dev/load/predictor.py

@@ -0,0 +1,29 @@
+import requests
+import time
+from cortex_internal.lib.log import logger as cortex_logger
+
+
+class PythonPredictor:
+    def __init__(self, config):
+        num_success = 0
+        num_fail = 0
+        for i in range(config["num_requests"]):
+            if i > 0:
+                time.sleep(config["sleep"])
+            response = requests.post(config["endpoint"], json=config["data"], timeout=3600)
+            if response.status_code == 200:
+                num_success += 1
+            else:
+                num_fail += 1
+                cortex_logger.error(
+                    "ERROR",
+                    extra={"error": True, "code": response.status_code, "body": response.text},
+                )
+
+        cortex_logger.warn(
+            "FINISHED",
+            extra={"finished": True, "num_success": num_success, "num_fail": num_fail},
+        )
+
+    def predict(self, payload):
+        return "ok"