Add CPU usage to host inventory

agent/ec_agent.py

@@ -30,7 +30,7 @@ import urllib.parse
 import urllib.request
 import uuid
 
-VERSION = "1.4.1"
+VERSION = "1.4.2"
 LABEL = "io.electric-cherry"
 MAX_FILE = 32 * 1024 * 1024
 LIBRARY_DOWNLOAD_WORKERS = 4
@@ -710,6 +710,8 @@ class Agent:
         self.docker = docker or Docker()
         self.stopping = threading.Event()
         self.current_operation = None
+        self.cpu_sample_lock = threading.Lock()
+        self.cpu_sample = None
         self.instance_locks = {}
         self.read_gate = ReadWriteGate()
         self.read_claim_lock = threading.Lock()
@@ -899,7 +901,36 @@ class Agent:
         except (OSError, ValueError):
             raise AgentError("Controller request failed") from None
 
+    def cpu_usage_percent(self):
+        """Host-wide busy time since the previous inventory sample, without waiting."""
+        with self.cpu_sample_lock:
+            previous = self.cpu_sample
+            self.cpu_sample = None
+            try:
+                with Path("/proc/stat").open() as source:
+                    fields = source.readline().split()
+                if not fields or fields[0] != "cpu" or len(fields) < 9:
+                    return None
+                # Guest time is already included in user/nice. Idle and I/O
+                # wait are not busy; steal counts as unavailable CPU capacity.
+                counters = tuple(int(value) for value in fields[1:9])
+                if any(value < 0 for value in counters):
+                    return None
+            except (OSError, ValueError):
+                return None
+            self.cpu_sample = counters
+            if previous is None:
+                return None
+            deltas = [current - old for current, old in zip(counters, previous)]
+            total = sum(deltas)
+            # Counters can reset and Linux can revise I/O wait downwards.
+            # Rebase rather than reporting a misleading utilization value.
+            if total <= 0 or any(delta < 0 for delta in deltas):
+                return None
+            return round(100 * (total - deltas[3] - deltas[4]) / total, 1)
+
     def inventory(self):
+        cpu_usage = self.cpu_usage_percent()
         usage = shutil.disk_usage(self.root)
         memory = {}
         try:
@@ -1004,7 +1035,7 @@ class Agent:
                     os_name = line.split("=", 1)[1].strip('"')
         update = self.update_inventory()
         capabilities = ["instance.cancel-create", "operations.read-lane"] + (["host.decommission"] if self.can_uninstall() else []) + (["agent.update"] if update["enabled"] else [])
-        return {"hostname": socket.gethostname(), "agentVersion": VERSION, "protocolVersion": 1, "capabilities": capabilities, "agentUpdate": update, "os": os_name, "osVersion": platform.release(), "architecture": platform.machine(), "cpu": platform.processor() or platform.machine(), "cpuCount": os.cpu_count() or 0, "memoryTotal": memory.get("MemTotal", 0), "memoryAvailable": memory.get("MemAvailable", 0), "storageTotal": usage.total, "storageAvailable": usage.free, "dockerVersion": docker_version, "dockerRunning": docker_running, "directoriesWritable": os.access(self.root, os.W_OK), "runtimeImages": sorted(set(images)), "libraries": libraries, "libraryCaches": caches, "currentOperation": self.current_operation, "portRange": self.config.get("portRange", {"min": 28960, "max": 29060})}
+        return {"hostname": socket.gethostname(), "agentVersion": VERSION, "protocolVersion": 1, "capabilities": capabilities, "agentUpdate": update, "os": os_name, "osVersion": platform.release(), "architecture": platform.machine(), "cpu": platform.processor() or platform.machine(), "cpuCount": os.cpu_count() or 0, **({"cpuUsagePercent": cpu_usage} if cpu_usage is not None else {}), "memoryTotal": memory.get("MemTotal", 0), "memoryAvailable": memory.get("MemAvailable", 0), "storageTotal": usage.total, "storageAvailable": usage.free, "dockerVersion": docker_version, "dockerRunning": docker_running, "directoriesWritable": os.access(self.root, os.W_OK), "runtimeImages": sorted(set(images)), "libraries": libraries, "libraryCaches": caches, "currentOperation": self.current_operation, "portRange": self.config.get("portRange", {"min": 28960, "max": 29060})}
 
     def safe_to_disrupt(self, instance_id, generation):
         health = self.inspect(instance_id, generation)