Expose stats

Author: neob91-close

tasktiger/stats.py

@@ -2,57 +2,66 @@
 import time
 from typing import TYPE_CHECKING, Optional
 
-from ._internal import g_fork_lock
+from ._internal import g_fork_lock, import_attribute
 
 if TYPE_CHECKING:
     from .worker import Worker
 
 
-class StatsThread(threading.Thread):
-    def __init__(self, tiger: "Worker") -> None:
-        super(StatsThread, self).__init__()
+class Stats:
+    def __init__(self, tiger, callback=None) -> None:
+        super().__init__()
+
         self.tiger = tiger
-        self._stop_event = threading.Event()
 
-        self._task_running = False
+        self._logging_thread: Optional[StatsLoggingThread] = None
+
         self._time_start = time.monotonic()
-        self._time_busy: float = 0.0
-        self._task_start_time: Optional[float] = None
-        self.daemon = True  # Exit process if main thread exits unexpectedly
+        self._time_busy = 0
+        self._task_start_time = None
 
         # Lock that protects stats computations from interleaving. For example,
         # we don't want report_task_start() to run at the same time as
-        # compute_stats(), as it might result in an inconsistent state.
+        # compute(), as it might result in an inconsistent state.
         self._computation_lock = threading.Lock()
 
-    def report_task_start(self) -> None:
+        # Callback to receive the duration of each completed task.
+        self._callback = (
+            import_attribute(callback)
+            if (callback := self.tiger.config["STATS_CALLBACK"])
+            else None
+        )
+
+    def report_task_start(self):
         now = time.monotonic()
         with self._computation_lock:
             self._task_start_time = now
-            self._task_running = True
 
     def report_task_end(self) -> None:
         now = time.monotonic()
+
+        if self._callback:
+            self._callback(now - self._task_start_time)
+
         with self._computation_lock:
-            assert self._task_start_time is not None
-            self._time_busy += now - self._task_start_time
-            self._task_running = False
+            self._record_time_busy(now)
             self._task_start_time = None
 
-    def compute_stats(self) -> None:
+    def log(self) -> None:
         now = time.monotonic()
 
         with self._computation_lock:
             time_total = now - self._time_start
             time_busy = self._time_busy
+
+            if self._task_start_time is not None:
+                # Add busy time for the currently running task
+                self._record_time_busy(now)
+
+            time_busy = self._time_busy
+
             self._time_start = now
             self._time_busy = 0
-            if self._task_running:
-                assert self._task_start_time is not None
-                time_busy += now - self._task_start_time
-                self._task_start_time = now
-            else:
-                self._task_start_time = None
 
         if time_total:
             utilization = 100.0 / time_total * time_busy
@@ -64,9 +73,33 @@ def compute_stats(self) -> None:
                     utilization=utilization,
                 )
 
+    def start_logging_thread(self):
+        if not self._logging_thread:
+            self._logging_thread = StatsLoggingThread(self)
+            self._logging_thread.start()
+
+    def stop_logging_thread(self):
+        if self._logging_thread:
+            self._logging_thread.stop()
+            self._logging_thread = None
+
+    def _record_time_busy(self, now: float) -> None:
+        self._time_busy += now - max(self._task_start_time, self._time_start)
+
+
+class StatsLoggingThread(threading.Thread):
+    def __init__(self, stats) -> None:
+        super().__init__()
+
+        self.tiger = stats.tiger
+        self._stats: Stats = stats
+        self._stop_event = threading.Event()
+
+        self.daemon = True  # Exit process if main thread exits unexpectedly
+
     def run(self) -> None:
         while not self._stop_event.wait(self.tiger.config["STATS_INTERVAL"]):
-            self.compute_stats()
+            self._stats.log()
 
     def stop(self) -> None:
         self._stop_event.set()

tasktiger/tasktiger.py

@@ -198,6 +198,12 @@ def init(
             "BATCH_QUEUES": {},
             # How often to print stats.
             "STATS_INTERVAL": 60,
+            # The function to call with the duration of each completed task.
+            # This can be useful to measure the worker's utilisation %.
+            # For example, the worker's utilisation over the last 30 minutes
+            # can be obtained by dividing the sum of task durations reported
+            # over the last 30 minutes by 30 minutes.
+            "STATS_CALLBACK": None,
             # Single worker queues can reduce redis activity in some use cases
             # by locking at the queue level instead of just at the task or task
             # group level. These queues will only allow a single worker to

tasktiger/worker.py

@@ -55,7 +55,7 @@
 )
 from .redis_semaphore import Semaphore
 from .runner import get_runner_class
-from .stats import StatsThread
+from .stats import Stats
 from .task import Task
 from .timeouts import JobTimeoutException
 from .utils import redis_glob_escape
@@ -107,7 +107,7 @@ def __init__(
         self._key = tiger._key
         self._did_work = True
         self._last_task_check = 0.0
-        self.stats_thread: Optional[StatsThread] = None
+        self.stats = Stats(self)
         self.id = str(uuid.uuid4())
 
         if queues:
@@ -1008,13 +1008,11 @@ def _execute_task_group(
         if not ready_tasks:
             return True, []
 
-        if self.stats_thread:
-            self.stats_thread.report_task_start()
+        self.stats.report_task_start()
         success = self._execute(
             queue, ready_tasks, log, locks, queue_lock, all_task_ids
         )
-        if self.stats_thread:
-            self.stats_thread.report_task_end()
+        self.stats.report_task_end()
 
         for lock in locks:
             try:
@@ -1280,9 +1278,7 @@ def run(self, once: bool = False, force_once: bool = False) -> None:
             self.log.warn("using old Redis version")
 
         if self.config["STATS_INTERVAL"]:
-            stats_thread = StatsThread(self)
-            self.stats_thread = stats_thread
-            stats_thread.start()
+            self.stats.start_logging_thread()
 
         # Queue any periodic tasks that are not queued yet.
         self._queue_periodic_tasks()
@@ -1328,9 +1324,7 @@ def run(self, once: bool = False, force_once: bool = False) -> None:
             raise
 
         finally:
-            if self.stats_thread:
-                self.stats_thread.stop()
-                self.stats_thread = None
+            self.stats.stop_logging_thread()
 
             # Free up Redis connection
             if self._pubsub: