Fixed callbacks' first iteration bug with scheduler refactor to use APScheduler

Author: DtDamianGrzes

dynatrace_extension/__about__.py

@@ -3,4 +3,4 @@
 # SPDX-License-Identifier: MIT
 
 
-__version__ = "1.7.2"
+__version__ = "1.7.3"

dynatrace_extension/sdk/callback.py

@@ -33,7 +33,7 @@ def __init__(
         self.interval: timedelta = interval
         self.logger = logger
         self.running: bool = False
-        self.status = Status(StatusValue.OK)
+        self.status = IgnoreStatus()
         self.executions_total = 0  # global counter
         self.executions_per_interval = 0  # counter per interval = 1 min by default
         self.duration = 0  # global counter
@@ -45,7 +45,6 @@ def __init__(
         self.ok_count = 0  # counter per interval = 1 min by default
         self.timeouts_count = 0  # counter per interval = 1 min by default
         self.exception_count = 0  # counter per interval = 1 min by default
-        self.iterations = 0  # how many times we ran the callback iterator for this callback
 
     def get_current_time_with_cluster_diff(self):
         return datetime.now() + timedelta(milliseconds=self.cluster_time_diff)
@@ -142,13 +141,3 @@ def clear_sfm_metrics(self):
         self.duration_interval_total = 0
         self.exception_count = 0
         self.executions_per_interval = 0
-
-    def get_next_execution_timestamp(self) -> float:
-        """
-        Get the timestamp for the next execution of the callback
-        This is done using execution total, the interval and the start timestamp
-        :return: datetime
-        """
-        return (
-            self.start_timestamp + timedelta(seconds=self.interval.total_seconds() * (self.iterations or 1))
-        ).timestamp()

dynatrace_extension/sdk/extension.py

@@ -8,9 +8,11 @@
 import sys
 import threading
 import time
+from apscheduler.schedulers.background import BackgroundScheduler
+from apscheduler.executors.pool import ThreadPoolExecutor
+from apscheduler.triggers.interval import IntervalTrigger
 from argparse import ArgumentParser
 from collections.abc import Callable
-from concurrent.futures import ThreadPoolExecutor
 from datetime import datetime, timedelta, timezone
 from enum import Enum
 from itertools import chain
@@ -172,6 +174,12 @@ def _add_sfm_metric(metric: Metric, sfm_metrics: list[Metric] | None = None):
     sfm_metrics.append(metric)
 
 
+class ExecutorType(str, Enum):
+    CALLBACKS = 'callbacks'
+    INTERNAL = 'internal'
+    HEARTBEAT = 'heartbeat'
+
+
 class Extension:
     """Base class for Python extensions.
 
@@ -240,21 +248,12 @@ def __init__(self, name: str = "") -> None:
         self._running_callbacks: dict[int, WrappedCallback] = {}
         self._running_callbacks_lock: Lock = Lock()
 
-        self._scheduler = sched.scheduler(time.time, time.sleep)
-
-        # Timestamps for scheduling of internal callbacks
-        self._next_internal_callbacks_timestamps: dict[str, datetime] = {
-            "timediff": datetime.now() + TIME_DIFF_INTERVAL,
-            "heartbeat": datetime.now() + HEARTBEAT_INTERVAL,
-            "metrics": datetime.now() + METRIC_SENDING_INTERVAL,
-            "events": datetime.now() + METRIC_SENDING_INTERVAL,
-            "sfm_metrics": datetime.now() + SFM_METRIC_SENDING_INTERVAL,
-        }
-
-        # Executors for the callbacks and internal methods
-        self._callbacks_executor = ThreadPoolExecutor(max_workers=CALLBACKS_THREAD_POOL_SIZE)
-        self._internal_executor = ThreadPoolExecutor(max_workers=INTERNAL_THREAD_POOL_SIZE)
-        self._heartbeat_executor = ThreadPoolExecutor(max_workers=HEARTBEAT_THREAD_POOL_SIZE)
+        # Scheduler and executors for the callbacks and internal methods
+        self._scheduler = BackgroundScheduler(executors={
+            ExecutorType.CALLBACKS: ThreadPoolExecutor(max_workers=CALLBACKS_THREAD_POOL_SIZE),
+            ExecutorType.INTERNAL: ThreadPoolExecutor(max_workers=INTERNAL_THREAD_POOL_SIZE),
+            ExecutorType.HEARTBEAT: ThreadPoolExecutor(max_workers=HEARTBEAT_THREAD_POOL_SIZE)
+        })
 
         # Extension metrics
         self._metrics_lock = RLock()
@@ -376,7 +375,12 @@ def _schedule_callback(self, callback: WrappedCallback):
         callback.cluster_time_diff = self._cluster_time_diff
         callback.running_in_sim = self._running_in_sim
         self._scheduled_callbacks.append(callback)
-        self._scheduler.enter(callback.initial_wait_time(), 1, self._callback_iteration, (callback,))
+
+        self._scheduler.add_job(self._run_callback, args=[callback],
+            executor=ExecutorType.CALLBACKS,
+            trigger=IntervalTrigger(seconds=callback.interval.total_seconds()),
+            next_run_time=datetime.now() + timedelta(seconds=callback.initial_wait_time())
+        )
 
     def schedule(
         self,
@@ -809,7 +813,10 @@ def _parse_args(self):
 
         if not self._is_fastcheck:
             try:
-                self._heartbeat_iteration()
+                # TODO: is it surely okay to schedule hearbeat this way? Originally it was scheduled in the very same scheduler, which would starve heartbeat if any callback took too long
+                # On the other hand, those callbacks inserted specific potentially risky jobs to different executors, so it should be okay?
+                # Why did heartbeat have a different priority (higher or lower?)
+                self._scheduler.add_job(self._heartbeat, executor=ExecutorType.HEARTBEAT, trigger=IntervalTrigger(seconds=HEARTBEAT_INTERVAL.total_seconds()))
                 self.initialize()
                 if not self.is_helper:
                     self.schedule(self.query, timedelta(minutes=1))
@@ -863,48 +870,39 @@ def _run_callback(self, callback: WrappedCallback):
             with self._running_callbacks_lock:
                 self._running_callbacks.pop(current_thread_id, None)
 
-    def _callback_iteration(self, callback: WrappedCallback):
-        self._callbacks_executor.submit(self._run_callback, callback)
-        callback.iterations += 1
-        next_timestamp = callback.get_next_execution_timestamp()
-        self._scheduler.enterabs(next_timestamp, 1, self._callback_iteration, (callback,))
 
     def _start_extension_loop(self):
         api_logger.debug(f"Starting main loop for monitoring configuration: '{self.monitoring_config_name}'")
 
         # These were scheduled before the extension started, schedule them now
         for callback in self._scheduled_callbacks_before_run:
             self._schedule_callback(callback)
-        self._metrics_iteration()
-        self._events_iteration()
-        self._sfm_metrics_iteration()
-        self._timediff_iteration()
-        self._scheduler.run()
-
-    def _timediff_iteration(self):
-        self._internal_executor.submit(self._update_cluster_time_diff)
-        next_timestamp = self._get_and_set_next_internal_callback_timestamp("timediff", TIME_DIFF_INTERVAL)
-        self._scheduler.enterabs(next_timestamp, 1, self._timediff_iteration)
-
-    def _heartbeat_iteration(self):
-        self._heartbeat_executor.submit(self._heartbeat)
-        next_timestamp = self._get_and_set_next_internal_callback_timestamp("heartbeat", HEARTBEAT_INTERVAL)
-        self._scheduler.enterabs(next_timestamp, 2, self._heartbeat_iteration)
-
-    def _metrics_iteration(self):
-        self._internal_executor.submit(self._send_metrics)
-        next_timestamp = self._get_and_set_next_internal_callback_timestamp("metrics", METRIC_SENDING_INTERVAL)
-        self._scheduler.enterabs(next_timestamp, 1, self._metrics_iteration)
-
-    def _events_iteration(self):
-        self._internal_executor.submit(self._send_buffered_events)
-        next_timestamp = self._get_and_set_next_internal_callback_timestamp("events", METRIC_SENDING_INTERVAL)
-        self._scheduler.enterabs(next_timestamp, 1, self._events_iteration)
-
-    def _sfm_metrics_iteration(self):
-        self._internal_executor.submit(self._send_sfm_metrics)
-        next_timestamp = self._get_and_set_next_internal_callback_timestamp("sfm_metrics", SFM_METRIC_SENDING_INTERVAL)
-        self._scheduler.enterabs(next_timestamp, 1, self._sfm_metrics_iteration)
+
+                
+        self._scheduler.add_job(self._send_metrics, executor=ExecutorType.INTERNAL,
+            trigger=IntervalTrigger(seconds=METRIC_SENDING_INTERVAL.total_seconds()),
+            next_run_time=datetime.now())
+
+        self._scheduler.add_job(self._send_buffered_events, executor=ExecutorType.INTERNAL,
+            trigger=IntervalTrigger(seconds=METRIC_SENDING_INTERVAL.total_seconds()),
+            next_run_time=datetime.now())
+
+        self._scheduler.add_job(self._send_sfm_metrics, executor=ExecutorType.INTERNAL,
+            trigger=IntervalTrigger(seconds=SFM_METRIC_SENDING_INTERVAL.total_seconds()),
+            next_run_time=datetime.now())
+
+        self._scheduler.add_job(self._update_cluster_time_diff, executor=ExecutorType.INTERNAL,
+            trigger=IntervalTrigger(seconds=TIME_DIFF_INTERVAL.total_seconds()),
+            next_run_time=datetime.now())
+
+        self._scheduler.start()
+        
+        try:
+            while self._scheduler.running:
+                time.sleep(1)
+        except Exception:
+            self._scheduler.shutdown()
+
 
     def _send_metrics(self):
         with self._metrics_lock:
@@ -1105,8 +1103,8 @@ def _heartbeat(self):
             api_logger.error(f"Heartbeat failed because {e}, response {response}", exc_info=True)
 
     def __del__(self):
-        self._callbacks_executor.shutdown()
-        self._internal_executor.shutdown()
+        if self._scheduler.running:
+            self._scheduler.shutdown()
 
     def _add_metric(self, metric: Metric):
         metric.validate()
@@ -1150,7 +1148,7 @@ def _send_events_internal(self, events: dict | list[dict]):
 
     def _send_events(self, events: dict | list[dict], send_immediately: bool = False):
         if send_immediately:
-            self._internal_executor.submit(self._send_events_internal, events)
+            self._scheduler.add_job(self._send_events_internal, args=[events], executor=ExecutorType.INTERNAL)
             return
         with self._logs_lock:
             if isinstance(events, dict):
@@ -1247,4 +1245,4 @@ def _send_sfm_logs(self, logs: dict | list[dict]):
             log["dt.extension.config.label"] = self.monitoring_config_name
             log.pop("monitoring.configuration", None)
 
-        self._internal_executor.submit(self._send_sfm_logs_internal, logs)
+        self._scheduler.add_job(self._send_sfm_logs_internal, args=[logs], executor=ExecutorType.INTERNAL)

dynatrace_extension/sdk/status.py

@@ -159,6 +159,7 @@ def __init__(self, send_sfm_logs_function: Callable) -> None:
         self._ep_records: dict[str, EndpointStatusRecord] = {}
         self._send_sfm_logs_function = send_sfm_logs_function
         self._logs_to_send: list[str] = []
+        self._datetime_now = datetime.now # Mockable datetime function
 
     def contains_any_status(self) -> bool:
         return len(self._ep_records) > 0
@@ -190,7 +191,7 @@ def send_ep_logs(self):
                             ep_record.ep_status.message,
                         )
                     )
-                    ep_record.last_sent = datetime.now()
+                    ep_record.last_sent = self._datetime_now()
                     ep_record.state = StatusState.ONGOING
 
         if logs_to_send:
@@ -202,7 +203,7 @@ def _should_be_reported(self, ep_record: EndpointStatusRecord):
         elif ep_record.state in (StatusState.INITIAL, StatusState.NEW):
             return True
         elif ep_record.state == StatusState.ONGOING and (
-            ep_record.last_sent is None or datetime.now() - ep_record.last_sent >= self.RESENDING_INTERVAL
+            ep_record.last_sent is None or self._datetime_now() - ep_record.last_sent >= self.RESENDING_INTERVAL
         ):
             return True
         else:

pyproject.toml

@@ -47,8 +47,7 @@ dependencies = [
     "pytest",
     "typer[all]",
     "pyyaml",
-    "dt-cli>=1.6.13",
-    "freezegun"
+    "dt-cli>=1.6.13"
 ]
 
 [tool.hatch.envs.default.scripts]
@@ -74,8 +73,7 @@ dependencies = [
     "ruff>=0.9.10",
     "typer[all]",
     "pyyaml",
-    "pytest",    
-    "freezegun"
+    "pytest"
 ]
 
 [tool.hatch.envs.lint.scripts]

tests/sdk/test_endpoints_sfm.py

@@ -1,52 +1,42 @@
-import threading
 import time
 import unittest
 from datetime import datetime, timedelta
 from unittest.mock import MagicMock
 
-from freezegun import freeze_time
-
 from dynatrace_extension import EndpointStatus, EndpointStatuses, Extension, Severity, StatusValue
 
 
-class KillSchedulerError(Exception):
-    pass
-
-
 class TestSfmPerEndpont(unittest.TestCase):
     def setUp(self, extension_name=""):
         self.ext = Extension(name=extension_name)
         self.ext.logger = MagicMock()
         self.ext._running_in_sim = True
         self.ext._client = MagicMock()
+        self.ext._client.send_sfm_logs = MagicMock(side_effect=lambda *args, **kwargs: [])
         self.ext._is_fastcheck = False
         self.i = 0
         self.test_cases = None
         self.time_machine_idx = None
 
         self.ext.schedule(self.callback, timedelta(seconds=1))
-        self.scheduler_thread = threading.Thread(target=self.scheduler_thread_fun)
-
-    def scheduler_thread_fun(self):
-        with self.assertRaises(KillSchedulerError):
-            self.ext._scheduler.run()
 
     def tearDown(self) -> None:
-        self.ext._scheduler.enter(delay=0, priority=1, action=lambda: (_ for _ in ()).throw(KillSchedulerError()))
-        self.scheduler_thread.join()
+        if Extension._instance and Extension._instance._scheduler.running:
+            Extension._instance._scheduler.shutdown(wait=True)
+
         Extension._instance = None
 
     def run_test(self):
-        self.scheduler_thread.start()
+        self.ext._scheduler.start()
         time.sleep(0.1)
 
         for case in self.test_cases[: self.time_machine_idx]:
             self.single_test_iteration(case)
 
         if self.time_machine_idx:
-            with freeze_time(datetime.now() + timedelta(hours=2), tick=True):
-                for case in self.test_cases[self.time_machine_idx :]:
-                    self.single_test_iteration(case)
+            self.ext._ep_statuses._datetime_now = lambda: datetime.now() + timedelta(hours=2)
+            for case in self.test_cases[self.time_machine_idx :]:
+                self.single_test_iteration(case)
 
     def single_test_iteration(self, case):
         self.ext._client.send_sfm_logs.reset_mock()
@@ -247,7 +237,7 @@ def test_endpoint_independent_ongoing(self):
                 ),
             },
             {
-                "status": EndpointStatus("1.2.3.4:1", StatusValue.WARNING, "Warning 1"),
+                "status": None,
                 "expected": self.expected_sfm_dict(
                     device_address="1.2.3.4:1",
                     level="WARN",
@@ -257,7 +247,7 @@ def test_endpoint_independent_ongoing(self):
                 ),
             },
             {
-                "status": EndpointStatus("1.2.3.4:2", StatusValue.WARNING, "Warning 2"),
+                "status": None,
                 "expected": self.expected_sfm_dict(
                     device_address="1.2.3.4:2",
                     level="WARN",
@@ -287,19 +277,21 @@ def test_endpoint_independent_ongoing(self):
             },
         ]
 
-        self.scheduler_thread.start()
+        self.ext._scheduler.start()
         time.sleep(0.1)
 
         self.single_test_iteration(self.test_cases[0])
 
-        with freeze_time(datetime.now() + timedelta(hours=1), tick=True):
-            self.single_test_iteration(self.test_cases[1])
+        self.ext._ep_statuses._datetime_now = lambda: datetime.now() + timedelta(hours=1)
+        self.single_test_iteration(self.test_cases[1])
+
+        self.ext._ep_statuses._datetime_now = lambda: datetime.now() + timedelta(hours=2)
+        self.single_test_iteration(self.test_cases[2])
 
-            with freeze_time(datetime.now() + timedelta(hours=1), tick=True):
-                self.single_test_iteration(self.test_cases[2])
+        self.ext._ep_statuses._datetime_now = lambda: datetime.now() + timedelta(hours=3)
+        self.single_test_iteration(self.test_cases[3])
 
-                with freeze_time(datetime.now() + timedelta(hours=1), tick=True):
-                    self.single_test_iteration(self.test_cases[3])
+        self.ext._ep_statuses._datetime_now = lambda: datetime.now() + timedelta(hours=5)
+        self.single_test_iteration(self.test_cases[4])
 
-                    with freeze_time(datetime.now() + timedelta(hours=2), tick=True):
-                        self.single_test_iteration(self.test_cases[4])
+        self.ext._scheduler.shutdown(wait=True)