dbworkload 0.8.2__tar.gz → 0.8.4__tar.gz

{dbworkload-0.8.2 → dbworkload-0.8.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: dbworkload
-Version: 0.8.2
+Version: 0.8.4
 Summary: Workload framework
 License: GPLv3+
 Author: Fabio Ghirardello

{dbworkload-0.8.2 → dbworkload-0.8.4}/dbworkload/models/run.py

@@ -15,6 +15,7 @@ from threading import Thread
 
 import numpy as np
 import tabulate
+from psutil import cpu_percent, virtual_memory
 
 import dbworkload.utils.common
 from dbworkload.cli.dep import ConnInfo
@@ -31,11 +32,13 @@ from dbworkload.cli.dep import ConnInfo
 DEFAULT_SLEEP = 3
 MAX_RETRIES = 3
 FREQUENCY = 10
+STATS_BUFFER = 8
 
 FIFO = "dbworkload.pipe"
 
 logger = logging.getLogger("dbworkload")
 
+force_exit = False
 
 HEADERS: list = [
     "elapsed",
@@ -95,6 +98,12 @@ def signal_handler(sig, frame):
         frame (_type_):
     """
     logger.info("KeyboardInterrupt signal detected. Stopping processes...")
+    global force_exit
+    if force_exit:
+        logger.warning("Forcibly quitting. You're rude!")
+        sys.exit(1)
+
+    force_exit = True
 
     # send the poison pill to each proc.
     # if dbworkload cannot graceful shutdown due
@@ -104,13 +113,15 @@ def signal_handler(sig, frame):
     # and raise the queue.Full exception, forcing to quit.
     for q in queues.values():
         try:
-            q.put("proc_end", timeout=0.1)
+            q.put("proc_end")
         except queue.Full:
             logger.error("Timed out")
             sys.exit(1)
 
     logger.debug("Sent poison pill to all procs")
-    os.remove(FIFO)
+
+    if os.path.exists(FIFO):
+        os.remove(FIFO)
 
 
 def cycle(iterable, backwards=False):
@@ -181,18 +192,15 @@ def run(
     log_level: str,
 ):
     def gracefully_shutdown(by_keyinterrupt: bool = False):
-        """
-        wait for final stat reports to come in,
-        then print final stats and quit
-        """
+        logger.debug("Gracefully shutting down...")
 
         end_time = int(time.time())
-        _s = stats_received
+        # _s = stats_received
 
         if not by_keyinterrupt:
             for q in queues.values():
                 try:
-                    q.put("proc_end", timeout=0.1)
+                    q.put("proc_end")
                 except queue.Full:
                     logger.error("Timed out")
                     sys.exit(1)
@@ -201,20 +209,23 @@ def run(
                 if x.is_alive():
                     x.join()
 
-        while True:
-            try:
-                msg = to_main_q.get(block=True, timeout=2.0)
-                if isinstance(msg, list):
-                    _s += 1
-                    stats.add_tds(msg)
-                    if _s >= active_connections:
-                        break
-                else:
-                    logger.error("Timed out, quitting")
-                    sys.exit(1)
-
-            except queue.Empty:
-                break
+        # Commenting the below as in theory there shouldn't be any stats that
+        # comes in *after* the PROC returns. That is, all threads send stats
+        # when all threads are returned, then the supervisor returns.
+        # while True:
+        #     try:
+        #         msg = to_main_q.get(block=True, timeout=2.0)
+        #         if isinstance(msg, list):
+        #             _s += 1
+        #             stats.add_tds(msg)
+        #             if _s >= active_connections:
+        #                 break
+        #         else:
+        #             logger.error("Timed out, quitting")
+        #             sys.exit(1)
+
+        #     except queue.Empty:
+        #         break
 
         # now that we have all stat reports, calculate the stats one last time.
         report = stats.calculate_stats(active_connections, end_time)
@@ -374,10 +385,10 @@ def run(
         )
         supervisors[x].start()
 
-    # report time happens 2 seconds after the stats are received.
+    # report time happens STATS_BUFFER seconds after the stats are received.
     # we add this buffer to make sure we get all the stats reports
     # from each thread before we aggregate and display
-    report_time = start_time + FREQUENCY + 2
+    report_time = start_time + FREQUENCY + STATS_BUFFER
 
     returned_procs = 0
     active_connections = 0
@@ -478,6 +489,7 @@ def run(
                     active_connections -= 1
                 elif msg == "proc_returned":
                     returned_procs += 1
+                    logger.debug(f"Stopped processes: {returned_procs}/{procs} ")
                 elif msg == "task_done":
                     returned_threads += 1
             except queue.Empty:
@@ -501,11 +513,15 @@ def run(
                     sys.exit(1)
 
             if time.time() >= report_time:
-                # if stats_received != active_connections:
-                #     logger.warning("didn't receive all stats reports yet")
+                cpu_util = cpu_percent()
+                vmem = virtual_memory().percent
+                if stats_received != active_connections or cpu_util > 70 or vmem > 70:
+                    logger.warning(
+                        f"{stats_received=}, expected={active_connections}. CPU Util={cpu_util}%, Memory={vmem}%"
+                    )
 
-                # remove the 2 seconds added
-                endtime = int(time.time()) - 2
+                # remove the STATS_BUFFER seconds added
+                endtime = int(time.time()) - STATS_BUFFER
 
                 report = stats.calculate_stats(active_connections, endtime)
 
@@ -580,7 +596,7 @@ def run(
                 report_time += FREQUENCY
 
             # pause briefly to prevent the loop from overheating the CPU
-            time.sleep(0.1)
+            time.sleep(0.001)
 
     gracefully_shutdown()
 

{dbworkload-0.8.2 → dbworkload-0.8.4}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "dbworkload"
-version = "0.8.2"
+version = "0.8.4"
 description = "Workload framework"
 authors = ["Fabio Ghirardello"]
 license = "GPLv3+"