signal handling improvements, more debug logging

Author: dmelikyan

setup.py

@@ -2,7 +2,7 @@
 
 setup(
     name = 'stackimpact',
-    version = '1.0.0',
+    version = '1.0.1',
     description = 'StackImpact Python Agent',
     author = 'StackImpact',
     author_email = 'devops@stackimpact.com',

stackimpact/agent.py

@@ -9,8 +9,9 @@
 import os
 import signal
 import atexit
+import platform
 
-from .runtime import min_version, runtime_info
+from .runtime import min_version, runtime_info, register_signal
 from .utils import timestamp, generate_uuid
 from .config import Config
 from .config_loader import ConfigLoader
@@ -25,7 +26,7 @@
 
 class Agent:
 
-    AGENT_VERSION = "1.0.0"
+    AGENT_VERSION = "1.0.1"
     SAAS_DASHBOARD_ADDRESS = "https://agent-api.stackimpact.com"
 
     def __init__(self, **kwargs):
@@ -62,6 +63,9 @@ def start(self, **kwargs):
         if not min_version(2, 7) and not min_version(3, 4):
             raise Exception('Supported Python versions 2.6 or highter and 3.4 or higher')
 
+        if platform.python_implementation() != 'CPython':
+            raise Exception('Supported Python interpreter is CPython')
+
         if self.agent_destroyed:
             self.log('Destroyed agent cannot be started')
             return
@@ -106,10 +110,12 @@ def _signal_handler(signum, frame):
                 except Exception:
                     self.exception()
 
-        signal.signal(signal.SIGUSR2, _signal_handler)
+                return True
+
+        register_signal(signal.SIGUSR2, _signal_handler)
 
         # destroy agent on exit
-        def _exit():
+        def _exit_handler(*arg):
             if not self.agent_started or self.agent_destroyed:
                 return
 
@@ -120,7 +126,12 @@ def _exit():
                 self.exception()
 
 
-        atexit.register(_exit)
+        atexit.register(_exit_handler)
+
+        register_signal(signal.SIGQUIT, _exit_handler)
+        register_signal(signal.SIGINT, _exit_handler)
+        register_signal(signal.SIGTERM, _exit_handler)
+        register_signal(signal.SIGHUP, _exit_handler)
 
         self.agent_started = True
         self.log('Agent started')

stackimpact/frame_selector.py

@@ -64,11 +64,7 @@ def start(self):
 
 
     def destroy(self):
-        self.agent_frame_cache = None
-        self.system_frame_cache = None
-        self.http_frame_cache = None
-
-        self.http_frame_regexp = None
+        pass
 
 
     def is_agent_frame(self, filename):

stackimpact/message_queue.py

@@ -25,8 +25,6 @@ def start(self):
 
 
     def destroy(self):
-        self.queue = []
-
         if self.flush_timer:
             self.flush_timer.cancel()
             self.flush_timer = None

stackimpact/reporters/allocation_reporter.py

@@ -63,6 +63,11 @@ def reset(self):
 
 
     def record(self, max_duration):
+        if self.agent.config.is_profiling_disabled():
+            return
+
+        self.agent.log('Activating memory allocation profiler.')
+
         def start():
             tracemalloc.start(self.MAX_TRACEBACK_SIZE)
         self.agent.run_in_main_thread(start)
@@ -76,6 +81,8 @@ def start():
             if tracemalloc.get_tracemalloc_memory() > self.MAX_MEMORY_OVERHEAD:
                 break
 
+        self.agent.log('Deactivating memory allocation profiler.')
+
         if tracemalloc.is_tracing():
             snapshot = tracemalloc.take_snapshot()
             self.agent.log('Allocation profiler memory overhead {0} bytes'.format(tracemalloc.get_tracemalloc_memory()))

stackimpact/reporters/block_reporter.py

@@ -103,10 +103,14 @@ def record(self, duration):
         if self.agent.config.is_profiling_disabled():
             return
 
+        self.agent.log('Activating blocking call profiler.')
+
         signal.setitimer(signal.ITIMER_REAL, self.SAMPLING_RATE, self.SAMPLING_RATE)
         time.sleep(duration)
         signal.setitimer(signal.ITIMER_REAL, 0)
 
+        self.agent.log('Deactivating blocking call profiler.')
+
         self.profile_duration += duration
 
         self.agent.log('Time profiler CPU overhead per activity second: {0} seconds'.format(self.block_profile._overhead / self.profile_duration))
@@ -152,17 +156,15 @@ def update_http_profile(self, stack, sample_time):
         include = False
         for frame in stack:
             if self.agent.frame_selector.is_http_frame(frame.filename):
-                frame._skip = True
                 include = True
 
         if include:
             current_node = self.http_profile
             current_node.increment(sample_time, 1)
 
             for frame in reversed(stack):
-                if not frame._skip:
-                    current_node = current_node.find_or_add_child(str(frame))
-                    current_node.increment(sample_time, 1)
+                current_node = current_node.find_or_add_child(str(frame))
+                current_node.increment(sample_time, 1)
 
 
     def report(self):

stackimpact/reporters/cpu_reporter.py

@@ -86,10 +86,14 @@ def record(self, duration):
         if self.agent.config.is_profiling_disabled():
             return
 
+        self.agent.log('Activating CPU profiler.')
+
         signal.setitimer(signal.ITIMER_PROF, self.SAMPLING_RATE, self.SAMPLING_RATE)
         time.sleep(duration)
         signal.setitimer(signal.ITIMER_PROF, 0)
 
+        self.agent.log('Deactivating CPU profiler.')
+
         self.profile_duration += duration
 
         self.agent.log('CPU profiler CPU overhead per activity second: {0} seconds'.format(self.profile._overhead / self.profile_duration))

stackimpact/reporters/error_reporter.py

@@ -53,8 +53,6 @@ def destroy(self):
 
         unpatch(sys, 'exc_info')
 
-        self.reset_profile()
-
         if self.process_timer:
             self.process_timer.cancel()
             self.process_timer = None

stackimpact/runtime.py

@@ -4,6 +4,7 @@
 import resource
 import re
 import os
+import signal
 from functools import wraps
 
 
@@ -117,3 +118,17 @@ def unpatch(obj, func_name):
         return
 
     setattr(obj, func_name, getattr(wrapper, '__stackimpact_orig__'))
+
+
+def register_signal(signal_number, handler_func):
+    prev_handler = signal.SIG_IGN
+    def _handler(signum, frame):
+        skip_prev = handler_func(signum, frame)
+
+        if not skip_prev and prev_handler not in [signal.SIG_IGN, signal.SIG_DFL] and callable(prev_handler):
+            prev_handler(signum, frame)
+
+    prev_handler = signal.signal(signal_number, signal.SIG_IGN)
+    if prev_handler != signal.SIG_IGN:
+        signal.signal(signal_number, _handler)
+