1 # Copyright (c) 2012 The Chromium OS Authors. All rights reserved. 2 # Use of this source code is governed by a BSD-style license that can be 3 # found in the LICENSE file. 4 5 import logging 6 import random 7 import signal 8 import sys 9 import threading 10 import time 11 12 from autotest_lib.client.common_lib import error 13 14 15 def install_sigalarm_handler(new_handler): 16 """ 17 Try installing a sigalarm handler. 18 19 In order to protect apache, wsgi intercepts any attempt to install a 20 sigalarm handler, so our function will feel the full force of a sigalarm 21 even if we try to install a pacifying signal handler. To avoid this we 22 need to confirm that the handler we tried to install really was installed. 23 24 @param new_handler: The new handler to install. This must be a callable 25 object, or signal.SIG_IGN/SIG_DFL which correspond to 26 the numbers 1,0 respectively. 27 @return: True if the installation of new_handler succeeded, False otherwise. 28 """ 29 if (new_handler is None or 30 (not callable(new_handler) and 31 new_handler != signal.SIG_IGN and 32 new_handler != signal.SIG_DFL)): 33 logging.warning('Trying to install an invalid sigalarm handler.') 34 return False 35 36 signal.signal(signal.SIGALRM, new_handler) 37 installed_handler = signal.getsignal(signal.SIGALRM) 38 return installed_handler == new_handler 39 40 41 def set_sigalarm_timeout(timeout_secs, default_timeout=60): 42 """ 43 Set the sigalarm timeout. 44 45 This methods treats any timeout <= 0 as a possible error and falls back to 46 using it's default timeout, since negative timeouts can have 'alarming' 47 effects. Though 0 is a valid timeout, it is often used to cancel signals; in 48 order to set a sigalarm of 0 please call signal.alarm directly as there are 49 many situations where a 0 timeout is considered invalid. 50 51 @param timeout_secs: The new timeout, in seconds. 52 @param default_timeout: The default timeout to use, if timeout <= 0. 53 @return: The old sigalarm timeout 54 """ 55 timeout_sec_n = int(timeout_secs) 56 if timeout_sec_n <= 0: 57 timeout_sec_n = int(default_timeout) 58 return signal.alarm(timeout_sec_n) 59 60 61 def sigalarm_wrapper(message): 62 """ 63 Raise a TimeoutException with the given message. Needed because the body 64 of a closure (lambda) can only be an expression, not a statement (such 65 as "raise") :P :P :P 66 67 @param message: the exception message. 68 """ 69 raise error.TimeoutException(message) 70 71 72 def custom_sigalarm_handler(func, timeout_sec): 73 """ 74 Returns a sigalarm handler which produces an exception with a custom 75 error message (function name and timeout length) instead of a generic 76 one. 77 78 @param func: the function that may time out 79 @param timeout_sec: timeout length in seconds 80 """ 81 try: 82 name = str(func.__name__) 83 except Exception as e: 84 name = '(unavailable function name: exception: %s)' % e 85 message = "sigalarm timeout (%d seconds) in %s" % (timeout_sec, name) 86 return lambda signum, frame: sigalarm_wrapper(message) 87 88 89 def timeout(func, args=(), kwargs={}, timeout_sec=60.0, default_result=None): 90 """ 91 This function run the given function using the args, kwargs and 92 return the given default value if the timeout_sec is exceeded. 93 94 @param func: function to be called. 95 @param args: arguments for function to be called. 96 @param kwargs: keyword arguments for function to be called. 97 @param timeout_sec: timeout setting for call to exit, in seconds. 98 @param default_result: default return value for the function call. 99 100 @return 1: is_timeout 2: result of the function call. If 101 is_timeout is True, the call is timed out. If the 102 value is False, the call is finished on time. 103 """ 104 old_alarm_sec = 0 105 old_handler = signal.getsignal(signal.SIGALRM) 106 handler = custom_sigalarm_handler(func, timeout_sec) 107 installed_handler = install_sigalarm_handler(handler) 108 if installed_handler: 109 old_alarm_sec = set_sigalarm_timeout(timeout_sec, default_timeout=60) 110 111 # If old_timeout_time = 0 we either didn't install a handler, or sigalrm 112 # had a signal.SIG_DFL handler with 0 timeout. In the latter case we still 113 # need to restore the handler/timeout. 114 old_timeout_time = (time.time() + old_alarm_sec) if old_alarm_sec > 0 else 0 115 116 try: 117 default_result = func(*args, **kwargs) 118 return False, default_result 119 except error.TimeoutException: 120 return True, default_result 121 finally: 122 # If we installed a sigalarm handler, cancel it since our function 123 # returned on time. If we can successfully restore the old handler, 124 # reset the old timeout, or, if the old timeout's deadline has passed, 125 # set the sigalarm to fire in one second. If the old_timeout_time is 0 126 # we don't need to set the sigalarm timeout since we have already set it 127 # as a byproduct of cancelling the current signal. 128 if installed_handler: 129 signal.alarm(0) 130 if install_sigalarm_handler(old_handler) and old_timeout_time: 131 set_sigalarm_timeout(int(old_timeout_time - time.time()), 132 default_timeout=1) 133 134 135 136 def retry(ExceptionToCheck, timeout_min=1.0, delay_sec=3, blacklist=None, 137 exception_to_raise=None, label=None): 138 """Retry calling the decorated function using a delay with jitter. 139 140 Will raise RPC ValidationError exceptions from the decorated 141 function without retrying; a malformed RPC isn't going to 142 magically become good. Will raise exceptions in blacklist as well. 143 144 If the retry is done in a child thread, timeout may not be enforced as 145 signal only works in main thread. Therefore, the retry inside a child 146 thread may run longer than timeout or even hang. 147 148 original from: 149 http://www.saltycrane.com/blog/2009/11/trying-out-retry-decorator-python/ 150 151 @param ExceptionToCheck: the exception to check. May be a tuple of 152 exceptions to check. 153 @param timeout_min: timeout in minutes until giving up. 154 @param delay_sec: pre-jittered delay between retries in seconds. Actual 155 delays will be centered around this value, ranging up to 156 50% off this midpoint. 157 @param blacklist: a list of exceptions that will be raised without retrying. 158 @param exception_to_raise: the exception to raise. Callers can specify the 159 exception they want to raise. 160 @param label: a label added to the exception message to help debug. 161 """ 162 def deco_retry(func): 163 """ 164 Decorator wrapper. 165 166 @param func: the function to be retried and timed-out. 167 """ 168 random.seed() 169 170 171 def delay(): 172 """ 173 'Jitter' the delay, up to 50% in either direction. 174 """ 175 random_delay = random.uniform(.5 * delay_sec, 1.5 * delay_sec) 176 logging.warning('Retrying in %f seconds...', random_delay) 177 time.sleep(random_delay) 178 179 180 def func_retry(*args, **kwargs): 181 """ 182 Used to cache exception to be raised later. 183 """ 184 exc_info = None 185 delayed_enabled = False 186 exception_tuple = () if blacklist is None else tuple(blacklist) 187 start_time = time.time() 188 remaining_time = timeout_min * 60 189 is_main_thread = isinstance(threading.current_thread(), 190 threading._MainThread) 191 if label: 192 details = 'label="%s"' % label 193 elif hasattr(func, '__name__'): 194 details = 'function="%s()"' % func.__name__ 195 else: 196 details = 'unknown function' 197 198 exception_message = ('retry exception (%s), timeout = %ds' % 199 (details, timeout_min * 60)) 200 201 while remaining_time > 0: 202 if delayed_enabled: 203 delay() 204 else: 205 delayed_enabled = True 206 try: 207 # Clear the cache 208 exc_info = None 209 if is_main_thread: 210 is_timeout, result = timeout(func, args, kwargs, 211 remaining_time) 212 if not is_timeout: 213 return result 214 else: 215 return func(*args, **kwargs) 216 except exception_tuple: 217 raise 218 except error.CrosDynamicSuiteException: 219 raise 220 except ExceptionToCheck as e: 221 logging.warning('%s(%s)', e.__class__, e) 222 # Cache the exception to be raised later. 223 exc_info = sys.exc_info() 224 225 remaining_time = int(timeout_min * 60 - 226 (time.time() - start_time)) 227 228 # The call must have timed out or raised ExceptionToCheck. 229 if not exc_info: 230 if exception_to_raise: 231 raise exception_to_raise(exception_message) 232 else: 233 raise error.TimeoutException(exception_message) 234 # Raise the cached exception with original backtrace. 235 if exception_to_raise: 236 raise exception_to_raise('%s: %s' % (exc_info[0], exc_info[1])) 237 raise exc_info[0], exc_info[1], exc_info[2] 238 239 240 return func_retry # true decorator 241 return deco_retry 242