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 env 13 from autotest_lib.client.common_lib import error 14 15 16 def install_sigalarm_handler(new_handler): 17 """ 18 Try installing a sigalarm handler. 19 20 In order to protect apache, wsgi intercepts any attempt to install a 21 sigalarm handler, so our function will feel the full force of a sigalarm 22 even if we try to install a pacifying signal handler. To avoid this we 23 need to confirm that the handler we tried to install really was installed. 24 25 @param new_handler: The new handler to install. This must be a callable 26 object, or signal.SIG_IGN/SIG_DFL which correspond to 27 the numbers 1,0 respectively. 28 @return: True if the installation of new_handler succeeded, False otherwise. 29 """ 30 # Installing signal handlers does not and is never expected to work if we're 31 # running in a mod_wsgi process. 32 if env.IN_MOD_WSGI: 33 return False 34 35 if (new_handler is None or 36 (not callable(new_handler) and 37 new_handler != signal.SIG_IGN and 38 new_handler != signal.SIG_DFL)): 39 logging.warning('Trying to install an invalid sigalarm handler.') 40 return False 41 42 signal.signal(signal.SIGALRM, new_handler) 43 installed_handler = signal.getsignal(signal.SIGALRM) 44 return installed_handler == new_handler 45 46 47 def set_sigalarm_timeout(timeout_secs, default_timeout=60): 48 """ 49 Set the sigalarm timeout. 50 51 This methods treats any timeout <= 0 as a possible error and falls back to 52 using it's default timeout, since negative timeouts can have 'alarming' 53 effects. Though 0 is a valid timeout, it is often used to cancel signals; in 54 order to set a sigalarm of 0 please call signal.alarm directly as there are 55 many situations where a 0 timeout is considered invalid. 56 57 @param timeout_secs: The new timeout, in seconds. 58 @param default_timeout: The default timeout to use, if timeout <= 0. 59 @return: The old sigalarm timeout 60 """ 61 timeout_sec_n = int(timeout_secs) 62 if timeout_sec_n <= 0: 63 timeout_sec_n = int(default_timeout) 64 return signal.alarm(timeout_sec_n) 65 66 67 def sigalarm_wrapper(message): 68 """ 69 Raise a TimeoutException with the given message. Needed because the body 70 of a closure (lambda) can only be an expression, not a statement (such 71 as "raise") :P :P :P 72 73 @param message: the exception message. 74 """ 75 raise error.TimeoutException(message) 76 77 78 def custom_sigalarm_handler(func, timeout_sec): 79 """ 80 Returns a sigalarm handler which produces an exception with a custom 81 error message (function name and timeout length) instead of a generic 82 one. 83 84 @param func: the function that may time out 85 @param timeout_sec: timeout length in seconds 86 """ 87 try: 88 name = str(func.__name__) 89 except Exception as e: 90 name = '(unavailable function name: exception: %s)' % e 91 message = "sigalarm timeout (%d seconds) in %s" % (timeout_sec, name) 92 return lambda signum, frame: sigalarm_wrapper(message) 93 94 95 def timeout(func, args=(), kwargs={}, timeout_sec=60.0, default_result=None): 96 """ 97 This function run the given function using the args, kwargs and 98 return the given default value if the timeout_sec is exceeded. 99 100 @param func: function to be called. 101 @param args: arguments for function to be called. 102 @param kwargs: keyword arguments for function to be called. 103 @param timeout_sec: timeout setting for call to exit, in seconds. 104 @param default_result: default return value for the function call. 105 106 @return 1: is_timeout 2: result of the function call. If 107 is_timeout is True, the call is timed out. If the 108 value is False, the call is finished on time. 109 """ 110 old_alarm_sec = 0 111 old_handler = signal.getsignal(signal.SIGALRM) 112 handler = custom_sigalarm_handler(func, timeout_sec) 113 installed_handler = install_sigalarm_handler(handler) 114 if installed_handler: 115 old_alarm_sec = set_sigalarm_timeout(timeout_sec, default_timeout=60) 116 117 # If old_timeout_time = 0 we either didn't install a handler, or sigalrm 118 # had a signal.SIG_DFL handler with 0 timeout. In the latter case we still 119 # need to restore the handler/timeout. 120 old_timeout_time = (time.time() + old_alarm_sec) if old_alarm_sec > 0 else 0 121 122 try: 123 default_result = func(*args, **kwargs) 124 return False, default_result 125 except error.TimeoutException: 126 return True, default_result 127 finally: 128 # If we installed a sigalarm handler, cancel it since our function 129 # returned on time. If we can successfully restore the old handler, 130 # reset the old timeout, or, if the old timeout's deadline has passed, 131 # set the sigalarm to fire in one second. If the old_timeout_time is 0 132 # we don't need to set the sigalarm timeout since we have already set it 133 # as a byproduct of cancelling the current signal. 134 if installed_handler: 135 signal.alarm(0) 136 if install_sigalarm_handler(old_handler) and old_timeout_time: 137 set_sigalarm_timeout(int(old_timeout_time - time.time()), 138 default_timeout=1) 139 140 141 142 def retry(ExceptionToCheck, timeout_min=1.0, delay_sec=3, blacklist=None, 143 exception_to_raise=None, label=None, callback=None): 144 """Retry calling the decorated function using a delay with jitter. 145 146 Will raise RPC ValidationError exceptions from the decorated 147 function without retrying; a malformed RPC isn't going to 148 magically become good. Will raise exceptions in blacklist as well. 149 150 If the retry is done in a child thread, timeout may not be enforced as 151 signal only works in main thread. Therefore, the retry inside a child 152 thread may run longer than timeout or even hang. 153 154 original from: 155 http://www.saltycrane.com/blog/2009/11/trying-out-retry-decorator-python/ 156 157 @param ExceptionToCheck: the exception to check. May be a tuple of 158 exceptions to check. 159 @param timeout_min: timeout in minutes until giving up. 160 @param delay_sec: pre-jittered delay between retries in seconds. Actual 161 delays will be centered around this value, ranging up to 162 50% off this midpoint. 163 @param blacklist: a list of exceptions that will be raised without retrying. 164 @param exception_to_raise: the exception to raise. Callers can specify the 165 exception they want to raise. 166 @param label: a label added to the exception message to help debug. 167 @param callback: a function to call before each retry. 168 """ 169 def deco_retry(func): 170 """ 171 Decorator wrapper. 172 173 @param func: the function to be retried and timed-out. 174 """ 175 random.seed() 176 177 178 def delay(): 179 """ 180 'Jitter' the delay, up to 50% in either direction. 181 """ 182 random_delay = random.uniform(.5 * delay_sec, 1.5 * delay_sec) 183 logging.warning('Retrying in %f seconds...', random_delay) 184 time.sleep(random_delay) 185 186 187 def func_retry(*args, **kwargs): 188 """ 189 Used to cache exception to be raised later. 190 """ 191 exc_info = None 192 delayed_enabled = False 193 exception_tuple = () if blacklist is None else tuple(blacklist) 194 start_time = time.time() 195 remaining_time = timeout_min * 60 196 is_main_thread = isinstance(threading.current_thread(), 197 threading._MainThread) 198 if label: 199 details = 'label="%s"' % label 200 elif hasattr(func, '__name__'): 201 details = 'function="%s()"' % func.__name__ 202 else: 203 details = 'unknown function' 204 205 exception_message = ('retry exception (%s), timeout = %ds' % 206 (details, timeout_min * 60)) 207 208 while remaining_time > 0: 209 if delayed_enabled: 210 delay() 211 else: 212 delayed_enabled = True 213 try: 214 # Clear the cache 215 exc_info = None 216 if is_main_thread: 217 is_timeout, result = timeout(func, args, kwargs, 218 remaining_time) 219 if not is_timeout: 220 return result 221 else: 222 return func(*args, **kwargs) 223 except exception_tuple: 224 raise 225 except error.CrosDynamicSuiteException: 226 raise 227 except ExceptionToCheck as e: 228 logging.warning('%s(%s)', e.__class__, e) 229 # Cache the exception to be raised later. 230 exc_info = sys.exc_info() 231 232 remaining_time = int(timeout_min * 60 - 233 (time.time() - start_time)) 234 235 if remaining_time > 0 and callback: 236 callback() 237 remaining_time = int(timeout_min * 60 - 238 (time.time() - start_time)) 239 240 241 # The call must have timed out or raised ExceptionToCheck. 242 if not exc_info: 243 if exception_to_raise: 244 raise exception_to_raise(exception_message) 245 else: 246 raise error.TimeoutException(exception_message) 247 # Raise the cached exception with original backtrace. 248 if exception_to_raise: 249 raise exception_to_raise('%s: %s' % (exc_info[0], exc_info[1])) 250 raise exc_info[0], exc_info[1], exc_info[2] 251 252 253 return func_retry # true decorator 254 return deco_retry 255