Edit File: mirror.py
# This library is free software; you can redistribute it and/or # modify it under the terms of the GNU Lesser General Public # License as published by the Free Software Foundation; either # version 2.1 of the License, or (at your option) any later version. # # This library is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU # Lesser General Public License for more details. # # You should have received a copy of the GNU Lesser General Public # License along with this library; if not, write to the # Free Software Foundation, Inc., # 59 Temple Place, Suite 330, # Boston, MA 02111-1307 USA # This file is part of urlgrabber, a high-level cross-protocol url-grabber # Copyright 2002-2004 Michael D. Stenner, Ryan Tomayko """Module for downloading files from a pool of mirrors DESCRIPTION This module provides support for downloading files from a pool of mirrors with configurable failover policies. To a large extent, the failover policy is chosen by using different classes derived from the main class, MirrorGroup. Instances of MirrorGroup (and cousins) act very much like URLGrabber instances in that they have urlread, urlgrab, and urlopen methods. They can therefore, be used in very similar ways. from urlgrabber.grabber import URLGrabber from urlgrabber.mirror import MirrorGroup gr = URLGrabber() mg = MirrorGroup(gr, ['http://foo.com/some/directory/', 'http://bar.org/maybe/somewhere/else/', 'ftp://baz.net/some/other/place/entirely/'] mg.urlgrab('relative/path.zip') The assumption is that all mirrors are identical AFTER the base urls specified, so that any mirror can be used to fetch any file. FAILOVER The failover mechanism is designed to be customized by subclassing from MirrorGroup to change the details of the behavior. In general, the classes maintain a master mirror list and a "current mirror" index. When a download is initiated, a copy of this list and index is created for that download only. The specific failover policy depends on the class used, and so is documented in the class documentation. Note that ANY behavior of the class can be overridden, so any failover policy at all is possible (although you may need to change the interface in extreme cases). CUSTOMIZATION Most customization of a MirrorGroup object is done at instantiation time (or via subclassing). There are four major types of customization: 1) Pass in a custom urlgrabber - The passed in urlgrabber will be used (by default... see #2) for the grabs, so options to it apply for the url-fetching 2) Custom mirror list - Mirror lists can simply be a list of stings mirrors (as shown in the example above) but each can also be a dict, allowing for more options. For example, the first mirror in the list above could also have been: {'mirror': 'http://foo.com/some/directory/', 'grabber': <a custom grabber to be used for this mirror>, 'kwargs': { <a dict of arguments passed to the grabber> }} All mirrors are converted to this format internally. If 'grabber' is omitted, the default grabber will be used. If kwargs are omitted, then (duh) they will not be used. kwarg 'max_connections' limits the number of concurrent connections to this mirror. When omitted or set to zero, the default limit (2) will be used. 3) Pass keyword arguments when instantiating the mirror group. See, for example, the failure_callback argument. 4) Finally, any kwargs passed in for the specific file (to the urlgrab method, for example) will be folded in. The options passed into the grabber's urlXXX methods will override any options specified in a custom mirror dict. """ import sys import random import thread # needed for locking to make this threadsafe from grabber import URLGrabError, CallbackObject, DEBUG, _to_utf8 from grabber import _run_callback, _do_raise from grabber import exception2msg from grabber import _TH def _(st): return st class GrabRequest: """This is a dummy class used to hold information about the specific request. For example, a single file. By maintaining this information separately, we can accomplish two things: 1) make it a little easier to be threadsafe 2) have request-specific parameters """ pass class MirrorGroup: """Base Mirror class Instances of this class are built with a grabber object and a list of mirrors. Then all calls to urlXXX should be passed relative urls. The requested file will be searched for on the first mirror. If the grabber raises an exception (possibly after some retries) then that mirror will be removed from the list, and the next will be attempted. If all mirrors are exhausted, then an exception will be raised. MirrorGroup has the following failover policy: * downloads begin with the first mirror * by default (see default_action below) a failure (after retries) causes it to increment the local AND master indices. Also, the current mirror is removed from the local list (but NOT the master list - the mirror can potentially be used for other files) * if the local list is ever exhausted, a URLGrabError will be raised (errno=256, No more mirrors). The 'errors' attribute holds a list of (full_url, errmsg) tuples. This contains all URLs tried and the corresponding error messages. OPTIONS In addition to the required arguments "grabber" and "mirrors", MirrorGroup also takes the following optional arguments: default_action A dict that describes the actions to be taken upon failure (after retries). default_action can contain any of the following keys (shown here with their default values): default_action = {'increment': 1, 'increment_master': 1, 'remove': 1, 'remove_master': 0, 'fail': 0} In this context, 'increment' means "use the next mirror" and 'remove' means "never use this mirror again". The two 'master' values refer to the instance-level mirror list (used for all files), whereas the non-master values refer to the current download only. The 'fail' option will cause immediate failure by re-raising the exception and no further attempts to get the current download. As in the "No more mirrors" case, the 'errors' attribute is set in the exception object. This dict can be set at instantiation time, mg = MirrorGroup(grabber, mirrors, default_action={'fail':1}) at method-execution time (only applies to current fetch), filename = mg.urlgrab(url, default_action={'increment': 0}) or by returning an action dict from the failure_callback return {'fail':0} in increasing precedence. If all three of these were done, the net result would be: {'increment': 0, # set in method 'increment_master': 1, # class default 'remove': 1, # class default 'remove_master': 0, # class default 'fail': 0} # set at instantiation, reset # from callback failure_callback this is a callback that will be called when a mirror "fails", meaning the grabber raises some URLGrabError. If this is a tuple, it is interpreted to be of the form (cb, args, kwargs) where cb is the actual callable object (function, method, etc). Otherwise, it is assumed to be the callable object itself. The callback will be passed a grabber.CallbackObject instance along with args and kwargs (if present). The following attributes are defined within the instance: obj.exception = < exception that was raised > obj.mirror = < the mirror that was tried > obj.tries = < the number of mirror tries so far > obj.relative_url = < url relative to the mirror > obj.url = < full url that failed > # .url is just the combination of .mirror # and .relative_url The failure callback can return an action dict, as described above. Like default_action, the failure_callback can be set at instantiation time or when the urlXXX method is called. In the latter case, it applies only for that fetch. The callback can re-raise the exception quite easily. For example, this is a perfectly adequate callback function: def callback(obj): raise obj.exception WARNING: do not save the exception object (or the CallbackObject instance). As they contain stack frame references, they can lead to circular references. Notes: * The behavior can be customized by deriving and overriding the 'CONFIGURATION METHODS' * The 'grabber' instance is kept as a reference, not copied. Therefore, the grabber instance can be modified externally and changes will take effect immediately. """ # notes on thread-safety: # A GrabRequest should never be shared by multiple threads because # it's never saved inside the MG object and never returned outside it. # therefore, it should be safe to access/modify grabrequest data # without a lock. However, accessing the mirrors and _next attributes # of the MG itself must be done when locked to prevent (for example) # removal of the wrong mirror. ############################################################## # CONFIGURATION METHODS - intended to be overridden to # customize behavior def __init__(self, grabber, mirrors, **kwargs): """Initialize the MirrorGroup object. REQUIRED ARGUMENTS grabber - URLGrabber instance mirrors - a list of mirrors OPTIONAL ARGUMENTS failure_callback - callback to be used when a mirror fails default_action - dict of failure actions See the module-level and class level documentation for more details. """ # OVERRIDE IDEAS: # shuffle the list to randomize order self.grabber = grabber self.mirrors = self._parse_mirrors(mirrors) self._next = 0 self._lock = thread.allocate_lock() self.default_action = None self._process_kwargs(kwargs) # use the same algorithm as parallel downloader to initially sort # the mirror list (sort by speed, but prefer live private mirrors) def estimate(m): speed, fail = _TH.estimate(m['mirror']) private = not fail and m.get('kwargs', {}).get('private', False) return private, speed # update the initial order. since sorting is stable, the relative # order of unknown (not used yet) hosts is retained. self.mirrors.sort(key=estimate, reverse=True) # if these values are found in **kwargs passed to one of the urlXXX # methods, they will be stripped before getting passed on to the # grabber options = ['default_action', 'failure_callback'] def _process_kwargs(self, kwargs): self.failure_callback = kwargs.get('failure_callback') self.default_action = kwargs.get('default_action') def _parse_mirrors(self, mirrors): parsed_mirrors = [] for m in mirrors: if isinstance(m, basestring): m = {'mirror': _to_utf8(m)} parsed_mirrors.append(m) return parsed_mirrors def _load_gr(self, gr): # OVERRIDE IDEAS: # shuffle gr list self._lock.acquire() gr.mirrors = list(self.mirrors) gr._next = self._next self._lock.release() def _get_mirror(self, gr): # OVERRIDE IDEAS: # return a random mirror so that multiple mirrors get used # even without failures. if not gr.mirrors: e = URLGrabError(256, _('No more mirrors to try.')) e.errors = gr.errors raise e return gr.mirrors[gr._next] def _failure(self, gr, cb_obj): # OVERRIDE IDEAS: # inspect the error - remove=1 for 404, remove=2 for connection # refused, etc. (this can also be done via # the callback) cb = gr.kw.get('failure_callback') or self.failure_callback if cb: if type(cb) == type( () ): cb, args, kwargs = cb else: args, kwargs = (), {} action = cb(cb_obj, *args, **kwargs) or {} else: action = {} # XXXX - decide - there are two ways to do this # the first is action-overriding as a whole - use the entire action # or fall back on module level defaults #action = action or gr.kw.get('default_action') or self.default_action # the other is to fall through for each element in the action dict a = dict(self.default_action or {}) a.update(gr.kw.get('default_action', {})) a.update(action) action = a self.increment_mirror(gr, action) if action and action.get('fail', 0): sys.exc_info()[1].errors = gr.errors raise def increment_mirror(self, gr, action={}): """Tell the mirror object increment the mirror index This increments the mirror index, which amounts to telling the mirror object to use a different mirror (for this and future downloads). This is a SEMI-public method. It will be called internally, and you may never need to call it. However, it is provided (and is made public) so that the calling program can increment the mirror choice for methods like urlopen. For example, with urlopen, there's no good way for the mirror group to know that an error occurs mid-download (it's already returned and given you the file object). remove --- can have several values 0 do not remove the mirror from the list 1 remove the mirror for this download only 2 remove the mirror permanently beware of remove=0 as it can lead to infinite loops """ badmirror = gr.mirrors[gr._next] self._lock.acquire() try: ind = self.mirrors.index(badmirror) except ValueError: pass else: if action.get('remove_master', 0): del self.mirrors[ind] elif self._next == ind and action.get('increment_master', 1): self._next += 1 if self._next >= len(self.mirrors): self._next = 0 self._lock.release() if action.get('remove', 1): del gr.mirrors[gr._next] elif action.get('increment', 1): gr._next += 1 if gr._next >= len(gr.mirrors): gr._next = 0 if DEBUG: grm = [m['mirror'] for m in gr.mirrors] DEBUG.info('GR mirrors: [%s] %i', ' '.join(grm), gr._next) selfm = [m['mirror'] for m in self.mirrors] DEBUG.info('MAIN mirrors: [%s] %i', ' '.join(selfm), self._next) ##################################################################### # NON-CONFIGURATION METHODS # these methods are designed to be largely workhorse methods that # are not intended to be overridden. That doesn't mean you can't; # if you want to, feel free, but most things can be done by # by overriding the configuration methods :) def _join_url(self, base_url, rel_url): if base_url.endswith('/') or rel_url.startswith('/'): return base_url + rel_url else: return base_url + '/' + rel_url def _mirror_try(self, func, url, kw): gr = GrabRequest() gr.func = func gr.url = url gr.kw = dict(kw) self._load_gr(gr) gr.errors = [] for k in self.options: try: del kw[k] except KeyError: pass tries = 0 while 1: tries += 1 mirrorchoice = self._get_mirror(gr) fullurl = self._join_url(mirrorchoice['mirror'], gr.url) grabber = mirrorchoice.get('grabber') or self.grabber # apply mirrorchoice kwargs on top of grabber.opts opts = grabber.opts.derive(**mirrorchoice.get('kwargs', {})) func_ref = getattr(grabber, func) if DEBUG: DEBUG.info('MIRROR: trying %s -> %s', url, fullurl) try: return func_ref( *(fullurl,), opts=opts, **kw ) except URLGrabError, e: if DEBUG: DEBUG.info('MIRROR: failed') gr.errors.append((fullurl, exception2msg(e))) obj = CallbackObject() obj.exception = e obj.mirror = mirrorchoice['mirror'] obj.relative_url = gr.url obj.url = fullurl obj.tries = tries self._failure(gr, obj) def urlgrab(self, url, filename=None, **kwargs): kw = dict(kwargs) kw['filename'] = filename if kw.get('async'): # enable mirror failovers in async path kw['mirror_group'] = self, [], {}, set() kw['relative_url'] = url else: kw.pop('failfunc', None) func = 'urlgrab' try: return self._mirror_try(func, url, kw) except URLGrabError, e: obj = CallbackObject(url=url, filename=filename, exception=e, **kwargs) return _run_callback(kwargs.get('failfunc', _do_raise), obj) def urlopen(self, url, **kwargs): kw = dict(kwargs) func = 'urlopen' return self._mirror_try(func, url, kw) def urlread(self, url, limit=None, **kwargs): kw = dict(kwargs) kw['limit'] = limit func = 'urlread' return self._mirror_try(func, url, kw) class MGRandomStart(MirrorGroup): """A mirror group that starts at a random mirror in the list. This behavior of this class is identical to MirrorGroup, except that it starts at a random location in the mirror list. """ def __init__(self, grabber, mirrors, **kwargs): """Initialize the object The arguments for intialization are the same as for MirrorGroup """ MirrorGroup.__init__(self, grabber, mirrors, **kwargs) self._next = random.randrange(len(mirrors)) class MGRandomOrder(MirrorGroup): """A mirror group that uses mirrors in a random order. This behavior of this class is identical to MirrorGroup, except that it uses the mirrors in a random order. Note that the order is set at initialization time and fixed thereafter. That is, it does not pick a random mirror after each failure. """ def __init__(self, grabber, mirrors, **kwargs): """Initialize the object The arguments for intialization are the same as for MirrorGroup """ MirrorGroup.__init__(self, grabber, mirrors, **kwargs) random.shuffle(self.mirrors) if __name__ == '__main__': pass