2 Manage pools of connections so that we can limit the number of requests per site and reuse
7 # Copyright (C) 2011, Thomas Leonard
8 # See the README file for details, or visit http://0install.net.
12 if sys
.version_info
[0] > 2:
13 from urllib
import parse
as urlparse
# Python 3
17 from collections
import defaultdict
18 import threading
, gobject
20 from zeroinstall
.support
import tasks
21 from zeroinstall
.injector
import download
33 class DownloadScheduler
:
34 """Assigns (and re-assigns on redirect) Downloads to Sites, allowing per-site limits and connection pooling.
37 self
._sites
= defaultdict(lambda: Site()) # (scheme://host:port) -> Site
40 def download(self
, dl
):
41 # (changed if we get redirected)
44 redirections_remaining
= 10
46 # Assign the Download to a Site based on its scheme, host and port. If the result is a redirect,
47 # reassign it to the appropriate new site. Note that proxy handling happens later; we want to group
48 # and limit by the target site, not treat everything as going to a single site (the proxy).
50 location_parts
= urlparse
.urlparse(current_url
)
52 site_key
= (location_parts
.scheme
,
53 location_parts
.hostname
,
54 location_parts
.port
or default_port
.get(location_parts
.scheme
, None))
58 step
.url
= current_url
59 blocker
= self
._sites
[site_key
].download(step
)
66 current_url
= step
.redirect
68 if redirections_remaining
== 0:
69 raise download
.DownloadError("Too many redirections {url} -> {current}".format(
71 current
= current_url
))
72 redirections_remaining
-= 1
73 # (else go around the loop again)
75 MAX_DOWNLOADS_PER_SITE
= 5
77 def _spawn_thread(step
):
78 from ._download
_child
import download_in_thread
80 thread_blocker
= tasks
.Blocker("wait for thread " + step
.url
)
81 def notify_done(status
, ex
= None, redirect
= None):
83 step
.redirect
= redirect
86 thread_blocker
.trigger(ex
)
88 gobject
.idle_add(wake_up_main
)
89 child
= threading
.Thread(target
= lambda: download_in_thread(step
.url
, step
.dl
.tempfile
, step
.dl
.modification_time
, notify_done
))
96 """Represents a service accepting download requests. All requests with the same scheme, host and port are
97 handled by the same Site object, allowing it to do connection pooling and queuing, although the current
98 implementation doesn't do either."""
104 def download(self
, step
):
105 if self
.active
== MAX_DOWNLOADS_PER_SITE
:
106 # Too busy to start a new download now. Queue this one and wait.
107 ticket
= tasks
.Blocker('queued download for ' + step
.url
)
108 self
.queue
.append(ticket
)
109 yield ticket
, step
.dl
._aborted
110 if step
.dl
._aborted
.happened
:
111 raise download
.DownloadAborted()
113 # Start a new thread for the download
114 thread_blocker
= _spawn_thread(step
)
118 # Wait for thread to complete download.
119 yield thread_blocker
, step
.dl
._aborted
122 if self
.active
< MAX_DOWNLOADS_PER_SITE
:
123 self
.process_next() # Start next queued download, if any
125 if step
.dl
._aborted
.happened
:
126 # Don't wait for child to finish (might be stuck doing IO)
127 raise download
.DownloadAborted()
129 tasks
.check(thread_blocker
)
131 if step
.status
== download
.RESULT_REDIRECT
:
133 return # DownloadScheduler will handle it
135 assert not step
.redirect
, step
.redirect
137 step
.dl
._finish
(step
.status
)
139 def process_next(self
):
140 assert self
.active
< MAX_DOWNLOADS_PER_SITE
143 nxt
= self
.queue
.pop()