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
20 from zeroinstall
import gobject
21 from zeroinstall
.support
import tasks
22 from zeroinstall
.injector
import download
34 class DownloadScheduler
:
35 """Assigns (and re-assigns on redirect) Downloads to Sites, allowing per-site limits and connection pooling.
38 self
._sites
= defaultdict(lambda: Site()) # (scheme://host:port) -> Site
41 def download(self
, dl
):
42 # (changed if we get redirected)
45 redirections_remaining
= 10
47 # Assign the Download to a Site based on its scheme, host and port. If the result is a redirect,
48 # reassign it to the appropriate new site. Note that proxy handling happens later; we want to group
49 # and limit by the target site, not treat everything as going to a single site (the proxy).
51 location_parts
= urlparse
.urlparse(current_url
)
53 site_key
= (location_parts
.scheme
,
54 location_parts
.hostname
,
55 location_parts
.port
or default_port
.get(location_parts
.scheme
, None))
59 step
.url
= current_url
60 blocker
= self
._sites
[site_key
].download(step
)
67 current_url
= step
.redirect
69 if redirections_remaining
== 0:
70 raise download
.DownloadError("Too many redirections {url} -> {current}".format(
72 current
= current_url
))
73 redirections_remaining
-= 1
74 # (else go around the loop again)
76 MAX_DOWNLOADS_PER_SITE
= 5
78 def _spawn_thread(step
):
79 from ._download
_child
import download_in_thread
81 thread_blocker
= tasks
.Blocker("wait for thread " + step
.url
)
82 def notify_done(status
, ex
= None, redirect
= None):
84 step
.redirect
= redirect
87 thread_blocker
.trigger(ex
)
89 gobject
.idle_add(wake_up_main
)
90 child
= threading
.Thread(target
= lambda: download_in_thread(step
.url
, step
.dl
.tempfile
, step
.dl
.modification_time
, notify_done
))
97 """Represents a service accepting download requests. All requests with the same scheme, host and port are
98 handled by the same Site object, allowing it to do connection pooling and queuing, although the current
99 implementation doesn't do either."""
105 def download(self
, step
):
106 if self
.active
== MAX_DOWNLOADS_PER_SITE
:
107 # Too busy to start a new download now. Queue this one and wait.
108 ticket
= tasks
.Blocker('queued download for ' + step
.url
)
109 self
.queue
.append(ticket
)
110 yield ticket
, step
.dl
._aborted
111 if step
.dl
._aborted
.happened
:
112 raise download
.DownloadAborted()
114 # Start a new thread for the download
115 thread_blocker
= _spawn_thread(step
)
119 # Wait for thread to complete download.
120 yield thread_blocker
, step
.dl
._aborted
123 if self
.active
< MAX_DOWNLOADS_PER_SITE
:
124 self
.process_next() # Start next queued download, if any
126 if step
.dl
._aborted
.happened
:
127 # Don't wait for child to finish (might be stuck doing IO)
128 raise download
.DownloadAborted()
130 tasks
.check(thread_blocker
)
132 if step
.status
== download
.RESULT_REDIRECT
:
134 return # DownloadScheduler will handle it
136 assert not step
.redirect
, step
.redirect
138 step
.dl
._finish
(step
.status
)
140 def process_next(self
):
141 assert self
.active
< MAX_DOWNLOADS_PER_SITE
144 nxt
= self
.queue
.pop()