Merge lp:~ralph-bean/pylibravatar/tcp-dns into lp:pylibravatar

[Ralph Bean]

Use TCP for SRV lookups

Use tcp for SRV lookups.

By default, the DNS module is using udp. I discovered this because it would
periodically time out on our server -- requests would sometimes take 30s that
would normally take less than 1s. I'd like to use tcp for our use case in
order to avoid timeouts. However, that alone is not reason enough to ask
everyone else to use tcp. They may want udp. It is faster, right?

While playing with this, I found that on average tcp is repeatably faster than
udp for these kinds of requests.

See tcp:

  $ time python -c """
  import DNS
  DNS.DiscoverNameServers()
  [DNS.Request(name='_avatars._tcp.ralph.id.fedoraproject.org', qtype='SRV', protocol='tcp').req() for i in range(128)]
  """
  python -c 0.10s user 0.03s system 10% cpu 1.270 total

Versus udp:

  $ time python -c """
  import DNS
  DNS.DiscoverNameServers()
  [DNS.Request(name='_avatars._tcp.ralph.id.fedoraproject.org', qtype='SRV', protocol='udp').req() for i in range(128)]
  """
  python -c 0.24s user 0.11s system 3% cpu 9.514 total

Which makes it a more suitable default for libravatar.py, no?

[François Marier (fmarier)]

Thanks!

review: Approve

=== modified file 'libravatar.py'
--- libravatar.py	2014-09-01 23:28:18 +0000
+++ libravatar.py	2015-06-26 18:40:19 +0000
@@ -165,7 +165,7 @@
     DNS.DiscoverNameServers()
     try:
         dns_request = DNS.Request(name=service_name(domain, https),
-                                  qtype='SRV').req()
+                                  qtype='SRV', protocol='tcp').req()
     except DNS.DNSError as message:
         print("DNS Error: %s" % message)
         return None