Honza Kral
30bbb2a0cd
Cut off the timeout increases after 5 failures by default
...
Contrary to popular beliefs there are Python processes in the wild that
run for weeks and increasing the timeout without bound is not useful for
those.
2013-06-14 16:59:36 +02:00
Honza Kral
c686cd2be3
Have error counting contained within ConnectionPool and don't leak out to Transport
...
Thanks @bleskes for the inspiration!
2013-06-14 16:46:58 +02:00
Honza Kral
dd19553ba7
Have sniff_hosts reuse existing connection to not waste open sockets
2013-05-24 01:29:31 +02:00
Honza Kral
1e81d09b14
Docstrings for connection_pool
2013-05-22 20:30:44 +02:00
Honza Kral
9e6714e2fa
Add a RandomSelector
2013-05-22 20:30:38 +02:00
Honza Kral
5af5e36c93
In case of race condition, don't block on resurrecting
2013-05-13 17:40:22 +02:00
Honza Kral
603ec53387
TODO is DONE
2013-05-05 04:51:32 +02:00
Honza Kral
379d443cfa
If a connection fails Nth time in a row put it on timeout for 2**(N-1) * default_timeout
2013-05-05 01:00:10 +02:00
Honza Kral
b4ba7433bd
Make sure dead connections are handled in a thread-safe manner
2013-05-05 00:36:39 +02:00
Honza Kral
8cd053799e
When resurrect happens it shouldn't return the risen connection, just return it to pool
2013-05-03 03:19:49 +02:00
Honza Kral
e3c76ecf01
ConnetionPool to hold all the connections
...
It provides load balancing and keeps track of the status of individual
connections (dead/alive) and penalizes dead connections with a timeout.
2013-05-02 17:54:26 +02:00
Honza Kral