Use LIFO for returning/retrieving connections from the pool #2085
Conversation
4ver
changed the title
There was a problem hiding this comment.
Please add this as a configurable option. A pool representing a read slave through a tcp load balancer works better with rr for instance, so there are likely folks who are relying on the current behavior. Both behaviors are valid, so it makes to add a configuration to choose.
|
Will do 👍 |
|
As for why it was initially implemented as rr it was before my time so I have no idea :) |
|
The implementation and test itself look good, though 👍 |
|
The test failures are a little funny, but it looks like I may have broken the |
|
Must have just been a randomly failing test. Should be good to go now 🙆 |
|
#1749 can probably be closed off if this pr gets merged. |
|
Gentle bump on this 😃. Had a chance to take a look over it @dougwilson ? |
|
Not since my initial look as I have been away. I will get back soon and take a look at the new changes |
|
All good 👍 Thanks for the message. |
dougwilson
force-pushed
the
master
branch
2 times, most recently
from
946727b
to
37fbbdd
Compare
|
Did you get a chance to take a look over it? @dougwilson |
The current way this works (cycling through all of the connections) means that you can't make use of mysql's
wait_timeoutsetting for closing inactive connections. Once a connection is created it will be persisted forever unless you have very low levels of activity.Was there a reason to cycle round-robin through the connections when this was initially implemented?