Consider setting connection_pool_maxsize to a higher value by default

[yuvipanda]

Currently, it is set to 1. Which means the moment you start doing concurrent requests to the same kubernetes master, you're going to end up thrashing the connection pool & reducing performance.

Since the most likely case is that we're going to be making many requests to the same IP rather than many requests to different IP, setting this to a larger value by default will (and does!) improve performance.

[pokoli]

I agree that a greater value will improve the performance but I don't know which will be a good default value.

@yuvipanda do you have any sugestion regarding the number of connections per api to keep?

[yuvipanda]

I like the ThreadPoolExecutor's default: https://docs.python.org/3/library/concurrent.futures.html#concurrent.futures.ThreadPoolExecutor, which is number of processors * 5.

…

On Wed, Jul 26, 2017 at 12:57 AM, Sergi Almacellas Abellana < ***@***.***> wrote: I agree that a greater value will improve the performance but I don't know which will be a good default value. @yuvipanda <https://github.com/yuvipanda> do you have any sugestion regarding the number of connections per api to keep? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#26 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAB23tLTIf1jbS8P__ffq4ATNL50lhZOks5sRvFbgaJpZM4OjZt1> .

-- Yuvi Panda T http://yuvi.in/blog

[pokoli]

I see we are using a None value as default, and the value of 1 comes from urllib3.
I like the idea of ussing None as default value, which means use the default value of the library.

IIUC the pool will create new connections if required (as by default block is set to false). See:

https://github.com/shazow/urllib3/blob/master/urllib3/connectionpool.py#L234

[yuvipanda]

@pokoli In http://urllib3.readthedocs.io/en/latest/advanced-usage.html:

By default only one connection will be saved for re-use. If you are making many requests to the same host simultaneously it might improve performance to increase this number:

>>> import urllib3
>>> http = urllib3.PoolManager(maxsize=10)
# Alternatively
>>> http = urllib3.HTTPConnectionPool('google.com', maxsize=10)

And in this case, we are doing exactly that - sending a lot of requests to one host (kubernetes master), rather than lots of requests to many hosts (what urllib3 optimizes for for default). So while our requests don't fail, it does mean that we aren't taking advantage of the pooling at all - as soon as a second concurrent request starts, it creates a new connection that's then discarded (not kept in the pool).

[pokoli]

@yuvipanda I just created pr #29 Can you please review/test? Thanks