Added connection_pool_size configuration property (preview)

[nfx]

We have three things that are configurable from requests connection pooling perspective:

• pool_connections - Number of urllib3 connection pools to cache before discarding the least recently used pool. Python requests default value is 10. This PR increases it to 20.
• pool_maxsize - The maximum number of connections to save in the pool. Improves performance in multithreaded situations. For now, we're setting it to the same value as connection_pool_size.
• pool_block - If pool_block is False, then more connections will are created, but not saved after the first use. Block when no free connections are available. urllib3 ensures that no more than pool_maxsize connections are used at a time. Prevents platform from flooding. By default, requests library doesn't block. We start with blocking.

This PR introduces two new configuration properties:

• connection_pool_size configuration property, which sets pool_connections. Defaults to 20.
• connection_pool_max_size, which sets pool_maxsize. Defaults to the same value as connection_pool_size.

[mgyucht]

+1 on increasing connection pool size & blocking. However, I don't know about increasing # of connection pools.

[mgyucht]

I understand the need to increase the connection pool size to support concurrent requests. However, why do we need multiple connection pools? Looking at requests, HttpAdapter's underlying PoolManager uses the same pool based on the <host,port,scheme> tuple, which for a single client will only be one endpoint.

[nfx]

I also do think that we need just one pool, but somehow the default client was using the same value for both.
how about exposing two configuration properties, marking them experimental, and figure out the best defaults later? single SDK client will query only one host.

how should we name pool_maxsize? connection_pool_max?

[mgyucht]

Default uses the same value for both because it uses the same constant for each parameter as default values in its constructor:

    def __init__(
        self,
        pool_connections=DEFAULT_POOLSIZE,
        pool_maxsize=DEFAULT_POOLSIZE,
        max_retries=DEFAULT_RETRIES,
        pool_block=DEFAULT_POOLBLOCK,
    ):

DEFAULT_POOLSIZE is set to 10.

[mgyucht]

Small nit on naming, but otherwise the change seems fine.

[mgyucht]

Here's my suggestion for the naming:

• max_connection_pools: maximum number of connection pools used by the HttpAdapter.
• max_connections_per_pool: maximum number of connections per pool used by the HttpAdapter.

[nfx]

sure. let's get it shipped

[codecov-commenter]

Codecov Report

Patch coverage: 100.00% and project coverage change: +0.02% 🎉

Comparison is base (b82fa01) 52.55% compared to head (6ed5b70) 52.57%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #276      +/-   ##
==========================================
+ Coverage   52.55%   52.57%   +0.02%     
==========================================
  Files          33       33              
  Lines       19823    19833      +10     
==========================================
+ Hits        10418    10428      +10     
  Misses       9405     9405              

Files Changed	Coverage Δ
databricks/sdk/core.py	67.14% <100.00%> (+0.43%)	⬆️

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.