When encountering a retryable error on a client derived from a pool, it can take a much longer time than normal to complete the retry on 1.17 than in previous versions. Interestingly, this has only been observed on evergreen hosts and not on any of my local machines.

Attached is a repro that times how long a single retry takes pooled vs non-pooled.

Output on 1.17.0:

======================== Non pooled version ====================
0.501870
======================== pooled version ====================
10.005833

Output on 1.16.3:

======================== Non pooled version ====================
0.501829
======================== pooled version ====================
0.502376