Performance regression when getting multiple connections with r2d2::Pool

[heyrict]

Setup

Versions

• Rust: rustc 1.62.0-nightly (e85edd9a8 2022-04-28)
• Diesel: v2.0.0-rc0
• Database: postgresql
• Operating System Linux 5.17.3-arch1-1

Feature Flags

• diesel: ["postgres", "r2d2"]

Problem Description

As is reported in #3144, performance regression is observed when upgrading diesel from v1.4.8 to v2.0.0-rc0. Aggregation queries (e.g. count(*) are particularly affected, while the performance hit on normal queries is minimal.

What are you trying to accomplish?

What is the expected output?

The performance of v2.0.0-rc0 should be on par with v1.4.8.

What is the actual output?

cargo bench output of table.filter(id.eq(1)).count().get_result():

hash	setup	perf
d0bbda8	diesel 1.4 + r2d2	185,936 ns/iter (+/- 15,282)
599c492	diesel 2.0 + r2d2	1,388,410 ns/iter (+/- 158,035)
d3c23ba	diesel 1.4 - r2d2	103,562 ns/iter (+/- 2,997)
58d1210	diesel 2.0 - r2d2	109,238 ns/iter (+/- 5,887)

Are you seeing any additional errors?

Steps to reproduce

The minimal reproduction sample is here.

Checklist

• I have already looked over the issue tracker and the disussion forum for similar possible closed issues.

• This issue can be reproduced on Rust's stable channel. (Your issue will be
  closed if this is not the case)
• This issue can be reproduced without requiring a third party crate

[weiznich]

This is caused by broken R2D2Connection::is_broken implementations in diesel. Those result in marking any connection as broken as soon as it is checked into the pool again. In turn that results in not reusing connections at all, so that the "pool" reestablish a new connection every time you checked out an connection from the pool.

#3159 addresses this. Can you verify that this solves the issue for you?

[heyrict]

I've changed the title to reflect the actual bug. Can confirm the bug is fixed with your PR. Thank you for debugging this 👍