Falling throughput when extracting path information

[morphy2k]

Your issue may already be reported!
Please search on the Actix Web issue tracker before creating one.

Expected Behavior

Throughput should be constant.

Current Behavior

Throughput drops with the number of requests.

Possible Solution

...

Steps to Reproduce (for bugs)

Link to example

1. Build and start the server
2. Benchmark the root endpoint where no path information is extracted
3. Benchmark the ID endpoint where path information is extracted
4. Compare the results

Context

I noticed the issue while developing a search backend.
As soon as I extract path or query information, the throughput drops continuously during benchmarking.
In addition, regardless of whether I extract information, I noticed that the memory consumption is continuously increasing, which could indicate a memory leak.

Your Environment

• CPU Ryzen 5 2600
• RAM 16GB DDR4-3200
• OS Fedora 32 Workstation (5.7.8-200.fc32.x86_64)

• Rust Version (I.e, output of rustc -V): rustc 1.44.1 (c7087fe00 2020-06-17)
• Actix Web Version: 3.0.0-beta.1

[fakeshadow]

I believe this line is the cause of your problem.

actix-web/src/app_service.rs

Line 248 in 2fd96c0

inner.app_data.push(self.data.clone());

Commenting it and your example runs OK for me. It would be good if someone can look into it and see if it's ok to remove it.

[robjtede]

for context: #1486 (comment)

[fakeshadow]

Or maybe we can clear the data pushed to HttpRequest when dropping it too(If it returns to pool).
I'm not familiar with how cached pool works so I could be wrong but it seems in certain cases we just keep push data and the vec become too long.
Well actually if a request is already in cache do we really need to push data again?

[morphy2k]

Can confirm that this causes the problem, as well as the high memory consumption #1606
As soon as I comment out the line, the memory consumption and throughput remains constant.