Performance is worse than terrible

[inv2004]

mongodb = "0.3.10"

rustc --version
rustc 1.30.0-nightly (4591a245c 2018-09-22)

mongod --version
db version v4.0.2

Rust:
https://gist.github.com/inv2004/87b49ee4217872d9260e13a09e6bf524

$ time ./target/release/bench_mongo
real    0m44.043s

Go:
https://gist.github.com/inv2004/d8e397ee7d0553940b60dff7f3005540

$ time ./bench_mongo
real    0m0.798s

---added---
mongo_driver:
real 0m0.583s

[vaffeine]

I took a quick look at the problem with callgrind. The first thing that caught my eye was a huge amount of allocations caused by in bson::OrderedDocument::clone(), and then the same amount of deallocations, when OrderedDocument is dropped. In case of your gist OrderedDocument::clone() is called 9046, which results in 243000 calls to both __rust__alloc and __rust_dealloc at the bottom of the call stack.
I was not able to try the same thing with Go, but I believe it doesn't allocate and then immediately drop document for each insert.
The other thing is that Collection::insert() itself took only 16% of total execution time. However, Cursor::query() took 59%. I am pretty new to this library and MongoDB itself, so I don't know if it is normal.
Going to take a look at Go and PyMongo to makes the difference.

[inv2004]

Thank you, its s good point, because mongo_driver uses doc by ref also. Unfortunately, I do not think that these allocations is the main point, because initial problem was found in the code, which generates uniq doc's.

Also, I did not make any special benchmarks, but I see that .find in mongo_driver is even much faster than mongodb. ~5k/1sec (native) vs ~15k/1sec (WSL)

[inv2004]

Would you like me to do another test without allocation?

Regards,