云鉴权>> warp-mem-leak-test>> 返回
项目作者： brianbruggeman

项目描述 ：
Investigate Warp's memory leak
高级语言： Shell
项目地址: git://github.com/brianbruggeman/warp-mem-leak-test.git
创建时间： 2020-06-18T22:32:35Z
项目社区：https://github.com/brianbruggeman/warp-mem-leak-test
开源协议：
下载


Warp Memory leak test

This tiny repo is intended to be useful for showing memory leaks in Hyper/Warp.

Motivation

In testing out my Warp application (private), I noticed a memory leak. This
lead me to look into Hyper and see #1790, which sounds like it’s been closed
because the issue has not been reproducable by the maintainers. Hopefully,
this repository allows someone to reproduce the memory leak consistently.

Getting started

Assuming all of the prerequisites have been installed, from the top of the repo,
just run:

  1. $ ./test-me.sh

Environment variables

The following variables are used to control the application and benchmarking.

  • HOST - the host name to use (default: 127.0.0.1)
  • PORT - the port name to use (default: 3030)
  • DURATION - the duration of the load test (default: 3 seconds)
  • WORKERS - the number of simultaneous workers (default: 100)

These can be tweaked by adding a .env file in the root of the repository.

Load Testing tool

The benchmark tool uses Hey, though I suspect that any benchmarking tool could
be used. More specifically, I noticed that this problem was relatively easy to
produce within a very short period of time (under 5 seconds), though it becomes
much more obvious over time.

For the mac, install was as simple as brew install hey.

Environment setup

I have a 2015 mac running Macos 10.15.5. I am using rustc 1.44.0 to produce the
results.

I use Tokio‘s “full” feature so that there’s nothing missing
(e.g. multithreading).

Interpreting numbers

For the following settings (.env):

  1. TRIES=1
  2. HOST=127.0.0.1
  3. PORT=3030
  4. DURATION=1s
  5. WORKERS=1

Running test-me.sh yields something like:

  1. ----------------------------------------
  2. Settings:
  3. ----------------------------------------
  4. Iterating 1 times
  5. Loading 127.0.0.1:3030
  6. Running for 1s each iteration
  7. Using 1 workers
  8. ----------------------------------------
  9. Server running on PID: 72930
  10. [Warp App] Serving on: 127.0.0.1:3030
  11. ===============================
  12. Iteration: 1 of 1
  13. -------------------------------
  14. Requests/sec: 5895.6067
  15. Latency (worst case): 0.0005 secs
  16. -------------------------------
  17. Starting memory: 2180 KB
  18. Ending memory: 2644 KB
  19. Memory leaked: 464 KB
  20. ===============================

In this case, we’re seeing a 464 KB increase in RAM. If we let this run for
100 iterations, some of them will appear to be 0. We’re using ps here, and
explicitly examining the rss (Resident Set Size) field. Because of that, the
numbers are in KB rather than B. So the 0’s being reported may actually be more
than 0 B.