Reuse bytes when cleaning the URL paths

[panjf2000]

Signed-off-by: Andy Pan panjf2000@gmail.com

[codecov]

Codecov Report

Merging #2179 into master will increase coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #2179      +/-   ##
==========================================
+ Coverage    98.5%   98.51%   +<.01%     
==========================================
  Files          41       41              
  Lines        2276     2284       +8     
==========================================
+ Hits         2242     2250       +8     
  Misses         18       18              
  Partials       16       16

Impacted Files	Coverage Δ
path.go	100% <100%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update fd8a65b...bf153c3. Read the comment docs.

[panjf2000]

/cc @thinkerou

[appleboy]

Any benchmark report?

[panjf2000]

Not yet, but I can add some tests and get benchmark-results later.

[panjf2000]

@appleboy

BenchmarkCleanPath-8                     1000000              1321 ns/op             416 B/op         40 allocs/op
BenchmarkCleanPathSyncPool-8             1000000              2482 ns/op             848 B/op         40 allocs/op
BenchmarkCleanPathBytesPool-8            1000000              1818 ns/op             112 B/op         17 allocs/op

Time cost increases a little bit(maybe those test cases of paths are not general cuz those are mocks) and the memory allocs decreases to 25% when using bytes pool.

[panjf2000]

@thinkerou @appleboy
PTAL

[appleboy]

@panjf2000 maybe you should create PR in https://github.com/julienschmidt/httprouter first.

[panjf2000]

Good point, I will open a PR to httprouter later.

[panjf2000]

Profiling
CPU:

      flat  flat%   sum%        cum   cum%
    1140ms 23.22% 23.22%     2500ms 50.92%  github.com/gin-gonic/gin.cleanPathRaw
    1010ms 20.57% 43.79%     1010ms 20.57%  runtime.pthread_cond_signal
     560ms 11.41% 55.19%     1010ms 20.57%  runtime.mallocgc
     340ms  6.92% 62.12%     1060ms 21.59%  github.com/gin-gonic/gin.cleanPath
     150ms  3.05% 65.17%      250ms  5.09%  github.com/gin-gonic/gin.bufApp
     150ms  3.05% 68.23%      530ms 10.79%  github.com/gin-gonic/gin.bufAppRaw
     150ms  3.05% 71.28%      150ms  3.05%  runtime.memmove
     130ms  2.65% 73.93%      620ms 12.63%  runtime.makeslice
     120ms  2.44% 76.37%      660ms 13.44%  runtime.slicebytetostring
     110ms  2.24% 78.62%      110ms  2.24%  runtime.acquirem

Mem:

      flat  flat%   sum%        cum   cum%
  343.01MB 38.99% 38.99%   578.01MB 65.70%  github.com/gin-gonic/gin.cleanPathRaw
  235.01MB 26.71% 65.70%   235.01MB 26.71%  github.com/gin-gonic/gin.bufAppRaw
  217.01MB 24.67% 90.37%   217.01MB 24.67%  github.com/gin-gonic/gin.glob..func4
   52.50MB  5.97% 96.34%    52.50MB  5.97%  github.com/gin-gonic/gin.cleanPath
      28MB  3.18% 99.52%   245.01MB 27.85%  github.com/gin-gonic/gin.cleanPathSyncPool
         0     0% 99.52%   578.01MB 65.70%  github.com/gin-gonic/gin.BenchmarkCleanPath
         0     0% 99.52%    52.50MB  5.97%  github.com/gin-gonic/gin.BenchmarkCleanPathBytesPool
         0     0% 99.52%   245.01MB 27.85%  github.com/gin-gonic/gin.BenchmarkCleanPathSyncPool
         0     0% 99.52%   875.52MB 99.52%  testing.(*B).launch
         0     0% 99.52%   875.52MB 99.52%  testing.(*B).runN

cleanPathRaw is the current function and cleanPath is the one that uses byte pool.
The memory usage has been reduced to 15% after using a byte pool.

[appleboy]

Add ref: julienschmidt/httprouter#292

[appleboy]

@panjf2000 missing:

1. julienschmidt/httprouter@7b49e86
2. julienschmidt/httprouter@8e4b52b

right?

[panjf2000]

Actually julienschmidt/httprouter@7b49e86 is already added in this PR, only the julienschmidt/httprouter@8e4b52b is missed, I will add it now.

[appleboy]

@panjf2000 thanks.

[panjf2000]

@appleboy
Done.

[thinkerou]

LGTM