`using AMQPClient` significantly slows down code

General Usage Performance
1

Hello,

I came across this weird performance issue, can someone please explain why using AMQPClient causes ~100x performance drop?

julia> m = Matrix{Any}(rand(1000,1000));

julia> function f(m, n)
           for i in 1:n
               m = vcat(m, zeros(1, 1000))
           end 
       end
f (generic function with 1 method)

julia> @time f(m, 10)
  0.280798 seconds (355.77 k allocations: 94.275 MiB, 23.08% gc time)

julia> @time f(m, 10)
  0.072561 seconds (10.04 k allocations: 76.945 MiB, 19.58% gc time)

julia> using AMQPClient

julia> @time f(m, 10)
  6.048357 seconds (132.75 k allocations: 82.892 MiB, 0.17% gc time)

julia> @time f(m, 10)
  6.084253 seconds (10.04 k allocations: 76.945 MiB, 0.17% gc time)
2

Henh. Strange. I don’t know what’s going on here. I can reproduce on julia 1.2.0 on Linux. But if I change m to

m = rand(1000,1000)

then using AMPQClient no longer causes a slowdown in f.

3

I was able to narrow it down to:

julia> m = Matrix{Any}(rand(1000,1000));

julia> function f(m, n)
           for i in 1:n
               m = vcat(m, zeros(1, 1000))
           end 
       end
f (generic function with 1 method)

julia> @time f(m, 10);
  0.306576 seconds (382.74 k allocations: 95.807 MiB, 37.96% gc time)

julia> @time f(m, 10);
  0.064143 seconds (10.04 k allocations: 76.945 MiB, 18.77% gc time)

julia> import Base.convert

julia> struct MyType
       end

julia> convert(::Type{Any}, o::MyType) = ""
convert (generic function with 186 methods)

julia> @time f(m, 10);
  9.189991 seconds (154.01 k allocations: 84.212 MiB, 0.13% gc time)
3 Likes
4

Very interesting. Nice find. Are you able to open an issue about this?

5

I have a PR to update AMQPClient now: https://github.com/JuliaComputing/AMQPClient.jl/pull/25