nsqd: add pool to minimize buffer allocation on message send

[imxyb]

No description provided.

[ploxiln]

Maybe the function names should be more specific, like GetMessageBuffer()?

Do you have a benchmark that shows a significant improvement in performance from this change?

[imxyb]

Maybe the function names should be more specific, like GetMessageBuffer()?

Do you have a benchmark that shows a significant improvement in performance from this change?

Thank you for your advice. I see the buffer in the paclage nsqd:buffer_ pool.go have similar codes, can I extract this part of the code for common use?

[ploxiln]

It looks like it is already extracted, and you can just use it, in one more place.

[imxyb]

have a benchmark that shows a significant improvement in performance from this change?

i use the `bench.sh' make this bench test:

before:

# using --mem-queue-size=1000000 --data-path= --size=200 --batch-size=200
# compiling/running nsqd
# creating topic/channel
# compiling bench_reader/bench_writer
PUB: [bench_writer] 2020/11/01 12:00:24 duration: 10.021804433s - 32.735mb/s - 171625.780ops/s - 5.827us/op
SUB: [bench_reader] 2020/11/01 12:00:34 duration: 10.00603778s - 32.787mb/s - 171896.213ops/s - 5.817us/op
waiting for pprof...

after:

# using --mem-queue-size=1000000 --data-path= --size=200 --batch-size=200
# compiling/running nsqd
# creating topic/channel
# compiling bench_reader/bench_writer
PUB: [bench_writer] 2020/11/01 12:15:40 duration: 10.027253623s - 33.219mb/s - 174165.336ops/s - 5.742us/op
SUB: [bench_reader] 2020/11/01 12:15:50 duration: 10.00180929s - 33.304mb/s - 174608.408ops/s - 5.727us/op
waiting for pprof...

[ploxiln]

Looks good to me, thanks. The commit title could be a bit better, I suggest something like

nsqd: use buffer pool for serializing message for client

[mreiferson]

LGTM other than that small question

[imxyb]

@ploxiln the latest benchmark:

# using --mem-queue-size=1000000 --data-path= --size=200 --batch-size=200
# compiling/running nsqd
# creating topic/channel
# compiling bench_reader/bench_writer
PUB: [bench_writer] 2020/11/12 10:21:33 duration: 10.029296172s - 33.213mb/s - 174129.866ops/s - 5.743us/op
SUB: [bench_reader] 2020/11/12 10:21:43 duration: 10.005897058s - 33.290mb/s - 174537.074ops/s - 5.729us/op

[ploxiln]

thanks!