decrease the memory footprint of request/reply procedure

[VladoLavor]

See issue #254

Instead of allocating new temporary request/reply objects for each message call and then leaving them for GC to collect, put them into sync.Pool instead. The follow-up call can pick previously allocated request/reply from the pool.

This approach is supposed to decrease the total number of allocations and STWs due to the GC runs.

Signed-off-by: Vladimir Lavor [email protected]

[VladoLavor]

Providing test results using the API memory benchmark (GoVPP + borrowed from Frouter). The API RPS tends to fluctuate but seems to be improved in most measurements. The memory footprint is decreased as demonstrated by example results below.

Full results can be found in attached files:
govpp_api_memory_rps_perf_current_code.txt
govpp_api_memory_rps_perf_patched.txt

Simple API:

Current code

┌──────────────────────────────────────────────────┐
│ Simple API 1000000 calls                         │
├──────────────────────────────────────────────────┤
│ Time: 47.252972943s                              │
│ ~API calls: 1000000                              │
│ Rate per second: 21,162                          │ <--
│ Result: PASSED                                   │
├────────────────┬────────────────┬────────────────┤
│                │    Measured    │    Threshold   │
├────────────────┼────────────────┼────────────────┤
│ Total alloc    │      744.12 MB │        1.20 GB │ <--
│ Memory Freed   │      743.11 MB │                │
│ Heap alloc     │        1.01 MB │          10 MB │
│ Objects alloc  │     13,498,928 │                │ <--
│ Obj freed      │     13,481,554 │                │
│ Objects remain │         17,374 │        200,000 │
│ Num goroutines │             11 │                │
└────────────────┴────────────────┴────────────────┘

Patched

┌──────────────────────────────────────────────────┐
│ Simple API 1000000 calls                         │
├──────────────────────────────────────────────────┤
│ Time: 41.480308176s                              │
│ ~API calls: 1000000                              │
│ Rate per second: 24,107                          │ <--
│ Result: PASSED                                   │
├────────────────┬────────────────┬────────────────┤
│                │    Measured    │    Threshold   │
├────────────────┼────────────────┼────────────────┤
│ Total alloc    │      680.42 MB │        1.20 GB │ <--
│ Memory Freed   │      678.03 MB │                │
│ Heap alloc     │        2.40 MB │          10 MB │
│ Objects alloc  │     12,500,734 │                │ <--
│ Obj freed      │     12,457,508 │                │
│ Objects remain │         43,226 │        200,000 │
│ Num goroutines │             13 │                │
└────────────────┴────────────────┴────────────────┘

Frouter2 configuration:

Current code

┌──────────────────────────────────────────────────┐
│ Frouter2 API 40000 iterations:                   │
├──────────────────────────────────────────────────┤
│ Time: 2m32.136766488s     Iterations: 40000      │
│ ~API calls: 1000000                              │
│ Rate per second: 6,573                           │ <--
│ Result: PASSED                                   │
├────────────────┬────────────────┬────────────────┤
│                │    Measured    │    Threshold   │
├────────────────┼────────────────┼────────────────┤
│ Total alloc    │      963.11 MB │        1.50 GB │ <--
│ Memory Freed   │      961.68 MB │                │
│ Heap alloc     │        1.44 MB │          10 MB │
│ Objects alloc  │     20,943,174 │                │ <--
│ Obj freed      │     20,912,491 │                │
│ Objects remain │         30,683 │        200,000 │
│ Num goroutines │             11 │                │
└────────────────┴────────────────┴────────────────┘

Patched

┌──────────────────────────────────────────────────┐
│ Frouter2 API 40000 iterations:                   │
├──────────────────────────────────────────────────┤
│ Time: 2m25.913932796s     Iterations: 40000      │
│ ~API calls: 1000000                              │
│ Rate per second: 6,853                           │ <--
│ Result: PASSED                                   │
├────────────────┬────────────────┬────────────────┤
│                │    Measured    │    Threshold   │
├────────────────┼────────────────┼────────────────┤
│ Total alloc    │      840.59 MB │        1.50 GB │ <--
│ Memory Freed   │      835.62 MB │                │
│ Heap alloc     │        4.96 MB │          10 MB │
│ Objects alloc  │     19,025,199 │                │ <--
│ Obj freed      │     18,913,016 │                │
│ Objects remain │        112,183 │        200,000 │
│ Num goroutines │             13 │                │
└────────────────┴────────────────┴────────────────┘