paint-brush
Go and sync.Poolby@kibizovd
4,181 reads
4,181 reads

Go and sync.Pool

by David KibizovSeptember 18th, 2023
Read on Terminal Reader
Read this story w/o Javascript
tldt arrow

Too Long; Didn't Read

The garbage collector (GC) does not collect garbage constantly, but at certain intervals. If your code allocates memory for some data structures and then frees them - and so on in a circle - this puts pressure on the GC, even forcing runtime to contact the OS to allocate new memory. Imagine this scenario: we select a chunk (e.g., []byte), work with it, then we release it. A certain amount of time will pass before the GC collect this chunk. If, during this time, we select another similar chunk and the memory allocated by the OS is insufficient, the application will need to request more memory from the OS. Regarding application time, the memory request from the OS can be time-consuming. Meanwhile, that previously 'used' chunk is left unused.
featured image - Go and sync.Pool
David Kibizov HackerNoon profile picture

The garbage collector (GC) does not collect garbage constantly, but at certain intervals. If your code allocates memory for some data structures and then frees them - and so on in a circle - this puts pressure on the GC, even forcing runtime to contact the OS to allocate new memory.


Imagine this scenario: we select a chunk (e.g., []byte), work with it, then we release it. A certain amount of time will pass before the GC collect this chunk. If, during this time, we select another similar chunk and the memory allocated by the OS is insufficient, the application will need to request more memory from the OS. Regarding application time, the memory request from the OS can be time-consuming. Meanwhile, that previously 'used' chunk is left unused.


What should you do? Here are your options:

  • create a pool
  • reset the state of the chunk
  • put waste chunks into the pool
  • take new chunks from the pool

Create a pool

import (
    "sync"
)

var bytesPool = sync.Pool{
    New: func() interface{} { return []byte{} },
} 

/*
In this example, the `New` function is not needed (see the explanation below). If the pool is empty,
and `New` is not `nil`, it will be used to create a new object. 
It will need to be converted from `interace{}` - converted to the required type
*/

Reset state

// let ary be a []byte of certain length and capacity
ary = ary[:0]
// truncate len, save cap

Put it into the pool

/*
Anyway, we may end up with too big chunks,
which we do not need (at least not often) - let's throw them away, 
otherwise, a 2048 byte chunk will be used where only 500-800 bytes are needed.
If the number is large, it will negatively impact memory.
This is what we aim to address.
*/

const maxCap = 1024

if cap(ary) <= maxCap {
    // putting limited-size chunks
    bytesPool.Put(ary)
}

Take from the pool

nextAry := bytesPool.Get().([]byte)

Explanation about New

The New function creates an empty []byte{}, as well as conversions to interface{} and back. In the case of []byte, we will most likely build it up using append, which basically makes this approach unprofitable because:

  • creation of []byte of zero capacity
  • double conversion to interface{} and back again
  • append will still create a new chunk
  • append can be fed nil, only of type []byte (not interface{})


It is much more convenient to create two functions that would deal with all the fuss with the pool:

// get
func getBytes() (b []byte) {
    ifc := bytesPool.Get()
    if ifc != nil {
        b = ifc.([]byte)
    }
    return
}

// put
func putBytes(b []byte) {
    if cap(b) <= maxCap {
        b = b[:0] // reset
        bytesPool.Put(b)
    }
}

Some final points to note

  • pool is goroutine-safe
  • pool will not necessarily release the data on the first GC wakeup, but it can release it at any time
  • there is no possibility to define and set the pool size
  • no need to worry about pool overflow
  • You don't need to create a pool everywhere you go; it was designed as a buffer for efficiently sharing common objects, not only within a package but also across multiple packages
  • you probably have or will have situations where the need/ability to help GC will be obvious