GoToSocial/vendor/codeberg.org/gruf/go-runners/pool.go
kim acc95923da
[performance] processing media and scheduled jobs improvements (#1482)
* replace media workers with just runners.WorkerPool, move to state structure, use go-sched for global task scheduling

* improved code comment

* fix worker tryUntil function, update go-runners/go-sched

* make preprocess functions package public, use these where possible to stop doubled up processing

* remove separate emoji worker pool

* limit calls to time.Now() during media preprocessing

* use Processor{} to manage singular runtime of processing media

* ensure workers get started when media manager is used

* improved error setting in processing media, fix media test

* port changes from processingmedia to processing emoji

* finish code commenting

* finish code commenting and comment-out client API + federator worker pools until concurrency worker pools replaced

* linterrrrrrrrrrrrrrrr

---------

Signed-off-by: kim <grufwub@gmail.com>
2023-02-13 18:40:48 +00:00

263 lines
6.0 KiB
Go

package runners
import (
"context"
"fmt"
"os"
"runtime"
"sync"
"codeberg.org/gruf/go-errors/v2"
)
// WorkerFunc represents a function processable by a worker in WorkerPool. Note
// that implementations absolutely MUST check whether passed context is <-ctx.Done()
// otherwise stopping the pool may block indefinitely.
type WorkerFunc func(context.Context)
// WorkerPool provides a means of enqueuing asynchronous work.
type WorkerPool struct {
fns chan WorkerFunc
svc Service
}
// Start will start the main WorkerPool management loop in a new goroutine, along
// with requested number of child worker goroutines. Returns false if already running.
func (pool *WorkerPool) Start(workers int, queue int) bool {
// Attempt to start the svc
ctx, ok := pool.svc.doStart()
if !ok {
return false
}
if workers <= 0 {
// Use $GOMAXPROCS as default.
workers = runtime.GOMAXPROCS(0)
}
if queue < 0 {
// Use reasonable queue default.
queue = workers * 10
}
// Allocate pool queue of given size.
//
// This MUST be set BEFORE we return and NOT in
// the launched goroutine, or there is a risk that
// the pool may appear as closed for a short time
// until the main goroutine has been entered.
fns := make(chan WorkerFunc, queue)
pool.fns = fns
go func() {
defer func() {
// unlock single wait
pool.svc.wait.Unlock()
// ensure stopped
pool.svc.Stop()
}()
var wait sync.WaitGroup
// Start goroutine worker functions
for i := 0; i < workers; i++ {
wait.Add(1)
go func() {
defer wait.Done()
// Run worker function (retry on panic)
for !worker_run(CancelCtx(ctx), fns) {
}
}()
}
// Wait on ctx
<-ctx
// Drain function queue.
//
// All functions in the queue MUST be
// run, so we pass them a closed context.
//
// This mainly allows us to block until
// the function queue is empty, as worker
// functions will also continue draining in
// the background with the (now) closed ctx.
for !drain_queue(fns) {
// retry on panic
}
// Now the queue is empty, we can
// safely close the channel signalling
// all of the workers to return.
close(fns)
wait.Wait()
}()
return true
}
// Stop will stop the WorkerPool management loop, blocking until stopped.
func (pool *WorkerPool) Stop() bool {
return pool.svc.Stop()
}
// Running returns if WorkerPool management loop is running (i.e. NOT stopped / stopping).
func (pool *WorkerPool) Running() bool {
return pool.svc.Running()
}
// Done returns a channel that's closed when WorkerPool.Stop() is called. It is the same channel provided to the currently running worker functions.
func (pool *WorkerPool) Done() <-chan struct{} {
return pool.svc.Done()
}
// Enqueue will add provided WorkerFunc to the queue to be performed when there is a free worker.
// This will block until function is queued or pool is stopped. In all cases, the WorkerFunc will be
// executed, with the state of the pool being indicated by <-ctx.Done() of the passed ctx.
// WorkerFuncs MUST respect the passed context.
func (pool *WorkerPool) Enqueue(fn WorkerFunc) {
// Check valid fn
if fn == nil {
return
}
select {
// Pool ctx cancelled
case <-pool.svc.Done():
fn(closedctx)
// Placed fn in queue
case pool.fns <- fn:
}
}
// EnqueueCtx is functionally identical to WorkerPool.Enqueue() but returns early in the
// case that caller provided <-ctx.Done() is closed, WITHOUT running the WorkerFunc.
func (pool *WorkerPool) EnqueueCtx(ctx context.Context, fn WorkerFunc) bool {
// Check valid fn
if fn == nil {
return false
}
select {
// Caller ctx cancelled
case <-ctx.Done():
return false
// Pool ctx cancelled
case <-pool.svc.Done():
return false
// Placed fn in queue
case pool.fns <- fn:
return true
}
}
// MustEnqueueCtx functionally performs similarly to WorkerPool.EnqueueCtx(), but in the case
// that the provided <-ctx.Done() is closed, it is passed asynchronously to WorkerPool.Enqueue().
// Return boolean indicates whether function was executed in time before <-ctx.Done() is closed.
func (pool *WorkerPool) MustEnqueueCtx(ctx context.Context, fn WorkerFunc) (ok bool) {
// Check valid fn
if fn == nil {
return false
}
select {
case <-ctx.Done():
// We failed to add this entry to the worker queue before the
// incoming context was cancelled. So to ensure processing
// we simply queue it asynchronously and return early to caller.
go pool.Enqueue(fn)
return false
case <-pool.svc.Done():
// Pool ctx cancelled
fn(closedctx)
return false
case pool.fns <- fn:
// Placed fn in queue
return true
}
}
// EnqueueNow attempts Enqueue but returns false if not executed.
func (pool *WorkerPool) EnqueueNow(fn WorkerFunc) bool {
// Check valid fn
if fn == nil {
return false
}
select {
// Pool ctx cancelled
case <-pool.svc.Done():
return false
// Placed fn in queue
case pool.fns <- fn:
return true
// Queue is full
default:
return false
}
}
// Queue returns the number of currently queued WorkerFuncs.
func (pool *WorkerPool) Queue() int {
var l int
pool.svc.While(func() {
l = len(pool.fns)
})
return l
}
// worker_run is the main worker routine, accepting functions from 'fns' until it is closed.
func worker_run(ctx context.Context, fns <-chan WorkerFunc) bool {
defer func() {
// Recover and drop any panic
if r := recover(); r != nil {
const msg = "worker_run: recovered panic: %v\n\n%s\n"
fmt.Fprintf(os.Stderr, msg, r, errors.GetCallers(2, 10))
}
}()
for {
// Wait on next func
fn, ok := <-fns
if !ok {
return true
}
// Run with ctx
fn(ctx)
}
}
// drain_queue will drain and run all functions in worker queue, passing in a closed context.
func drain_queue(fns <-chan WorkerFunc) bool {
defer func() {
// Recover and drop any panic
if r := recover(); r != nil {
const msg = "drain_queue: recovered panic: %v\n\n%s\n"
fmt.Fprintf(os.Stderr, msg, r, errors.GetCallers(2, 10))
}
}()
for {
select {
// Run with closed ctx
case fn := <-fns:
fn(closedctx)
// Queue is empty
default:
return true
}
}
}