[performance] update remaining worker pools to use queues (#2865)

* start replacing client + federator + media workers with new worker + queue types

* refactor federatingDB.Delete(), drop queued messages when deleting account / status

* move all queue purging to the processor workers

* undo toolchain updates

* code comments, ensure dereferencer worker pool gets started

* update gruf libraries in readme

* start the job scheduler separately to the worker pools

* reshuffle ordering or server.go + remove duplicate worker start / stop

* update go-list version

* fix vendoring

* move queue invalidation to before wipeing / deletion, to ensure queued work not dropped

* add logging to worker processing functions in testrig, don't start workers in unexpected places

* update go-structr to add (+then rely on) QueueCtx{} type

* ensure more worker pools get started properly in tests

* fix remaining broken tests relying on worker queue logic

* fix account test suite queue popping logic, ensure noop workers do not pull from queue

* move back accidentally shuffled account deletion order

* ensure error (non nil!!) gets passed in refactored federatingDB{}.Delete()

* silently drop deletes from accounts not permitted to

* don't warn log on forwarded deletes

* make if else clauses easier to parse

* use getFederatorMsg()

* improved code comment

* improved code comment re: requesting account delete checks

* remove boolean result from worker start / stop since false = already running or already stopped

* remove optional passed-in http.client

* remove worker starting from the admin CLI commands (we don't need to handle side-effects)

* update prune cli to start scheduler but not all of the workers

* fix rebase issues

* remove redundant return statements

* i'm sorry sir linter
This commit is contained in:
kim 2024-04-26 13:50:46 +01:00 committed by GitHub
commit c9c0773f2c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
79 changed files with 1892 additions and 834 deletions

141
internal/queue/simple.go Normal file
View file

@ -0,0 +1,141 @@
// GoToSocial
// Copyright (C) GoToSocial Authors admin@gotosocial.org
// SPDX-License-Identifier: AGPL-3.0-or-later
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU Affero General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU Affero General Public License for more details.
//
// You should have received a copy of the GNU Affero General Public License
// along with this program. If not, see <http://www.gnu.org/licenses/>.
package queue
import (
"context"
"sync"
"codeberg.org/gruf/go-list"
)
// SimpleQueue provides a simple concurrency safe
// queue using generics and a memory pool of list
// elements to reduce overall memory usage.
type SimpleQueue[T any] struct {
l list.List[T]
p []*list.Elem[T]
w chan struct{}
m sync.Mutex
}
// Push will push given value to the queue.
func (q *SimpleQueue[T]) Push(value T) {
q.m.Lock()
elem := q.alloc()
elem.Value = value
q.l.PushElemFront(elem)
if q.w != nil {
close(q.w)
q.w = nil
}
q.m.Unlock()
}
// Pop will attempt to pop value from the queue.
func (q *SimpleQueue[T]) Pop() (value T, ok bool) {
q.m.Lock()
if ok = (q.l.Tail != nil); ok {
tail := q.l.Tail
value = tail.Value
q.l.Remove(tail)
q.free(tail)
}
q.m.Unlock()
return
}
// PopCtx will attempt to pop value from queue, else blocking on context.
func (q *SimpleQueue[T]) PopCtx(ctx context.Context) (value T, ok bool) {
// Acquire lock.
q.m.Lock()
var elem *list.Elem[T]
for {
// Get next elem.
elem = q.l.Tail
if ok = (elem != nil); ok {
break
}
if q.w == nil {
// Create new wait channel.
q.w = make(chan struct{})
}
// Get current
// ch pointer.
ch := q.w
// Done with lock.
q.m.Unlock()
select {
// Context canceled.
case <-ctx.Done():
return
// Pushed!
case <-ch:
}
// Relock queue.
q.m.Lock()
}
// Extract value.
value = elem.Value
// Remove element.
q.l.Remove(elem)
q.free(elem)
// Done with lock.
q.m.Unlock()
return
}
// Len returns the current length of the queue.
func (q *SimpleQueue[T]) Len() int {
q.m.Lock()
l := q.l.Len()
q.m.Unlock()
return l
}
// alloc will allocate new list element (relying on memory pool).
func (q *SimpleQueue[T]) alloc() *list.Elem[T] {
if len(q.p) > 0 {
elem := q.p[len(q.p)-1]
q.p = q.p[:len(q.p)-1]
return elem
}
return new(list.Elem[T])
}
// free will free list element and release to pool.
func (q *SimpleQueue[T]) free(elem *list.Elem[T]) {
var zero T
elem.Next = nil
elem.Prev = nil
elem.Value = zero
q.p = append(q.p, elem)
}

View file

@ -18,7 +18,7 @@
package queue
import (
"sync/atomic"
"context"
"codeberg.org/gruf/go-structr"
)
@ -26,15 +26,14 @@ import (
// StructQueue wraps a structr.Queue{} to
// provide simple index caching by name.
type StructQueue[StructType any] struct {
queue structr.Queue[StructType]
queue structr.QueueCtx[StructType]
index map[string]*structr.Index
wait atomic.Pointer[chan struct{}]
}
// Init initializes queue with structr.QueueConfig{}.
func (q *StructQueue[T]) Init(config structr.QueueConfig[T]) {
q.index = make(map[string]*structr.Index, len(config.Indices))
q.queue = structr.Queue[T]{}
// q.queue = structr.QueueCtx[T]{}
q.queue.Init(config)
for _, cfg := range config.Indices {
q.index[cfg.Fields] = q.queue.Index(cfg.Fields)
@ -43,13 +42,22 @@ func (q *StructQueue[T]) Init(config structr.QueueConfig[T]) {
// Pop: see structr.Queue{}.PopFront().
func (q *StructQueue[T]) Pop() (value T, ok bool) {
return q.queue.PopFront()
values := q.queue.PopFrontN(1)
if ok = (len(values) > 0); !ok {
return
}
value = values[0]
return
}
// Push wraps structr.Queue{}.PushBack() to awaken those blocking on <-.Wait().
// PopCtx: see structr.QueueCtx{}.PopFront().
func (q *StructQueue[T]) PopCtx(ctx context.Context) (value T, ok bool) {
return q.queue.PopFront(ctx)
}
// Push: see structr.Queue.PushBack().
func (q *StructQueue[T]) Push(values ...T) {
q.queue.PushBack(values...)
q.broadcast()
}
// Delete pops (and drops!) all queued entries under index with key.
@ -66,31 +74,5 @@ func (q *StructQueue[T]) Len() int {
// Wait returns current wait channel, which may be
// blocked on to awaken when new value pushed to queue.
func (q *StructQueue[T]) Wait() <-chan struct{} {
var ch chan struct{}
for {
// Get channel ptr.
ptr := q.wait.Load()
if ptr != nil {
return *ptr
}
if ch == nil {
// Allocate new channel.
ch = make(chan struct{})
}
// Try set the new wait channel ptr.
if q.wait.CompareAndSwap(ptr, &ch) {
return ch
}
}
}
// broadcast safely closes wait channel if
// currently set, releasing waiting goroutines.
func (q *StructQueue[T]) broadcast() {
if ptr := q.wait.Swap(nil); ptr != nil {
close(*ptr)
}
return q.queue.Wait()
}