mirror of
https://gitee.com/milvus-io/milvus.git
synced 2025-12-07 17:48:29 +08:00
Fix CDC OOM by: 1. free msg buffer manually. 2. limit max msg buffer size. 3. reduce scanner msg hander buffer size. issue: https://github.com/milvus-io/milvus/issues/44123 Signed-off-by: bigsheeper <yihao.dai@zilliz.com>
228 lines
6.8 KiB
Go
228 lines
6.8 KiB
Go
// Licensed to the LF AI & Data foundation under one
|
|
// or more contributor license agreements. See the NOTICE file
|
|
// distributed with this work for additional information
|
|
// regarding copyright ownership. The ASF licenses this file
|
|
// to you under the Apache License, Version 2.0 (the
|
|
// "License"); you may not use this file except in compliance
|
|
// with the License. You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
|
|
package replicatestream
|
|
|
|
import (
|
|
"context"
|
|
"sync"
|
|
|
|
message "github.com/milvus-io/milvus/pkg/v2/streaming/util/message"
|
|
)
|
|
|
|
// MsgQueue exposes the required operations. We include context support to meet the
|
|
// blocking + cancel requirements.
|
|
// The names mirror the user's suggestion with small adjustments for clarity.
|
|
type MsgQueue interface {
|
|
// Enqueue appends a message. Blocks if capacity is full until space appears
|
|
// (via CleanupConfirmedMessages) or ctx is canceled.
|
|
Enqueue(ctx context.Context, msg message.ImmutableMessage) error
|
|
|
|
// ReadNext returns the next message from the current read cursor and advances
|
|
// the cursor by one. It does NOT delete the message from the queue storage.
|
|
// Blocks when there are no readable messages (i.e., cursor is at tail) until
|
|
// a new message is Enqueued or ctx is canceled.
|
|
ReadNext(ctx context.Context) (message.ImmutableMessage, error)
|
|
|
|
// SeekToHead moves the read cursor to the first not-yet-deleted message.
|
|
SeekToHead()
|
|
|
|
// CleanupConfirmedMessages permanently removes all messages whose Timetick()
|
|
// <= lastConfirmedTimeTick. This may free capacity and wake Enqueue waiters.
|
|
// Returns the messages that were cleaned up.
|
|
CleanupConfirmedMessages(lastConfirmedTimeTick uint64) []message.ImmutableMessage
|
|
|
|
// Len returns the number of stored (not yet deleted) messages.
|
|
Len() int
|
|
|
|
// Cap returns the maximum capacity of the queue.
|
|
Cap() int
|
|
}
|
|
|
|
var _ MsgQueue = (*msgQueue)(nil)
|
|
|
|
// msgQueue is a bounded, in-memory implementation of MsgQueue.
|
|
// It uses a simple slice for storage and an integer read cursor that is always
|
|
// within [0, len(buf)].
|
|
type msgQueue struct {
|
|
mu sync.Mutex
|
|
notEmpty *sync.Cond
|
|
notFull *sync.Cond
|
|
|
|
buf []message.ImmutableMessage
|
|
bufBytes int
|
|
readIdx int
|
|
cap int
|
|
maxSize int
|
|
closed bool
|
|
}
|
|
|
|
type MsgQueueOptions struct {
|
|
Capacity int
|
|
MaxSize int
|
|
}
|
|
|
|
// NewMsgQueue creates a queue with a fixed capacity (>0).
|
|
func NewMsgQueue(options MsgQueueOptions) *msgQueue {
|
|
if options.Capacity <= 0 {
|
|
panic("capacity must be > 0")
|
|
}
|
|
if options.MaxSize <= 0 {
|
|
panic("max size must be > 0")
|
|
}
|
|
q := &msgQueue{cap: options.Capacity, maxSize: options.MaxSize}
|
|
q.notEmpty = sync.NewCond(&q.mu)
|
|
q.notFull = sync.NewCond(&q.mu)
|
|
return q
|
|
}
|
|
|
|
func (q *msgQueue) Len() int {
|
|
q.mu.Lock()
|
|
defer q.mu.Unlock()
|
|
return len(q.buf)
|
|
}
|
|
|
|
func (q *msgQueue) Cap() int { return q.cap }
|
|
|
|
// Enqueue implements a blocking producer. It respects ctx cancellation.
|
|
func (q *msgQueue) Enqueue(ctx context.Context, msg message.ImmutableMessage) error {
|
|
q.mu.Lock()
|
|
defer q.mu.Unlock()
|
|
|
|
for len(q.buf) >= q.cap || q.bufBytes >= q.maxSize {
|
|
if ctx.Err() != nil {
|
|
return context.Canceled
|
|
}
|
|
q.waitWithContext(ctx, q.notFull)
|
|
if ctx.Err() != nil {
|
|
return context.Canceled
|
|
}
|
|
}
|
|
|
|
// Optional runtime check: enforce non-decreasing timetick
|
|
// if n := len(q.buf); n > 0 {
|
|
// if q.buf[n-1].Timetick() > msg.Timetick() {
|
|
// return fmt.Errorf("enqueue timetick order violation: last=%d new=%d", q.buf[n-1].Timetick(), msg.Timetick())
|
|
// }
|
|
// }
|
|
|
|
q.buf = append(q.buf, msg)
|
|
q.bufBytes += msg.EstimateSize()
|
|
|
|
// New data is available for readers.
|
|
q.notEmpty.Signal()
|
|
return nil
|
|
}
|
|
|
|
// ReadNext returns the next message at the read cursor. Does not delete it.
|
|
func (q *msgQueue) ReadNext(ctx context.Context) (message.ImmutableMessage, error) {
|
|
q.mu.Lock()
|
|
defer q.mu.Unlock()
|
|
|
|
for q.readIdx >= len(q.buf) { // no readable messages
|
|
if ctx.Err() != nil {
|
|
return nil, context.Canceled
|
|
}
|
|
q.waitWithContext(ctx, q.notEmpty)
|
|
if ctx.Err() != nil {
|
|
return nil, context.Canceled
|
|
}
|
|
}
|
|
|
|
m := q.buf[q.readIdx]
|
|
q.readIdx++ // advance read cursor only; storage remains intact
|
|
|
|
// Readers advancing may not directly free capacity, but producers may still
|
|
// be waiting due to full buffer; signal in case cleanup made space earlier.
|
|
q.notFull.Signal()
|
|
return m, nil
|
|
}
|
|
|
|
// SeekToHead resets the read cursor to the first existing element.
|
|
func (q *msgQueue) SeekToHead() {
|
|
q.mu.Lock()
|
|
q.readIdx = 0
|
|
// Let potential consumers know there's readable data (if any exists).
|
|
if len(q.buf) > 0 {
|
|
q.notEmpty.Broadcast()
|
|
}
|
|
q.mu.Unlock()
|
|
}
|
|
|
|
// CleanupConfirmedMessages permanently drops messages with Timetick <= watermark.
|
|
// This frees capacity and may move the read cursor backward proportionally to the
|
|
// number of deleted messages (but clamped at 0).
|
|
// Returns the messages that were cleaned up.
|
|
func (q *msgQueue) CleanupConfirmedMessages(lastConfirmedTimeTick uint64) []message.ImmutableMessage {
|
|
q.mu.Lock()
|
|
defer q.mu.Unlock()
|
|
|
|
if len(q.buf) == 0 {
|
|
return nil
|
|
}
|
|
|
|
// Find first index whose Timetick() > lastConfirmedTimeTick
|
|
cut := 0
|
|
for cut < len(q.buf) && q.buf[cut].TimeTick() <= lastConfirmedTimeTick {
|
|
cut++
|
|
}
|
|
|
|
var cleanedMessages []message.ImmutableMessage
|
|
if cut > 0 {
|
|
// Collect the messages that will be cleaned up
|
|
cleanedMessages = make([]message.ImmutableMessage, cut)
|
|
copy(cleanedMessages, q.buf[:cut])
|
|
|
|
// Drop the prefix [0:cut)
|
|
// Release memory of [0:cut) by zeroing references before reslicing
|
|
for i := 0; i < cut; i++ {
|
|
q.bufBytes -= q.buf[i].EstimateSize()
|
|
q.buf[i] = nil
|
|
}
|
|
q.buf = q.buf[cut:]
|
|
// Adjust read cursor relative to the new slice
|
|
q.readIdx -= cut
|
|
if q.readIdx < 0 {
|
|
q.readIdx = 0
|
|
}
|
|
// Free space became available; wake blocked producers.
|
|
q.notFull.Broadcast()
|
|
}
|
|
|
|
return cleanedMessages
|
|
}
|
|
|
|
// waitWithContext waits on cond while allowing ctx cancellation to wake it up.
|
|
// Implementation detail:
|
|
// - cond.Wait() requires holding q.mu; Wait atomically unlocks and re-locks.
|
|
// - We spawn a short-lived goroutine that Broadcasts on ctx.Done(). This is
|
|
// safe and bounded: after Wait returns, we close a local channel to let the
|
|
// goroutine exit immediately (even if ctx wasn't canceled).
|
|
func (q *msgQueue) waitWithContext(ctx context.Context, cond *sync.Cond) {
|
|
done := make(chan struct{})
|
|
go func() {
|
|
select {
|
|
case <-ctx.Done():
|
|
q.mu.Lock()
|
|
cond.Broadcast()
|
|
q.mu.Unlock()
|
|
case <-done:
|
|
}
|
|
}()
|
|
cond.Wait() // releases q.mu while waiting; re-locks before returning
|
|
close(done)
|
|
}
|