[venus-messager] 使用内存逐渐增加

[simlecode]

链服务模块 / Chain Service Components

• venus
• venus-auth
• venus-gateway
• venus-messager
• venus-miner
• 文档 / docs

订单服务模块 / Deal Service Components

• venus-market
• 文档 / docs

算力服务模块 / Storage Power Service Components

• venus-sector-manager
• venus-worker
• 文档 / docs

版本 / Version

venus-messager -v
venus message version v1.7.0+git.8dd0f4f

描述 / Describe the Bug

使用的内存逐渐增加，直至程序被kill

日志 / Logging Information

无

重现步骤 / Repo Steps

抓取的proof：
pprof.venus-messager.alloc_objects.alloc_space.inuse_objects.inuse_space.002.pb.gz

[simlecode]

[LinZexiao]

可以确定内存泄露发生在messager获取tipset更新本地数据的过程中,推测有可能是获取的tipset没有释放导致的,审阅相关代码,目前还不能确定具体位置

[diwufeiwen]

大概率是与MessageState中的messageCache *cache.Cache有关，master分支删了这个，就没有一直增长的现象了

[diwufeiwen]

type cache struct {
defaultExpiration time.Duration
items map[string]Item
mu sync.RWMutex
onEvicted func(string, interface{})
janitor *janitor
}
type Item struct {
Object interface{}
Expiration int64
}

这里的： items map[string]Item，value是struct，在delete时并不会回收内存。见下面示例：

package main

import (
	"log"
	"runtime"
)

type Test struct {
	Name interface{}
	Sex  bool
}

var lastTotalFreed uint64
var testMap map[int]Test
var cnt = 100000

func main() {
	printMemStats()

	initMap()
	runtime.GC()
	printMemStats()
	log.Println(len(testMap))

	for i := 0; i < cnt; i++ {
		delete(testMap, i)
	}
	runtime.GC()
	printMemStats()
        log.Println(len(testMap))

	testMap = nil
	runtime.GC()
	printMemStats()
}

func initMap() {
	testMap = make(map[int]Test, cnt)

	for i := 0; i < cnt; i++ {
		testMap[i] = Test{
			Name: "name",
		}
	}
}

func printMemStats() {
	var m runtime.MemStats
	runtime.ReadMemStats(&m)
	log.Printf("Alloc = %v TotalAlloc = %v  Just Freed = %v Sys = %v NumGC = %v\n",
		m.Alloc/1024, m.TotalAlloc/1024, ((m.TotalAlloc-m.Alloc)-lastTotalFreed)/1024, m.Sys/1024, m.NumGC)

	lastTotalFreed = m.TotalAlloc - m.Alloc
}

go run result:
2022/10/13 11:51:50 Alloc = 146 TotalAlloc = 146 Just Freed = 0 Sys = 6548 NumGC = 0
2022/10/13 11:51:51 Alloc = 5244 TotalAlloc = 5305 Just Freed = 61 Sys = 15644 NumGC = 2
2022/10/13 11:51:51 100000
2022/10/13 11:51:51 Alloc = 5245 TotalAlloc = 5306 Just Freed = 0 Sys = 15644 NumGC = 3
2022/10/13 11:51:51 0
2022/10/13 11:51:51 Alloc = 143 TotalAlloc = 5308 Just Freed = 5103 Sys = 15644 NumGC = 4

[diwufeiwen]

删除map要想马上被gc内存，需要用替换的方式，创建个新的map，把保留元素加进去，然后赋值给map变量

[diwufeiwen]

据运维反馈，最近master已没有了