Redis偶發(fā)連接失敗案例實(shí)戰(zhàn)記錄
前言
本文主要給大家介紹了關(guān)于Redis偶發(fā)連接失敗的相關(guān)內(nèi)容,分享出來(lái)供大家參考學(xué)習(xí),下面話不多說(shuō)了,來(lái)一起看看詳細(xì)的介紹吧
【作者】
張延?。簲y程技術(shù)保障中心資深DBA,對(duì)數(shù)據(jù)庫(kù)架構(gòu)和疑難問(wèn)題分析排查有濃厚的興趣。
壽向晨:攜程技術(shù)保障中心高級(jí)DBA,主要負(fù)責(zé)攜程Redis及DB的運(yùn)維工作,在自動(dòng)化運(yùn)維,流程化及監(jiān)控排障等方面有較多的實(shí)踐經(jīng)驗(yàn),喜歡深入分析問(wèn)題,提高團(tuán)隊(duì)運(yùn)維效率。
【問(wèn)題描述】
生產(chǎn)環(huán)境有一個(gè)Redis會(huì)偶爾發(fā)生連接失敗的報(bào)錯(cuò),報(bào)錯(cuò)的時(shí)間點(diǎn)、客戶端IP并沒(méi)有特別明顯的規(guī)律,過(guò)一會(huì)兒,報(bào)錯(cuò)會(huì)自動(dòng)恢復(fù)。
以下是客戶端報(bào)錯(cuò)信息:
CRedis.Client.RExceptions.ExcuteCommandException: Unable to Connect redis server: ---> CRedis.Third.Redis.RedisException: Unable to Connect redis server: 在 CRedis.Third.Redis.RedisNativeClient.CreateConnectionError() 在 CRedis.Third.Redis.RedisNativeClient.SendExpectData(Byte[][] cmdWithBinaryArgs) 在 CRedis.Client.Entities.RedisServer.<>c__DisplayClassd`1.
從報(bào)錯(cuò)的信息來(lái)看,應(yīng)該是連接不上Redis所致。Redis的版本是2.8.19。雖然版本有點(diǎn)老,但基本運(yùn)行穩(wěn)定。
線上環(huán)境只有這個(gè)集群有偶爾報(bào)錯(cuò)。這個(gè)集群的一個(gè)比較明顯的特征是客戶端服務(wù)器比較多,有上百臺(tái)。
【問(wèn)題分析】
從報(bào)錯(cuò)的信息來(lái)看,客戶端連接不到服務(wù)端。常見(jiàn)的原因有以下幾點(diǎn):
- 一個(gè)常見(jiàn)的原因是由于端口耗盡,對(duì)網(wǎng)絡(luò)連接進(jìn)行排查,在出問(wèn)題的點(diǎn)上,TCP連接數(shù)遠(yuǎn)沒(méi)有達(dá)到端口耗盡的場(chǎng)景,因此這個(gè)不是Redis連接不上的根本原因。
- 另外一種常見(jiàn)的場(chǎng)景是在服務(wù)端有慢查詢,導(dǎo)致Redis服務(wù)阻塞。我們?cè)赗edis服務(wù)端,把運(yùn)行超過(guò)10毫秒的語(yǔ)句進(jìn)行抓取,也沒(méi)有抓到運(yùn)行慢的語(yǔ)句。
從服務(wù)端的部署的監(jiān)控來(lái)看,出問(wèn)題的點(diǎn)上,連接數(shù)有一個(gè)突然飆升,從3500個(gè)連接突然飆升至4100個(gè)連接。如下圖顯示:

同時(shí)間,服務(wù)器端顯示Redis服務(wù)端有丟包現(xiàn)象:345539 – 344683 = 856個(gè)包。
Sat Apr 7 10:41:40 CST 2018 1699 outgoing packets dropped 92 dropped because of missing route 344683 SYNs to LISTEN sockets dropped 344683 times the listen queue of a socket overflowed
Sat Apr 7 10:41:41 CST 2018 1699 outgoing packets dropped 92 dropped because of missing route 345539 SYNs to LISTEN sockets dropped 345539 times the listen queue of a socket overflowed
客戶端報(bào)錯(cuò)的原因基本確定,是因?yàn)榻ㄟB速度太快,導(dǎo)致服務(wù)端backlog隊(duì)列溢出,連接被server端reset。
【關(guān)于backlog overflow】
在高并發(fā)的短連接服務(wù)中,這是一種很常見(jiàn)的tcp報(bào)錯(cuò)類型。一個(gè)正常的tcp建連過(guò)程如下:
1.client發(fā)送一個(gè)(SYN)給server
2.server返回一個(gè)(SYN,ACK)給client
3.client返回一個(gè)(ACK)
三次握手結(jié)束,對(duì)client來(lái)說(shuō)建連成功,client可以繼續(xù)發(fā)送數(shù)據(jù)包給server,但是這個(gè)時(shí)候server端未必ready,如下圖所示 :

在BSD版本內(nèi)核實(shí)現(xiàn)的tcp協(xié)議中,server端建連過(guò)程需要兩個(gè)隊(duì)列,一個(gè)是SYN queue,一個(gè)是accept queue。前者叫半開(kāi)連接(或者半連接)隊(duì)列,在接收到client發(fā)送的SYN時(shí)加入隊(duì)列。(一種常見(jiàn)的網(wǎng)絡(luò)攻擊方式就是不斷發(fā)送SYN但是不發(fā)送ACK從而導(dǎo)致server端的半開(kāi)隊(duì)列撐爆,server端拒絕服務(wù)。)后者叫全連接隊(duì)列,server返回(SYN,ACK),在接收到client發(fā)送ACK后(此時(shí)client會(huì)認(rèn)為建連已經(jīng)完成,會(huì)開(kāi)始發(fā)送PSH包),如果accept queue沒(méi)有滿,那么server從SYN queue把連接信息移到accept queue;如果此時(shí)accept queue溢出的話,server的行為要看配置。如果tcp_abort_on_overflow為0(默認(rèn)),那么直接drop掉client發(fā)送的PSH包,此時(shí)client會(huì)進(jìn)入重發(fā)過(guò)程,一段時(shí)間后server端重新發(fā)送SYN,ACK,重新從建連的第二步開(kāi)始;如果tcp_abort_on_overflow為1,那么server端發(fā)現(xiàn)accept queue滿之后直接發(fā)送reset。
通過(guò)wireshark搜索發(fā)現(xiàn)在一秒內(nèi)有超過(guò)2000次對(duì)Redis Server端發(fā)起建連請(qǐng)求。我們嘗試修改tcp backlog大小,從511調(diào)整到2048, 問(wèn)題并沒(méi)有得到解決。所以此類微調(diào),并不能徹底的解決問(wèn)題。
【網(wǎng)絡(luò)包分析】
我們用wireshark來(lái)識(shí)別網(wǎng)絡(luò)擁塞的準(zhǔn)確時(shí)間點(diǎn)和原因。我們已經(jīng)有了準(zhǔn)確的報(bào)錯(cuò)時(shí)間點(diǎn),先用editcap把超大的tcp包裁剪一下,裁成30秒間隔,并通過(guò)wireshark I/O 100ms間隔分析網(wǎng)絡(luò)阻塞的準(zhǔn)確時(shí)間點(diǎn):

根據(jù)圖標(biāo)可以明顯看到tcp的packets來(lái)往存在block。
對(duì)該block前后的網(wǎng)絡(luò)包進(jìn)行明細(xì)分析,網(wǎng)絡(luò)包來(lái)往情況如下:
| Time | Source | Dest | Description |
|---|---|---|---|
| 12:01:54.6536050 | Redis-Server | Clients | TCP:Flags=…AP… |
| 12:01:54.6538580 | Redis-Server | Clients | TCP:Flags=…AP… |
| 12:01:54.6539770 | Redis-Server | Clients | TCP:Flags=…AP… |
| 12:01:54.6720580 | Redis-Server | Clients | TCP:Flags=…A..S.. |
| 12:01:54.6727200 | Redis-Server | Clients | TCP:Flags=…A…… |
| 12:01:54.6808480 | Redis-Server | Clients | TCP:Flags=…AP….. |
| 12:01:54.6910840 | Redis-Server | Clients | TCP:Flags=…A…S., |
| 12:01:54.6911950 | Redis-Server | Clients | TCP:Flags=…A…… |
| … | … | … | … |
| 12:01:56.1181350 | Redis-Server | Clients | TCP:Flags=…AP…. |
12:01:54.6808480, Redis Server端向客戶端發(fā)送了一個(gè)Push包,也就是對(duì)于查詢請(qǐng)求的一個(gè)結(jié)果返回。后面的包都是在做連接處理,包括Ack包,Ack確認(rèn)包,以及重置的RST包,緊接著下面一個(gè)Push包是在12:01:56.1181350發(fā)出的。中間的間隔是1.4372870秒。也就是說(shuō),在這1.4372870秒期間,Redis的服務(wù)器端,除了做一個(gè)查詢,其他的操作都是在做建連,或拒絕連接。
客戶端報(bào)錯(cuò)的前后邏輯已經(jīng)清楚了,redis-server卡了1.43秒,client的connection pool被打滿,瘋狂新建連接,server的accept queue滿,直接拒絕服務(wù),client報(bào)錯(cuò)。開(kāi)始懷疑client發(fā)送了特殊命令,這時(shí)需要確認(rèn)一下client的最后幾個(gè)命令是什么,找到redis-server卡死前的第一個(gè)包,裝一個(gè)wireshark的redis插件,看到最后幾個(gè)命令是簡(jiǎn)單的get,并且key-value都很小,不至于需要耗費(fèi)1.43秒才能完成。服務(wù)端也沒(méi)有slow log,此時(shí)排障再次陷入僵局。
【進(jìn)一步分析】
為了了解這1.43秒之內(nèi),Redis Server在做什么事情,我們用pstack來(lái)抓取信息。Pstack本質(zhì)上是gdb attach. 高頻率的抓取會(huì)影響redis的吞吐。死循環(huán)0.5秒一次無(wú)腦抓,在redis-server卡死的時(shí)候抓到堆棧如下(過(guò)濾了沒(méi)用的棧信息):
Thu May 31 11:29:18 CST 2018
Thread 1 (Thread 0x7ff2db6de720 (LWP 8378)):
#0 0x000000000048cec4 in ?? ()
#1 0x00000000004914a4 in je_arena_ralloc ()
#2 0x00000000004836a1 in je_realloc ()
#3 0x0000000000422cc5 in zrealloc ()
#4 0x00000000004213d7 in sdsRemoveFreeSpace ()
#5 0x000000000041ef3c in clientsCronResizeQueryBuffer ()
#6 0x00000000004205de in clientsCron ()
#7 0x0000000000420784 in serverCron ()
#8 0x0000000000418542 in aeProcessEvents ()
#9 0x000000000041873b in aeMain ()
#10 0x0000000000420fce in main ()
Thu May 31 11:29:19 CST 2018
Thread 1 (Thread 0x7ff2db6de720 (LWP 8378)):
#0 0x0000003729ee5407 in madvise () from /lib64/libc.so.6
#1 0x0000000000493a4e in je_pages_purge ()
#2 0x000000000048cf70 in ?? ()
#3 0x00000000004914a4 in je_arena_ralloc ()
#4 0x00000000004836a1 in je_realloc ()
#5 0x0000000000422cc5 in zrealloc ()
#6 0x00000000004213d7 in sdsRemoveFreeSpace ()
#7 0x000000000041ef3c in clientsCronResizeQueryBuffer ()
#8 0x00000000004205de in clientsCron ()
#9 0x0000000000420784 in serverCron ()
#10 0x0000000000418542 in aeProcessEvents ()
#11 0x000000000041873b in aeMain ()
#12 0x0000000000420fce in main ()
Thu May 31 11:29:19 CST 2018
Thread 1 (Thread 0x7ff2db6de720 (LWP 8378)):
#0 0x000000000048108c in je_malloc_usable_size ()
#1 0x0000000000422be6 in zmalloc ()
#2 0x00000000004220bc in sdsnewlen ()
#3 0x000000000042c409 in createStringObject ()
#4 0x000000000042918e in processMultibulkBuffer ()
#5 0x0000000000429662 in processInputBuffer ()
#6 0x0000000000429762 in readQueryFromClient ()
#7 0x000000000041847c in aeProcessEvents ()
#8 0x000000000041873b in aeMain ()
#9 0x0000000000420fce in main ()
Thu May 31 11:29:20 CST 2018
Thread 1 (Thread 0x7ff2db6de720 (LWP 8378)):
#0 0x000000372a60e7cd in write () from /lib64/libpthread.so.0
#1 0x0000000000428833 in sendReplyToClient ()
#2 0x0000000000418435 in aeProcessEvents ()
#3 0x000000000041873b in aeMain ()
#4 0x0000000000420fce in main ()
重復(fù)多次抓取后,從堆棧中發(fā)現(xiàn)可疑堆棧clientsCronResizeQueryBuffer位置,屬于serverCron()函數(shù)下,這個(gè)redis-server內(nèi)部的定時(shí)調(diào)度,并不在用戶線程下,這個(gè)解釋了為什么卡死的時(shí)候沒(méi)有出現(xiàn)慢查詢。
查看redis源碼,確認(rèn)到底redis-server在做什么:
clientsCron(server.h):
#define CLIENTS_CRON_MIN_ITERATIONS 5
void clientsCron(void) {
/* Make sure to process at least numclients/server.hz of clients
* per call. Since this function is called server.hz times per second
* we are sure that in the worst case we process all the clients in 1
* second. */
int numclients = listLength(server.clients);
int iterations = numclients/server.hz;
mstime_t now = mstime();
/* Process at least a few clients while we are at it, even if we need
* to process less than CLIENTS_CRON_MIN_ITERATIONS to meet our contract
* of processing each client once per second. */
if (iterations < CLIENTS_CRON_MIN_ITERATIONS)
iterations = (numclients < CLIENTS_CRON_MIN_ITERATIONS) ?
numclients : CLIENTS_CRON_MIN_ITERATIONS;
while(listLength(server.clients) && iterations--) {
client *c;
listNode *head;
/* Rotate the list, take the current head, process.
* This way if the client must be removed from the list it's the
* first element and we don't incur into O(N) computation. */
listRotate(server.clients);
head = listFirst(server.clients);
c = listNodeValue(head);
/* The following functions do different service checks on the client.
* The protocol is that they return non-zero if the client was
* terminated. */
if (clientsCronHandleTimeout(c,now)) continue;
if (clientsCronResizeQueryBuffer(c)) continue;
}
}
clientsCron首先判斷當(dāng)前client的數(shù)量,用于控制一次清理連接的數(shù)量,生產(chǎn)服務(wù)器單實(shí)例的連接數(shù)量在5000不到,也就是一次清理的連接數(shù)是50個(gè)。
clientsCronResizeQueryBuffer(server.h):
/* The client query buffer is an sds.c string that can end with a lot of
* free space not used, this function reclaims space if needed.
*
* The function always returns 0 as it never terminates the client. */
int clientsCronResizeQueryBuffer(client *c) {
size_t querybuf_size = sdsAllocSize(c->querybuf);
time_t idletime = server.unixtime - c->lastinteraction;
/* 只在以下兩種情況下會(huì)Resize query buffer:
* 1) Query buffer > BIG_ARG(在server.h 中定義#define PROTO_MBULK_BIG_ARG (1024*32))
且這個(gè)Buffer的小于一段時(shí)間的客戶端使用的峰值.
* 2) 客戶端空閑超過(guò)2s且Buffer size大于1k. */
if (((querybuf_size > PROTO_MBULK_BIG_ARG) &&
(querybuf_size/(c->querybuf_peak+1)) > 2) ||
(querybuf_size > 1024 && idletime > 2))
{
/* Only resize the query buffer if it is actually wasting space. */
if (sdsavail(c->querybuf) > 1024) {
c->querybuf = sdsRemoveFreeSpace(c->querybuf);
}
}
/* Reset the peak again to capture the peak memory usage in the next
* cycle. */
c->querybuf_peak = 0;
return 0;
}
如果redisClient對(duì)象的query buffer滿足條件,那么就直接resize掉。滿足條件的連接分成兩種,一種是真的很大的,比該客戶端一段時(shí)間內(nèi)使用的峰值還大;還有一種是很閑(idle>2)的,這兩種都要滿足一個(gè)條件,就是buffer free的部分超過(guò)1k。那么redis-server卡住的原因就是正好有那么50個(gè)很大的或者空閑的并且free size超過(guò)了1k大小連接的同時(shí)循環(huán)做了resize,由于redis都屬于單線程工作的程序,所以block了client。那么解決這個(gè)問(wèn)題辦法就很明朗了,讓resize 的頻率變低或者resize的執(zhí)行速度變快。
既然問(wèn)題出在query buffer上,我們先看一下這個(gè)東西被修改的位置:
readQueryFromClient(networking.c):
redisClient *createClient(int fd) {
redisClient *c = zmalloc(sizeof(redisClient));
/* passing -1 as fd it is possible to create a non connected client.
* This is useful since all the Redis commands needs to be executed
* in the context of a client. When commands are executed in other
* contexts (for instance a Lua script) we need a non connected client. */
if (fd != -1) {
anetNonBlock(NULL,fd);
anetEnableTcpNoDelay(NULL,fd);
if (server.tcpkeepalive)
anetKeepAlive(NULL,fd,server.tcpkeepalive);
if (aeCreateFileEvent(server.el,fd,AE_READABLE,
readQueryFromClient, c) == AE_ERR)
{
close(fd);
zfree(c);
return NULL;
}
}
selectDb(c,0);
c->id = server.next_client_id++;
c->fd = fd;
c->name = NULL;
c->bufpos = 0;
c->querybuf = sdsempty(); 初始化是0
readQueryFromClient(networking.c):
void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mask) {
redisClient *c = (redisClient*) privdata;
int nread, readlen;
size_t qblen;
REDIS_NOTUSED(el);
REDIS_NOTUSED(mask);
server.current_client = c;
readlen = REDIS_IOBUF_LEN;
/* If this is a multi bulk request, and we are processing a bulk reply
* that is large enough, try to maximize the probability that the query
* buffer contains exactly the SDS string representing the object, even
* at the risk of requiring more read(2) calls. This way the function
* processMultiBulkBuffer() can avoid copying buffers to create the
* Redis Object representing the argument. */
if (c->reqtype == REDIS_REQ_MULTIBULK && c->multibulklen && c->bulklen != -1
&& c->bulklen >= REDIS_MBULK_BIG_ARG)
{
int remaining = (unsigned)(c->bulklen+2)-sdslen(c->querybuf);
if (remaining < readlen) readlen = remaining;
}
qblen = sdslen(c->querybuf);
if (c->querybuf_peak < qblen) c->querybuf_peak = qblen;
c->querybuf = sdsMakeRoomFor(c->querybuf, readlen); 在這里會(huì)被擴(kuò)大
由此可見(jiàn)c->querybuf在連接第一次讀取命令后的大小就會(huì)被分配至少1024*32,所以回過(guò)頭再去看resize的清理邏輯就明顯存在問(wèn)題,每個(gè)被使用到的query buffer的大小至少就是1024*32,但是清理的時(shí)候判斷條件是>1024,也就是說(shuō),所有的idle>2的被使用過(guò)的連接都會(huì)被resize掉,下次接收到請(qǐng)求的時(shí)候再重新分配到1024*32,這個(gè)其實(shí)是沒(méi)有必要的,在訪問(wèn)比較頻繁的群集,內(nèi)存會(huì)被頻繁得回收重分配,所以我們嘗試將清理的判斷條件改造為如下,就可以避免大部分沒(méi)有必要的resize操作:
if (((querybuf_size > REDIS_MBULK_BIG_ARG) &&
(querybuf_size/(c->querybuf_peak+1)) > 2) ||
(querybuf_size > 1024*32 && idletime > 2))
{
/* Only resize the query buffer if it is actually wasting space. */
if (sdsavail(c->querybuf) > 1024*32) {
c->querybuf = sdsRemoveFreeSpace(c->querybuf);
}
}
這個(gè)改造的副作用是內(nèi)存的開(kāi)銷,按照一個(gè)實(shí)例5k連接計(jì)算,5000*1024*32=160M,這點(diǎn)內(nèi)存消耗對(duì)于上百G內(nèi)存的服務(wù)器完全可以接受。
【問(wèn)題重現(xiàn)】
在使用修改過(guò)源碼的Redis server后,問(wèn)題仍然重現(xiàn)了,客戶端還是會(huì)報(bào)同類型的錯(cuò)誤,且報(bào)錯(cuò)的時(shí)候,服務(wù)器內(nèi)存依然會(huì)出現(xiàn)抖動(dòng)。抓取內(nèi)存堆棧信息如下:
Thu Jun 14 21:56:54 CST 2018
#3 0x0000003729ee893d in clone () from /lib64/libc.so.6
Thread 1 (Thread 0x7f2dc108d720 (LWP 27851)):
#0 0x0000003729ee5400 in madvise () from /lib64/libc.so.6
#1 0x0000000000493a1e in je_pages_purge ()
#2 0x000000000048cf40 in arena_purge ()
#3 0x00000000004a7dad in je_tcache_bin_flush_large ()
#4 0x00000000004a85e9 in je_tcache_event_hard ()
#5 0x000000000042c0b5 in decrRefCount ()
#6 0x000000000042744d in resetClient ()
#7 0x000000000042963b in processInputBuffer ()
#8 0x0000000000429762 in readQueryFromClient ()
#9 0x000000000041847c in aeProcessEvents ()
#10 0x000000000041873b in aeMain ()
#11 0x0000000000420fce in main ()
Thu Jun 14 21:56:54 CST 2018
Thread 1 (Thread 0x7f2dc108d720 (LWP 27851)):
#0 0x0000003729ee5400 in madvise () from /lib64/libc.so.6
#1 0x0000000000493a1e in je_pages_purge ()
#2 0x000000000048cf40 in arena_purge ()
#3 0x00000000004a7dad in je_tcache_bin_flush_large ()
#4 0x00000000004a85e9 in je_tcache_event_hard ()
#5 0x000000000042c0b5 in decrRefCount ()
#6 0x000000000042744d in resetClient ()
#7 0x000000000042963b in processInputBuffer ()
#8 0x0000000000429762 in readQueryFromClient ()
#9 0x000000000041847c in aeProcessEvents ()
#10 0x000000000041873b in aeMain ()
#11 0x0000000000420fce in main ()
顯然,Querybuffer被頻繁resize的問(wèn)題已經(jīng)得到了優(yōu)化,但是還是會(huì)出現(xiàn)客戶端報(bào)錯(cuò)。這就又陷入了僵局。難道還有其他因素導(dǎo)致query buffer resize變慢?我們?cè)俅巫トstack。但這時(shí),jemalloc引起了我們的注意。此時(shí)回想Redis的內(nèi)存分配機(jī)制,Redis為避免libc內(nèi)存不被釋放導(dǎo)致大量?jī)?nèi)存碎片的問(wèn)題,默認(rèn)使用的是jemalloc用作內(nèi)存分配管理,這次報(bào)錯(cuò)的堆棧信息中都是je_pages_purge () redis在調(diào)用jemalloc回收臟頁(yè)。我們看下jemalloc做了些什么:
arena_purge(arena.c)
static void
arena_purge(arena_t *arena, bool all)
{
arena_chunk_t *chunk;
size_t npurgatory;
if (config_debug) {
size_t ndirty = 0;
arena_chunk_dirty_iter(&arena->chunks_dirty, NULL,
chunks_dirty_iter_cb, (void *)&ndirty);
assert(ndirty == arena->ndirty);
}
assert(arena->ndirty > arena->npurgatory || all);
assert((arena->nactive >> opt_lg_dirty_mult) < (arena->ndirty -
arena->npurgatory) || all);
if (config_stats)
arena->stats.npurge++;
npurgatory = arena_compute_npurgatory(arena, all);
arena->npurgatory += npurgatory;
while (npurgatory > 0) {
size_t npurgeable, npurged, nunpurged;
/* Get next chunk with dirty pages. */
chunk = arena_chunk_dirty_first(&arena->chunks_dirty);
if (chunk == NULL) {
arena->npurgatory -= npurgatory;
return;
}
npurgeable = chunk->ndirty;
assert(npurgeable != 0);
if (npurgeable > npurgatory && chunk->nruns_adjac == 0) {
arena->npurgatory += npurgeable - npurgatory;
npurgatory = npurgeable;
}
arena->npurgatory -= npurgeable;
npurgatory -= npurgeable;
npurged = arena_chunk_purge(arena, chunk, all);
nunpurged = npurgeable - npurged;
arena->npurgatory += nunpurged;
npurgatory += nunpurged;
}
}
Jemalloc每次回收都會(huì)判斷所有實(shí)際應(yīng)該清理的chunck并對(duì)清理做count,這個(gè)操作對(duì)于高響應(yīng)要求的系統(tǒng)是很奢侈的,所以我們考慮通過(guò)升級(jí)jemalloc的版本來(lái)優(yōu)化purge的性能。Redis 4.0版本發(fā)布后,性能有很大的改進(jìn),并可以通過(guò)命令回收內(nèi)存,我們線上也正準(zhǔn)備進(jìn)行升級(jí),跟隨4.0發(fā)布的jemalloc版本為4.1,jemalloc的版本使用的在jemalloc的4.0之后版本的arena_purge()做了很多優(yōu)化,去掉了計(jì)數(shù)器的調(diào)用,簡(jiǎn)化了很多判斷邏輯,增加了arena_stash_dirty()方法合并了之前的計(jì)算和判斷邏輯,增加了purge_runs_sentinel,用保持臟塊在每個(gè)arena LRU中的方式替代之前的保持臟塊在arena樹(shù)的dirty-run-containing chunck中的方式,大幅度減少了臟塊purge的體積,并且在內(nèi)存回收過(guò)程中不再移動(dòng)內(nèi)存塊。代碼如下:
arena_purge(arena.c)
static void
arena_purge(arena_t *arena, bool all)
{
chunk_hooks_t chunk_hooks = chunk_hooks_get(arena);
size_t npurge, npurgeable, npurged;
arena_runs_dirty_link_t purge_runs_sentinel;
extent_node_t purge_chunks_sentinel;
arena->purging = true;
/*
* Calls to arena_dirty_count() are disabled even for debug builds
* because overhead grows nonlinearly as memory usage increases.
*/
if (false && config_debug) {
size_t ndirty = arena_dirty_count(arena);
assert(ndirty == arena->ndirty);
}
assert((arena->nactive >> arena->lg_dirty_mult) < arena->ndirty || all);
if (config_stats)
arena->stats.npurge++;
npurge = arena_compute_npurge(arena, all);
qr_new(&purge_runs_sentinel, rd_link);
extent_node_dirty_linkage_init(&purge_chunks_sentinel);
npurgeable = arena_stash_dirty(arena, &chunk_hooks, all, npurge,
&purge_runs_sentinel, &purge_chunks_sentinel);
assert(npurgeable >= npurge);
npurged = arena_purge_stashed(arena, &chunk_hooks, &purge_runs_sentinel,
&purge_chunks_sentinel);
assert(npurged == npurgeable);
arena_unstash_purged(arena, &chunk_hooks, &purge_runs_sentinel,
&purge_chunks_sentinel);
arena->purging = false;
}
【解決問(wèn)題】
實(shí)際上我們有多個(gè)選項(xiàng)。可以使用Google的tcmalloc來(lái)代替jemalloc,可以升級(jí)jemalloc的版本等等。我們根據(jù)上面的分析,嘗試通過(guò)升級(jí)jemalloc版本,實(shí)際操作為升級(jí)Redis版本來(lái)解決。我們將Redis的版本升級(jí)到4.0.9之后觀察,線上客戶端連接超時(shí)這個(gè)棘手的問(wèn)題得到了解決。
【問(wèn)題總結(jié)】
Redis在生產(chǎn)環(huán)境中因其支持高并發(fā),響應(yīng)快,易操作被廣泛使用,對(duì)于運(yùn)維人員而言,其響應(yīng)時(shí)間的要求帶來(lái)了各種各樣的問(wèn)題,Redis的連接超時(shí)問(wèn)題是其中比較典型的一種,從發(fā)現(xiàn)問(wèn)題,客戶端連接超時(shí),到通過(guò)抓取客戶端與服務(wù)端的網(wǎng)絡(luò)包,內(nèi)存堆棧定位問(wèn)題,也被其中一些假象所迷惑,最終通過(guò)升級(jí)jemalloc(Redis)的版本解決問(wèn)題,這次最值得總結(jié)和借鑒的是整個(gè)分析的思路。
總結(jié)
以上就是這篇文章的全部?jī)?nèi)容了,希望本文的內(nèi)容對(duì)大家的學(xué)習(xí)或者工作具有一定的參考學(xué)習(xí)價(jià)值,如果有疑問(wèn)大家可以留言交流,謝謝大家對(duì)腳本之家的支持。
相關(guān)文章
基于redis 7.2.3的makefile源碼解讀學(xué)習(xí)
這篇文章主要為大家介紹了基于redis 7.2.3的makefile源碼解讀學(xué)習(xí),有需要的朋友可以借鑒參考下,希望能夠有所幫助,祝大家多多進(jìn)步,早日升職加薪2023-12-12
Redis偶發(fā)連接失敗案例實(shí)戰(zhàn)記錄
這篇文章主要給大家介紹了關(guān)于Redis偶發(fā)連接失敗的相關(guān)資料,文中通過(guò)示例代碼介紹的非常詳細(xì),對(duì)大家學(xué)習(xí)或者使Redis具有一定的參考學(xué)習(xí)價(jià)值,用需要的朋友們下面隨著小編來(lái)一起學(xué)習(xí)學(xué)習(xí)吧2018-10-10
Redis中一些最常見(jiàn)的面試問(wèn)題總結(jié)
Redis在互聯(lián)網(wǎng)技術(shù)存儲(chǔ)方面使用如此廣泛,幾乎所有的后端技術(shù)面試官都要在Redis的使用和原理方面對(duì)小伙伴們進(jìn)行各種刁難。下面這篇文章主要給大家總結(jié)介紹了關(guān)于Redis中一些最常見(jiàn)的面試問(wèn)題,需要的朋友可以參考下2018-09-09
NoSQL和Redis簡(jiǎn)介及Redis在Windows下的安裝和使用教程
這篇文章主要介紹了NoSQL和Redis簡(jiǎn)介及Redis在Windows下的安裝和使用教程,本文同時(shí)講解了python操作redis,并給出了操作實(shí)例,需要的朋友可以參考下2015-01-01
redis requires ruby version2.2.2的解決方案
本文主要介紹了redis requires ruby version2.2.2的解決方案,文中通過(guò)示例代碼介紹的非常詳細(xì),需要的朋友們下面隨著小編來(lái)一起學(xué)習(xí)學(xué)習(xí)吧2021-07-07

