sequence自增原理是什么?

為什么get命令看到的cversion和實(shí)際存儲(chǔ)的cversion不一樣?

sequence自增怎么保證無(wú)并發(fā)問(wèn)題?單節(jié)點(diǎn)和多節(jié)點(diǎn)?

1. sequence自增原理是什么?

   // 代碼取自(服務(wù)端代碼)org.apache.zookeeper.server.PrepRequestProcessor#pRequest2TxnCreate
   ChangeRecord parentRecord = getRecordForPath(parentPath);
   checkACL(zks, parentRecord.acl, ZooDefs.Perms.CREATE, request.authInfo);
   int parentCVersion = parentRecord.stat.getCversion();
   if (createMode.isSequential()) {
       // 如果是順序節(jié)點(diǎn)的話,從父節(jié)點(diǎn)中取出cversion的值作為sequence.
       path = path + String.format(Locale.ENGLISH, "%010d", parentCVersion);
   }
   

2. 為什么get命令看到的cversion和實(shí)際存儲(chǔ)的cversion不一樣?

   知道了sequence取值邏輯后,用get命令查看父節(jié)點(diǎn)的cversion字段,卻發(fā)現(xiàn)總是和新生成節(jié)點(diǎn)的sequence不一致?最后定位到get命令的代碼.

   // 代碼取自(服務(wù)端代碼)org.apache.zookeeper.server.DataNode#copyStat
   synchronized public void copyStat(Stat to) {
           to.setAversion(stat.getAversion());
           to.setCtime(stat.getCtime());
           to.setCzxid(stat.getCzxid());
           to.setMtime(stat.getMtime());
           to.setMzxid(stat.getMzxid());
           to.setPzxid(stat.getPzxid());
           to.setVersion(stat.getVersion());
           to.setEphemeralOwner(getClientEphemeralOwner(stat));
           to.setDataLength(data == null ? 0 : data.length);
           int numChildren = 0;
           if (this.children != null) {
               numChildren = children.size();
           }
        // 重點(diǎn)!
           // when we do the Cversion we need to translate from the count of the creates
           // to the count of the changes (v3 semantics)
           // for every create there is a delete except for the children still present
        // 當(dāng)我們進(jìn)行Cversion時(shí)，我們需要從create的數(shù)量中進(jìn)行翻譯更改計(jì)數(shù)（v3語(yǔ)義）對(duì)于每個(gè)創(chuàng)建都有一個(gè)刪除，除了仍然存在的孩子(Google 翻譯)
           to.setCversion(stat.getCversion()*2 - numChildren);
           to.setNumChildren(numChildren);
   }
   

3. sequence自增怎么保證無(wú)并發(fā)問(wèn)題?單節(jié)點(diǎn)和多節(jié)點(diǎn)?

   先來(lái)說(shuō)多節(jié)點(diǎn)吧,(參考網(wǎng)上通常的說(shuō)法)當(dāng)一個(gè)客戶端進(jìn)行寫(xiě)數(shù)據(jù)請(qǐng)求時(shí),會(huì)指定ZooKeeper集群中的一個(gè)Server節(jié)點(diǎn),如果該節(jié)點(diǎn)為Follower,則該節(jié)點(diǎn)會(huì)把寫(xiě)請(qǐng)求轉(zhuǎn)發(fā)給Leader,Leader通過(guò)內(nèi)部的協(xié)議進(jìn)行原子廣播,直到一半以上的server節(jié)點(diǎn)都成功寫(xiě)入數(shù)據(jù),這次寫(xiě)請(qǐng)求便算是成功,然后Leader便會(huì)通過(guò)通知相應(yīng)的Follower節(jié)點(diǎn)寫(xiě)請(qǐng)求成功,該節(jié)點(diǎn)向client返回寫(xiě)入成功.
   本質(zhì)上說(shuō)還是把多節(jié)點(diǎn)寫(xiě)操作轉(zhuǎn)換成了單節(jié)點(diǎn)的寫(xiě)操作,這樣就可以在單節(jié)點(diǎn)進(jìn)行并發(fā)控制了,效率更高.而在單節(jié)點(diǎn)做并發(fā)控制的話就方便的多了,通過(guò)PrepRequestProcessor->SyncRequestProcessor->FinalRequestProcessor->ZooKeeperServer->ZKDatabase->DataTree鏈路處理.有興趣可以自己翻翻代碼哦.