We have deployed an Oracle RAC cluster which consists of three Oracle instances. As we are considering the HA issue, and as far as I know, the only way of Oracle HA is DataGuard. So is it possible that we deploy DataGuard in another data center as backup for the master RAC cluster and if master RAC cluster down, We can switch the DataGuard to master which enable read and write operations. Someone told me that it is a very complicated process. Is that true? how much time does this process cost? Is there any other solutions on Oracle HA or any other load balance solution to replace RAC cluster? Does Mysql have the same capability?

  • RAC is a HA solution: if a node goes down, you still have the others. Data Guard is a DR solution: if your storage goes down or all nodes become unreachable, you have a copy you can failover to. Performing a Data Guard role transition can be done by issuing a single command after connecting to the database, and can finish in seconds. It is not at all complicated. Of course some DBAs are still stuck on manual labor, they do not configure the broker, perform each step manually in 10 minutes, instead of using the broker and doing the same in 10 seconds. It can even happen automatically. – Balazs Papp Dec 1 at 13:50
  • @BalazsPapp Would you please explain that how to configure the broker? I am not familiar with it. Another question, if we have a Oracle RAC cluster which consists of 3 oracle instances, and we also want to a DR solution, we should set 3 Data Guard Oracle instances for each instance in RAC, is that right? – yifan Dec 1 at 22:44
  • The configuration steps are far beyond the scope of a simple comment here. For your other question: there is no such requirement. The number of instances can differ in different databases in a Data Guard configuration. So you can have a 3 node RAC primary database with a let's say, 2 node RAC standby, or even a single instance standby database. The question is: in case of a role transition, can the single instance standby handle the same load that the 3 node RAC primary could handle? – Balazs Papp Dec 4 at 18:07
  • @BalazsPapp Thank you for your instruction. Do you have any resources about the configuration of role transition? I am a developer. As our DBA thought it was impossible to complete this proccess in 10 seconds, I want to know more about it and have a try by myself. – yifan Dec 5 at 7:49
  • docs.oracle.com/en/database/oracle/oracle-database/18/dgbkr/… And here is an output of a switchover with broker: pastebin.com/rACkimPA It can be seen from the alert log, that the switchover procedure on the standby (becoming an open primary from the mounted standby) site took 11 seconds (ok, not 10...). – Balazs Papp Dec 5 at 10:27

(MySQL)

There are several MySQL Replication topologies. For HA, deploying at least one server in each of 3 datacenters and using Galera Cluster, failover is essentially automatic if a datacenter goes down. Some form of proxy or load balancer could take care of redirecting users to the still-alive nodes. When the datacenter comes back up, that node will get back in sync.

If a node crashes and burns, wheel in a new server; the live nodes will pour a copy of the data set into it, and you are back on the air.

And you did not get paged.

(Now perhaps we will hear from users for which it was not that seemless.)

Group Replication is another HA product on top of MySQL. I have seen 3 nodes being set up during a conference talk. It took one person less than 5 minutes.

Caveat: I cannot say whether either of these is easier or harder than Oracle's offering(s).

  • As far as I know, Galera Cluster is not the official product of MySQL. I am not familiar with the features of this product. The distance between our data centers is quite long. So the quality of data transmission may be a big issue. By the way, How does these MySQL products handle the data consistency if a switchover occurs. – yifan Dec 1 at 3:25
  • @yifan - Data consistency -- That's rather complex. And differs between Galera and Group Replication. I don't want to dive into the details here. I should point out that Galera allows writing to all nodes. And a COMMIT can fail if a conflict is encountered on a remote node. By long distance how many milliseconds ping time are you talking about. – Rick James Dec 1 at 4:54
  • for how many milliseconds ping time am I talking about, It depends on many factors. We have data centers in Aisa and north america. In normal condition, the transmission is ok. But the quality of communition in such a long distance can not be guaranteed. – yifan Dec 1 at 9:22

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.