One way for oracle12c to manage job resources: oracle12c job resources

One way for oracle12c to manage job resources: oracle12c job resources

Database: 12.1.0.2, rac, cdb Mode

 

I am responsible for moving two cdb clusters of 12.1.0.2, one on aix and the other on linux. Unfortunately, they are both hybrid and have more than 100 TB of data.

Because 12c was already delivered by other departments, I was not very familiar with 12c before, but I still want to see if I can maximize performance without sharding.

The result is not good, because some applications that prefer OLTP often feel slow.

 

What should I do? You have to specify a node through the SERVICE. Each SERVICE must have at least two nodes. The SERVICE adopts the taf configuration, one master node and one slave node.

For example, a service created for an oltp is as follows:

Srvctl add service-db wgdb-service nsn_flowdb-preferred wgdb2-available wgdb4-tafpolicy preconnect-policy automatic-failovertype session-failovermethod basic-Rule 180-rule 2-pdb nsn_flow

In addition, node 2 is only allocated to all oltp applications and other olap applications are allocated to other nodes.

I personally think that, in this regard, oracle databases are far better than rdbms and bigdata, which can save a lot of resources and costs for enterprises.

The Resource Management of other rdbms is a bit complicated, which is far less convenient than that of oracle.

In other words, oracle is doing so conveniently to move closer to cloud.

 

Later, most of the time, oltp applications are very fast, but occasionally slow, such as a morning or a certain time point in the afternoon.

What's going on? After analyzing the running logs of the job, it is found that the job runs on all nodes instead of the created service, and runs on each node according to the oracle database's own balancing algorithm.

I also studied oracle jobs and found that the service can be controlled through the scheduling class. I am very happy that no complicated operations are needed.

But the problem comes again:

1. existing jobs use the default scheduling class. The default scheduling class does not specify nodes.

2. In the future, developers will still use the default scheduling class if they are not forced.

 

Therefore, we decided to modify the default scheduling class of other pdb instances:

begin  dbms_scheduler.set_attribute(name => 'DEFAULT_JOB_CLASS',attribute => 'comments',value => 'This is for all olap app');  dbms_scheduler.set_attribute(name => 'DEFAULT_JOB_CLASS',attribute => 'service',value => 'wybigdata');  end;

As for the slow process of running OLap on a specific node, it does not matter. olap can tolerate this situation.

 

Given that oracle's scheduling is so powerful, it was subsequently decided to force various vendors to use scheduling to execute their jobs instead of jobs. The latter was too bad and should have been thrown into the history garbage.

In addition to controlling resource allocation through services, oracle can also manage resource allocation for scheduling jobs in other ways, such as through resource groups.

 

Therefore, it is very important to select the cluster configuration method at the initial stage of cluster construction.