Servers leaving or joining a cluster resulting a reconfiguration that is essentially a synchronization event to recover all the changes made by the failed instance. Oracle RAC has reduced the time, the sessions wait on this event during reconfiguration.
In Oracle RAC 23ai, smart reconfiguration reduces the impact of a service disruption from both planned and unplanned operations, utilizing several features, such as Recovery Buddy and PDB and service isolation and smooth reconfiguration, resulting in a faster reconfiguration than previous releases. So we’re going to take a look at some of the features introduced in previous releases, and then we’re going to get into smooth reconfiguration that was introduced in 23ai.
Let’s review global resource management. One user makes a connection to one of the RAC instances, and user can submit SQL statement requesting a set of blocks. In order to catch database blocks, the buffers must be allocated and also master metadata must be allocated to be able to describe changes to those buffers.
An internal algorithm is used to decide which instance should contain the master metadata structure for that entity. In our example, the master metadata structure are distributed across instance 1 and instance 2. During the instance startup, this information on the master metadata structure for the entity has persisted in the data dictionary and reused during instance startup. And also the global resources are managed for unplanned instance crash and also planned service relocation as well.
Now, let’s review PDB and service isolation. Let’s assume, there are three PDBs– PDB1, PDB2, PDB3.
PDB1 is running on instance 1 and PDB2 is running on instance 1, instance 2, and instance 3. And also PDB3, it is available in instance 2 and instance 3. So when you make any changes to PDB1, then the metadata structure owned by PDB1 is only available in instance 1.
When you make any changes to PDB3, the master metadata structure for PDB3 will be distributed across the instances where PDB3 is up and running. In this example, in instance 2 and instance 3. So let’s take a look at RAC reconfiguration.
The PDB in RAC embodiment is reconfigured only needed if PDB1 is open on instance 2 as an example. So, for example, originally, PDB1 was available on instance 1. When you start PDB1 even in instance 2, the master metadata is redistributed across instance 1 and instance 2.
And also if CDB 2 goes up, then what will happen? All the PDBs that were running out of instance 2 are no longer available. So the master metadata that was kept in instance 2 must be redistributed across surviving instances, in this example, instance 1 and instance 2.
And impact is isolated to the affected PDBs only if the PDB is unaffected when CDB instance crashes. And also PDB1 is open on instance 2, or fourth instance is brought up. So these cases, the impact is only isolated at the PDB level.
The PDB and service isolation. This is a feature that is used for CDB, and this is an enhancement from the service-oriented buffer cache access. And this feature improves performance by reducing distributed lock manager operations for services not offered in all PDB instances.
The next topic is Buddy Recovery for reconfiguration. The Recovery Buddy feature is a feature that reduces the waiting time during reconfiguration. In prior releases, Oracle RAC instances identified and recovered the changes made by the failed instance by reading the redo logs.
So, for example, for instance 1, PDB1 goes down. In order to recover blocks, the heavily modified on PDB1, one of the surviving instances must access the redo log file owned by PROD 1 and then identify blocks to be recovered. So that’s in a physical I/O. And it is a time-consuming operation.
With the Recovery Buddy feature, we can reduce this I/O because of in-memory log and also because of Recovery Buddy concept. So, for example, in the three-node direct database, we actually assign the Buddy Recovery for each instance. For example, PDB1 is a Recovery Buddy of PROD 3, and also PROD 2 is a Recovery Buddy of a PROD 1, PROD 3 is a Recovery Buddy of PROD 2.
So which means that when you make any changes to the blocks in instance 1, PROD 1, then changes will be captured directly in PROD 1 but also the same changes will be maintained in the Recovery Buddy memory. So in-memory log. And same thing– when you make any changes to PROD 2, this change will be maintained not only locally but also in the Buddy Recovery instance as well.
So here’s an example. So we connect to instance 1 and request the blocks like this and then make changes. So we make changes like this. Now, when you make any changes to PROD 1, these changes are maintained in PROD 1 but also the same change is maintained in the Recovery Buddy instance.
So if PROD 1 goes down, instead of having access to online redo log file owned by Prod 1, we can directly access to in-memory log preserved in the Recovery Buddy instance and then read it to identify blocks to be recovered. So once we identify blocks to be recovered and apply changes, then it recover. So this feature reduces the time required for reconfiguration.
Smooth reconfiguration. Smooth reconfiguration of Oracle Real Application Clusters instance reduces brownout time during cluster reconfiguration. So here’s an example. Suppose that you run srvctl command to stop instance, in the previous version, as soon as you run srvctl stop instance command, your instance is just stopped.
And also until the metadata that was kept in the stop instance is redistributed, your database was frozen for that amount of time until global resources are recovered. However, in 23ai, we changed the algorithm slightly. So you request a stop instance.
However, instead of stopping instance immediately, we perform the resource remastering operation first. So we distribute the metadata before performing stopping of an instance. So after redistributing instance, then we’re going to actually shut down instance.
So when you actually look at the differences between version 19c, for example, and then 23ai, we slightly change the order, and that reduced the time required for the cluster reconfiguration. So in 19c, as soon as you issue stop instance command, srvctl stop instance, your instance must be killed and stopped. And then the global resource must be remastered. So for a short amount of time, your database wasn’t able to perform any activities.
However, in 23ai, when user requests the srvctl stop instance, instead of stopping an instance, we remaster the resource first. And then after the resources are remastered, then we could actually shut down instance. That reduces reconfiguration time.
So this feature it distributes resource coordinator. So resource coordinator is same as the master resource, the owner, or the resource master. So same terminology. So we call it now resource coordinator, before shutting down instances for planned maintenance.
So this feature, smooth reconfiguration, works for planned maintenance, such as when user requests a certain operation by issuing user command. And also it reduces the reconfiguration time as the resources do not need to be remastered. So before shutting down, we remaster resource first and then shut down. And applications benefit from reduced the burnout time during planned maintenance operation.
Recent Comments