By default, infrastructure maintenance updates the Exadata database server hosts in a rolling fashion, followed by updating the storage servers.

You can also choose non-rolling maintenance to update database and storage servers. The non-rolling maintenance method first updates your storage servers at the same time, then your database servers at the same time. Although non-rolling maintenance minimizes maintenance time, it incurs full system downtime while the storage servers and database servers are being updated.

Rolling infrastructure maintenance begins with the Exadata database server hosts. For the rolling maintenance method, database servers are updated one at a time. Each of the database server host's VMs is shut down, the host is updated, restarted, and then the VMs are started, while other database servers remain operational. This rolling maintenance impact older applications not written to handle a rolling instance outage. This process continues until all servers are updated.

After database server maintenance is complete, storage server maintenance begins. For the rolling maintenance method, storage servers are updated one at a time and do not impact VM cluster VM's availability. However, the rolling storage server maintenance can result in reduced IO performance as storage servers are taken offline (reducing available IO capacity) and resynced when brought back online (small overhead on database servers). Properly sizing the database and storage infrastructure to accommodate increased work distributed to database and storage servers not under maintenance will minimize (or eliminate) any performance impact.

Note that while databases are expected to be available during the rolling maintenance process, the automated maintenance verifies Oracle Clusterware is running but does not verify that all database services and pluggable databases (PDBs) are available after a server is brought back online. The availability of database services and PDBs after maintenance can depend on the application service definition. For example, a database service, configured with certain preferred and available nodes, may be relocated during the maintenance and wouldn't automatically be relocated back to its original node after the maintenance completes. Oracle recommends reviewing the documentation on Achieving Continuous Availability for Your Applications on Exadata Cloud Systems to reduce the potential for impact to your applications. By following the documentation's guidelines, the impact of infrastructure maintenance will be only minor service degradation as database servers are sequentially updated.

Oracle recommends that you follow the Maximum Availability Architecture (MAA) best practices and use Data Guard to ensure the highest availability for your critical applications. For databases with Data Guard enabled, Oracle recommends that you separate the maintenance windows for the infrastructure instances running the primary and standby databases. You may also perform a switchover prior to the maintenance operations for the infrastructure instance hosting the primary database. This allows you to avoid any impact on your primary database during infrastructure maintenance.

Prechecks are performed on the Exadata Cloud@Customer infrastructure components prior to the start of the maintenance window. The goal of the prechecks is to identify issues that may prevent the infrastructure maintenance from succeeding. The Exadata infrastructure and all components remain online during the prechecks. An initial precheck is run approximately two weeks prior to the maintenance start and another precheck is run approximately 24 hours prior to maintenance start. If the prechecks identify an issue that requires rescheduling the maintenance notification is sent to the maintenance contacts.

The time taken to update infrastructure components varies depending on the number of database servers and storage servers in the Exadata infrastructure, the maintenance method, and whether custom action has been enabled. The approximate times provided are estimates. Time for custom action, if configured, is not included in the estimates below. Database server maintenance time may vary depending on the time required to shutdown each VM before the update and then start each VM and associated resources after the update of each node before proceeding to the next node. The storage server maintenance time will vary depending on the time required for the ASM rebalance, which is not included in the estimates below. If issues are encountered during maintenance this may also delay completion beyond the approximate time listed. In such a situation, if Oracle cloud operations determine resolution would extend beyond the expected window, they will send a notification and may reschedule the maintenance.

• Rolling:
  • Each database server takes 90 minutes on average.
  • Each storage server takes 60 minutes on average.
  • Each InfiniBand or RoCE fabric switch takes 30 minutes on average.

  • The approximate total time for infrastructure maintenance is as follows:

    • Base and Quarter Rack (2 Database Servers/3 Storage Servers): Approximately 7 hours

      2 Database Servers X 90 = 180 minutes

      3 Storage Servers X 60 = 180 minutes

      2 InfiniBand or RoCE Fabric Switch X 30 = 60 minutes

    • Half Rack (4 Database Servers/6 Storage Servers): Approximately 13 hours

      4 Database Servers X 90 = 360 minutes

      6 Storage Servers X 60 = 360 minutes

      2 InfiniBand or RoCE Fabric Switch X 30 = 60 minutes

    • Full Rack (8 Database Servers/12 Storage Servers): Approximately 26 hours

      8 Database Servers X 90 = 720 minutes

      12 Storage Servers X 60 = 720 minutes

      2 InfiniBand or RoCE Fabric Switch X 30 = 60 minutes

• Non-Rolling:
  • All database servers take 180 minutes on average.
  • All storage servers take 60 minutes on average.
  • Storage Servers and Database servers are brought back online prior to starting fabric switch maintenance.
  • Network fabric switches are still updated in a rolling method and take 30 minutes each on average.
  • The approximate total time for infrastructure maintenance is 5 hours regardless of shape:
    • All Database Servers = 180 minutes
    • All Storage Servers = 60 minutes
    • 2 InfiniBand or RoCE Fabric Switch = 60 minutes

To edit your Oracle Exadata Database Service on Cloud@Customer infrastructure quarterly maintenance preferences, be prepared to provide values for the infrastructure configuration. The changes you make will only apply to future maintenance runs, not those already scheduled.

Learn how to view and edit the time of the next scheduled maintenance.

1. Open the navigation menu. Under Oracle Database, click Exadata Database Service on Cloud@Customer.
2. Choose your Compartment.
3. Click Exadata Infrastructure.
4. In the list of Exadata Infrastructures, find the infrastructure you want to set the next scheduled maintenance window for and click its highlighted name.

   The Infrastructure Details page displays information about the selected Oracle Exadata infrastructure.

   Note
   
   

   An information block is displayed 6 hours before the start of a maintenance run, regardless of whether you've chosen rolling or non-rolling maintenance method. When the maintenance begins, it is automatically removed.

5. On the Infrastructure Details page, under Maintenance, click the view link in the Next Quarterly Maintenance field.

   The Exadata Infrastructure Maintenance page is displayed.

6. On the Exadata Infrastructure Maintenance page, scheduled maintenance details are listed.

   Target DB Server Version and Target Storage Server Version: These fields display the Exadata software version to be applied by the scheduled maintenance. The version applied will be the most recent certified update for Exadata infrastructures in the cloud. If the next quarterly update is not yet certified when the maintenance is scheduled, then the versions may show "LATEST" until the new quarterly update becomes available. Once the update becomes available the new version will be displayed.

   To find information on the Database Server Exadata software version or the Storage Server Exadata software version, see My Oracle Support note Exadata Database Machine and Exadata Storage Server Supported Versions (Doc ID 888828.1).

7. To change the next scheduled maintenance settings, click Edit Maintenance Run.
   On the Edit Maintenance page, do the following:

   • Select a maintenance method, Rolling or Non-rolling.
     Note
     
     

     If you select the Non-rolling option, an information block appears stating that components will be updated simultaneously, resulting in full system downtime.

   • Enable custom action before performing maintenance on DB servers: Enable custom action only if you want to perform additional actions outside of Oracle’s purview. For maintenance configured with a rolling software update, enabling this option will force the maintenance run to wait for a custom action with a configured timeout before starting maintenance on each DB server. For maintenance configured with non-rolling software updates, the maintenance run will wait for a custom action with a configured timeout before starting maintenance across all DB servers. The maintenance run, while waiting for the custom action, may also be resumed prior to the timeout.
     • Custom action timeout (in minutes): Maximum timeout available to perform custom action before starting maintenance on the DB Servers.

       Default: 30 minutes

       Maximum: 120 minutes

   • To reschedule the next quarterly maintenance run, enter a date and time in the Scheduled Start time field.
     The following restrictions apply:

     • Oracle expects to be able to perform infrastructure maintenance at least once per quarter. You should not defer maintenance beyond the end of a maintenance quarter unless unexpected issues prevent your accommodating it before the next maintenance quarter.
     • In the event unexpected issues prevent your accommodating the scheduled infrastructure maintenance run, you can reschedule the infrastructure maintenance to another date no more than 180 days from the prior infrastructure maintenance. Since normal maintenance should be performed quarterly, this provides approximately 90 additional days for you to reschedule the infrastructure maintenance. Oracle strongly recommends you not schedule maintenance at or close to the 180 day limit, as you will have no flexibility to reschedule further if additional unexpected issues arise.
     • If a new maintenance release is announced prior to your rescheduled maintenance run, the newer release will be applied on your specified date.
     • You can reschedule your maintenance to take place earlier than it is currently scheduled. You cannot reschedule the maintenance if the current time is within 2 hours of the scheduled maintenance start time.
     • Oracle reserves certain dates each quarter for internal maintenance operations, and you cannot schedule your maintenance on these dates.
   • Click Save Changes.
8. To view estimated maintenance time details for various components, click the View link is displayed in the Total Estimated Maintenance Time field.

   The View link is displayed in the Total Estimated Maintenance Time field only if the Maintenance Method is Rolling.

   The Estimated Maintenance Time Details page is displayed with details that include:

   • Total Estimated Maintenance Time
   • Database Servers Estimated Maintenance Time
   • Storage Servers Estimated Maintenance Time
   • Network Switches Estimated Maintenance Time
   • Order in which components are updated. In rolling maintenance, components are updated in the sequence displayed

   1. To view the number of VMs that will be restarted as part of Database Server maintenance, click the Show details link.

      The VM Location dialog is displayed.

   2. In the VM Cluster Name field, you can find out what VM cluster a particular VM belongs to.
   3. Click Close.
9. Click Close to close the Estimated Maintenance Time Details page.

Learn how to view the maintenance history for an Oracle Exadata Database Service on Cloud@Customer Infrastructure.

1. Open the navigation menu. Under Oracle Database, click Exadata Database Service on Cloud@Customer.
2. Choose your Compartment.
3. Click Exadata Infrastructure.
4. In the list of Exadata Infrastructures, find the infrastructure you want to view the maintenance history and click its highlighted name.

   The Infrastructure Details page displays information about the selected Oracle Exadata infrastructure.

5. On the Infrastructure Details page, under Maintenance, click the view link in the Next Quarterly Maintenance field.

   The Exadata Infrastructure Maintenance page is displayed.

6. On the Exadata Infrastructure Maintenance page, click Maintenance History to see a list of past maintenance events including details on their completion state and the target database and storage server versions.

   To find information on the Database Server Exadata software version or the Storage Server Exadata software version, see My Oracle Support note Exadata Database Machine and Exadata Storage Server Supported Versions (Doc ID 888828.1).