Oracle disaster recovery licensing in 2026. When the standby needs its own license.

Oracle gives disaster recovery one narrow concession: a single spare node in a shared storage cluster, for ten separate 24 hour periods a year. Almost every standby and mirror site outside it needs a full license, options included.

What does Oracle's failover rule allow for disaster recovery?

It allows one unlicensed spare node in a cluster to run the licensed database for up to ten separate 24 hour periods in a calendar year while the primary is down. That is the whole concession. Oracle has no general disaster recovery exemption, and reading the failover rule as one is the most expensive mistake in this subject.

Oracle's policy text calls it the 10 day rule, first set out in the software investment guide, and the current wording sits in Oracle's Licensing Data Recovery Environments document, dated December 15, 2024. Both are policy documents outside your contract and have been revised more than once, so check the current version against your ordering documents before relying on it.

Which conditions must all hold at once?

• Shared storage. The failover node and the production node sit in one cluster and attach to the same logical disk array.
• One data center. That array is located in a single data center, which rules out a second site by definition.
• One node. Only a single spare in each cluster qualifies, however many are configured.
• Whole period counting. Any part of a 24 hour period the node runs counts as a full period.
• Failover only. The node runs production's workload because the primary is down. Using it as somewhere to run reports is licensed use.
• Return to normal. When the primary is repaired, you switch the workload back to it, or you designate the repaired server as the new failover node.

How are the ten days counted?

By separate 24 hour periods, and partial use rounds up. Oracle's own example counts two hours of failover on Tuesday and three hours on Friday as two periods. Once the total passes ten in a calendar year, the failover node must be licensed.

The 2024 wording also says downtime for maintenance counts toward the ten. That changes the arithmetic for patching. On that reading, a team that runs production on the spare for a few hours each month while it patches the primary uses 12 periods a year, and breaks the limit before any real failure happens.

Which Oracle disaster recovery setups need a full license?

Every setup except that single spare node needs a full license for the database and for every option running on it, and so does the spare once it passes the allowance. The rule is also silent on several things buyers assume it covers. The table sets out the configurations we meet most often.

Are planned failover drills covered?

The wording does not settle it. The data recovery document carries a narrow testing right: an unlicensed server, including one in a cloud environment, may run the program up to four times in a calendar year, for no more than two days each time, to test physical copies of backups.

That right is written for restoring backups. A Data Guard switchover drill or a cluster failover test is a different activity, and reasonable people have read its treatment both ways. The clause that counts maintenance downtime suggests that, at best, a planned drill uses up part of the ten periods.

Why should you not settle that question yourself?

An auditor will read any gap in the wording against you. Ask Oracle for written confirmation of how planned drills are treated, keep the reply with your license records, and design the runbook around the conservative reading until you have it.

How does a standby in the cloud change the picture?

The failover section of Oracle's document does not mention cloud at all. A standby running in AWS or Azure is counted under Oracle's authorized cloud policy on its vCPUs, with the same edition and options as production. Only the backup testing right names cloud servers explicitly.

Does an Oracle Data Guard standby need its own license?

Yes, in almost every real configuration. A physical standby keeps its own copy of the data on its own storage, so it fails the shared storage condition before any other test applies. Oracle's document says that in copying, synchronizing and mirroring setups every program installed or running must be licensed under standard policy.

Why does an idle standby still count?

The trigger is installation and capability. Software installed on a server that can run it is licensable, which is why a standby sitting in mount mode, applying redo and serving no users, still counts. The node being idle changes nothing about the position.

The same logic applies to storage replication. If array replication copies the Oracle home to a DR server that can mount it and start an instance, count that server as installed when you reconcile your own records.

Which options does the standby need?

A standby applying redo needs the same edition license as the primary, and Data Guard itself is included in Enterprise Edition. If the primary runs a priced option such as Partitioning, the standby carries those objects block for block and needs the option too. Oracle's document goes further and requires license metrics and program options to match on both servers.

• One exception for Real Application Clusters. The policy names two exceptions to the matching rule. RAC is not needed on the recovery server unless it is used there, and the other exception applies only to production licensed through certain Oracle cloud database services.
• Same metric. If production is licensed by Named User Plus, so is the standby, and the Enterprise Edition minimum of 25 Named User Plus per processor applies to the standby's own hardware.
• Active Data Guard. Opening the standby for queries while redo apply runs adds a separate option that is licensed on the primary as well. Our active standby guide works through that line, and the replication technology is documented on the Data Guard product pages.

What does a standby cost at list price?

Say production runs on a two socket Intel server with 24 cores in total and uses Partitioning and Advanced Compression. At Oracle's 0.5 core factor, 24 cores count as 12 processors. The standby sits on identical hardware at a second site, so the failover rule never applies.

A discount lowers every line by a similar share, so the proportions hold at any negotiated price, and the two option lines come to $276,000 of license and $60,720 a year of support. Those are the lines that go missing when the standby is licensed as a bare database. Check current figures against the price list before you budget.

What have we seen in Oracle disaster recovery reviews?

Across roughly 30 to 40 Oracle environments I reviewed in 2024 and 2025, disaster recovery was the most misunderstood area of the licensing. The most common single finding was a fully replicated standby treated as exempt. Four patterns recurred.

• Unlicensed standby. In about 6 out of 10 environments the standby or mirror site carried no license at all, on the belief that disaster recovery is free.
• Drills under the concession. Teams stretched the failover rule to cover routine drills, which the wording does not clearly allow and which no one had confirmed with Oracle in writing.
• Shared storage failed. Where the concession did apply on paper, the cluster failed the shared storage condition, so the relief was never available.
• Options left off. In 7 environments the standby was licensed for the database but not for the options on it, usually partitioned or otherwise optioned objects that had never been entitled separately.

The unlicensed standby is the large, expensive finding. The options gap is smaller and was budgeted by no one, and it survives the first remediation because everybody fixes the database line and stops.

Disaster recovery is where Oracle licensing costs the most money for the least deliberate reason.

Why we reject the idea that a standby site is free

The usual advice is that disaster recovery falls under the failover concession, so a standby site costs nothing, and we disagree. The concession covers one node in one cluster in one data center, for ten periods a year. A second site fails it on location, and a replicated standby fails it on storage.

The better course is to classify every recovery node against the six conditions, license the ones that fail, get the drills treatment in writing and reconcile the standby's options against the primary. Our Oracle practice runs that classification before an auditor opens the question.

What evidence closes an Oracle DR licensing question early?

A configuration record that answers each condition before anyone asks. The conditions are objective, so you can assemble the evidence once and keep it current as the design changes.

Which four records should you be able to produce on request?

A physical standby's data dictionary is a copy of the primary's, so the primary's feature usage history is the best record of which options production uses and the standby must match. Review it the way you would review the feature usage report before any audit script runs.

How do you match your design to Oracle's published names?

The architecture is public. Oracle documents the configurations on the high availability pages, and the licensing treatment of each feature in the Oracle Database 19c edition of the licensing information manual. With each node described in Oracle's own terms, any disagreement is about published text, which is easier to settle.

What will Oracle say, and how should you answer?

• "The standby has been installed since the site was built, so it has needed a license since then." Agree with the principle, then check the install dates in your own change records before accepting any backdated support charge.
• "Every option used on the primary is required on the standby." The policy supports that, with the RAC exception. Ask for usage per option on the primary with first and last usage dates, and challenge anything triggered only by a default setting, because an option that production does not need is one the standby does not need either.
• "Your drills exceeded the ten day allowance." Produce your day log and ask which dates Oracle is counting, and from what evidence.
• "A wider agreement would cover the DR servers." Price the DR gap on its own first, so you know what the larger deal is replacing.

Which mistakes cost the most?

• Calling a second site's server the spare. It fails the single data center condition, so the full license was owed from installation.
• Running reports on the spare while it waits. One reporting job makes the node licensable for every day of the year.
• Patching through the spare every month. Twelve maintenance windows exceed the ten periods on their own.
• Building the standby on a shared virtualization cluster. Oracle's partitioning position counts every host the standby could run on, so the DR bill can exceed production's. The virtualization guide covers the containment options.
• Buying standby licenses on a different metric. Named User Plus on production and processor on the standby, or the reverse, breaks the matching rule even when the counts look generous.

What to do next

1. Classify every recovery node against the six conditions. The concession covers one node in one cluster in one data center, so list each standby, mirror and spare and mark which condition it fails.
2. Check the storage topology first. It is the quickest test to run, and it settles the status of most standbys and mirrors in one pass.
3. Keep a whole day log for every spare. Any part of a 24 hour period counts in full against the published ten, and maintenance windows count too.
4. Get the treatment of planned drills confirmed in writing. Design the runbook around the conservative reading until that reply exists.
5. Reconcile the option inventory on the standby against the primary. Match the metric as well, including the Named User Plus minimum on the standby hardware, and leave RAC off the standby only if it is never used there.
6. Assemble the configuration record before it is requested. The Oracle practice can build it with you and price any gap before Oracle does.

Frequently asked questions