0

Microsoft documentation for Read-Scale (clusterless) failover: "If the primary replica is not available and can't immediately be recovered, then you need to force a failover to the secondary replica with data loss."

https://learn.microsoft.com/en-us/sql/database-engine/availability-groups/windows/configure-read-scale-availability-groups?view=sql-server-ver16

Question: How is data loss possible after failover if we have sync-commit mode configured and primary becomes unavailable? Databases are SYNCHRONIZED before failover.

qqq
  • 1
  • 1
  • If you read through the documentation in order, you'll see that earlier they imply you would be running in async to all secondaries _most of the time_, and only switch to sync to a specific secondary before a _planned_ failover. The section about forced manual failover with data loss is describing the scenario where you were in async and the primary disappeared. You didn't plan for that and wouldn't have had everything in sync. I don't know if running read-scale in sync all the time is supported - the docs create them as async for a reason, I assume. – Aaron Bertrand May 28 '23 at 22:12
  • Yes, that makes sense. Read-Scale sync-commit is supported, but they probably wanted to describe Read-Scale as non-HA solution. I need Read-Scale for synchronization between FCI and Standalone instance, and because of that I will use SYNC-COMMIT mode with manual failover without data loss. Thanks. – qqq May 28 '23 at 23:08
  • I mean, it's right at the top of the doc: `There are two types of architecture for availability groups: An architecture for high availability that uses a cluster manager...` and `An architecture that supports only read-scale workloads.` e.g. the latter is not meant for HA. – Aaron Bertrand May 28 '23 at 23:14
  • Yes, you are right. – qqq May 28 '23 at 23:47

0 Answers0