Looking for replicas for search documentation

[rafalbroll-datapharm]

Hello,

In the recent release i've seen there's a replicas for search feature enabled
https://github.com/microsoft/fhir-server/releases/tag/release/4.0.51

Is there any documentation which explains that topic? I'm wondering if that has something to do with having multiple instances to increase read operations performance and volume.

[brendankowitz]

While the code is there, we're working through some details on rolling out this capability.
Most of the changes where in #3663 and under the hood aims to set the ApplicationIntent=ReadOnly; which is explained here: https://learn.microsoft.com/en-us/azure/azure-sql/database/service-tier-hyperscale-replicas?view=azuresql&tabs=portal#connect-to-an-ha-replica

The main consideration is that the replicas aren't 100% guaranteed to have the absolute latest data which will be something that the app/business must consider what is acceptable.
To handle some of this, the part that will need documentation is the logic that selects when to use the replicas:

fhir-server/src/Microsoft.Health.Fhir.SqlServer/Features/Storage/SqlRetry/SqlRetryService.cs

Line 430 in 869dbab

var useWriteConnection = unchecked(Interlocked.Increment(ref _usageCounter)) % (int)(1 / (1 - _replicaTrafficRatio)) == 1; // examples for ratio -> % divider = { 0.9 -> 10, 0.8 -> 5, 0.75 - 4, 0.67 - 3, 0.5 -> 2, <0.5 -> 1}

Which is currently controlled by a DB setting:

fhir-server/src/Microsoft.Health.Fhir.SqlServer/Features/Storage/SqlRetry/SqlRetryService.cs

Line 474 in 869dbab

using var cmd = new SqlCommand("IF object_id('dbo.Parameters') IS NOT NULL SELECT Number FROM dbo.Parameters WHERE Id = 'ReplicaTrafficRatio'", conn);

[rafalbroll-datapharm]

Thank you for the explanation, Brendan.

One more thing: If i get it right, the FHIR handling is all about picking the right connection for doing the job, however the actual synchronisation between replicas is going to be achieved using native SQL mechanism, correct?

[brendankowitz]

Correct, the underlying capability in the docs above is provided by Hyperscale