Performance | Introduce lower-allocation AlwaysEncrypted primitives #3554

edwardneal · 2025-08-09T15:52:10Z

Description

I'm trying to improve query performance in AE scenarios, and part of this means a review of the column encryption providers. These were written prior to a number of cryptography improvements in .NET 5.0; as a result of these improvements, we can save a little memory.

Besides the performance improvements, each provider reproduces the same serialization logic. These primitives centralise that logic; they use the column master key parameters to perform encryption and decryption of a column encryption key with an RSA instance. The only thing that the three SqlColumnEncryptionKeyStoreProvider derivatives need to do is retrieve an RSA instance using the master key path.

I considered moving this logic into a protected method on SqlColumnEncryptionKeyStoreProvider, but a third-party provider isn't guaranteed to perform encryption/decryption of the CEK using an RSA instance (or indeed, on the same server - SqlColumnEncryptionAzureKeyVaultProvider passes that duty to Azure Key Vault.)

A subsequent PR will refactor the column encryption providers to use these primitives, and the benchmark results for this are below. To summarise: no significant variations in execution time, and around a 15% reduction in AlwaysEncrypted memory usage.

Benchmarks

`SqlColumnEncryptionCertificateStoreProvider`

Method	Branch	Mean	Error	StdDev	Ratio	Allocated	Alloc Ratio
DecryptColumnEncryptionKey	main	13.79 ms	0.256 ms	0.227 ms	1.00	18.17 KB	1.00
DecryptColumnEncryptionKey	PR	13.16 ms	0.180 ms	0.150 ms	0.95	15.43 KB	0.85
EncryptColumnEncryptionKey	main	12.59 ms	0.224 ms	0.375 ms	1.00	18.94 KB	1.00
EncryptColumnEncryptionKey	PR	12.58 ms	0.246 ms	0.450 ms	1.00	16.01 KB	0.85

`SqlColumnEncryptionCngProvider`

Method	Branch	Mean	Error	StdDev	Ratio	Allocated	Alloc Ratio
DecryptColumnEncryptionKey	main	3.303 ms	0.0639 ms	0.1661 ms	1.00	2.51 KB	1.00
DecryptColumnEncryptionKey	PR	3.285 ms	0.0648 ms	0.0842 ms	0.99	901 B	0.36
EncryptColumnEncryptionKey	main	3.214 ms	0.0395 ms	0.0330 ms	1.00	3.11 KB	1.00
EncryptColumnEncryptionKey	PR	3.332 ms	0.0645 ms	0.0603 ms	1.03	1.61 KB	0.52

`SqlColumnEncryptionCspProvider`

Method	Branch	Mean	Error	StdDev	Ratio	Allocated	Alloc Ratio
DecryptColumnEncryptionKey	main	2.775 ms	0.0411 ms	0.0520 ms	1.00	3.59 KB	1.00
DecryptColumnEncryptionKey	PR	2.737 ms	0.0256 ms	0.0200 ms	0.99	3.01 KB	0.84
EncryptColumnEncryptionKey	main	2.815 ms	0.0556 ms	0.0571 ms	1.00	4.01 KB	1.00
EncryptColumnEncryptionKey	PR	2.847 ms	0.0464 ms	0.0434 ms	1.01	3.23 KB	0.81

Testing

These primitives aren't used anywhere yet, so there's no test coverage. The existing providers have this coverage though, so the new code will gain coverage from the follow-up PR. I have a local branch where the column encryption providers use this code, and almost all tests pass. The two failing tests are capitalisation errors in exception messages produced by the CSP provider, which will be fixed at the same time.

Wraith2 · 2025-08-11T22:50:39Z

...osoft.Data.SqlClient/src/Microsoft/Data/SqlClient/AlwaysEncrypted/ColumnMasterKeyMetadata.cs

+        providerName.AsSpan().ToLowerInvariant(masterKeyMetadataSpan);
+        masterKeyPath.AsSpan().ToLowerInvariant(masterKeyMetadataSpan.Slice(providerName.Length));


Are the characters in the input variable restricted such that it is certain that the output lower invariant replacement version is certain to be of exactly the same length?

Yes, indirectly. This is due to be used by the certificate store provider, and the constructor takes an RSA instance. The only way to get this instance is by using the master key path to select a valid certificate, and the only valid master key paths are ASCII text in a few formats:

[LocalMachine|CurrentUser]/My/[40 character SHA1 thumbprint]

My/[40 character SHA1 thumbprint]

[40 character SHA1 thumbprint]

Excellent. It might be worth leaving a comment in the code somewhere so anyone who reads the code in future knows that there are input restrictions.

Add a comment and a debug assertion to make it clear that the invariant lowercase value of masterKeyPath should be the same length as the original string.

paulmedynski · 2025-08-27T12:55:56Z

/azp run

azure-pipelines · 2025-08-27T12:56:19Z

Azure Pipelines successfully started running 2 pipeline(s).

paulmedynski

Consolidation looks good overall, with some questions/suggestions.

...osoft.Data.SqlClient/src/Microsoft/Data/SqlClient/AlwaysEncrypted/ColumnMasterKeyMetadata.cs

...lient/src/Microsoft/Data/SqlClient/AlwaysEncrypted/EncryptedColumnEncryptionKeyParameters.cs

Add/updated XML documentation to the public APIs. Add intermediary variable containing the signature size for clarity. Use AlgorithmOffset rather than hardcoded offset in EncryptedColumnEncryptionKeyParameters. Specifically define using block in ColumnMasterKeyMetadata.

class -> struct

paulmedynski · 2025-08-28T11:08:10Z

/azp run

azure-pipelines · 2025-08-28T11:08:32Z

Azure Pipelines successfully started running 2 pipeline(s).

paulmedynski

Thanks for the previous updates. Just a couple more and one question about disposal.

...osoft.Data.SqlClient/src/Microsoft/Data/SqlClient/AlwaysEncrypted/ColumnMasterKeyMetadata.cs

...lient/src/Microsoft/Data/SqlClient/AlwaysEncrypted/EncryptedColumnEncryptionKeyParameters.cs

...osoft.Data.SqlClient/src/Microsoft/Data/SqlClient/AlwaysEncrypted/ColumnMasterKeyMetadata.cs

edwardneal · 2025-08-28T18:20:40Z

Thanks. Responded, and merged following the merge of #3014.

paulmedynski

Thanks for the updates! RSA.SignHash() will throw CryptographicException if "An error occurred creating the signature.", so need to document that one as well.

edwardneal · 2025-08-28T22:47:18Z

I'd missed that - added to ColumnMasterKeyMetadata.Sign and EncryptedColumnEncryptionKeyParameters.Encrypt.

paulmedynski · 2025-08-29T12:16:04Z

/azp run

azure-pipelines · 2025-08-29T12:16:26Z

Azure Pipelines successfully started running 2 pipeline(s).

Introduce low-allocation AlwaysEncrypted primitives

a9308cf

edwardneal requested a review from a team as a code owner August 9, 2025 15:52

edwardneal changed the title ~~Performance | Introduce low-allocation AlwaysEncrypted primitives~~ Performance | Introduce lower-allocation AlwaysEncrypted primitives Aug 9, 2025

Wraith2 reviewed Aug 11, 2025

View reviewed changes

Respond to code review

e30ccae

Add a comment and a debug assertion to make it clear that the invariant lowercase value of masterKeyPath should be the same length as the original string.

paulmedynski self-assigned this Aug 27, 2025

paulmedynski requested changes Aug 27, 2025

View reviewed changes

apoorvdeshmukh self-assigned this Aug 27, 2025

edwardneal added 2 commits August 27, 2025 21:12

Typo correction

7defd2f

class -> struct

paulmedynski requested changes Aug 28, 2025

View reviewed changes

edwardneal added 2 commits August 28, 2025 19:13

Review and add exceptions to XML documentation

a311495

Merge main

7990d0a

paulmedynski requested changes Aug 28, 2025

View reviewed changes

Add additional instances where CryptographicException can be thrown

cb143aa

paulmedynski approved these changes Aug 29, 2025

View reviewed changes

		providerName.AsSpan().ToLowerInvariant(masterKeyMetadataSpan);
		masterKeyPath.AsSpan().ToLowerInvariant(masterKeyMetadataSpan.Slice(providerName.Length));

Performance | Introduce lower-allocation AlwaysEncrypted primitives #3554

Are you sure you want to change the base?

Performance | Introduce lower-allocation AlwaysEncrypted primitives #3554

Uh oh!

Conversation

edwardneal commented Aug 9, 2025

Description

Benchmarks

SqlColumnEncryptionCertificateStoreProvider

SqlColumnEncryptionCngProvider

SqlColumnEncryptionCspProvider

Testing

Uh oh!

Wraith2 Aug 11, 2025

Choose a reason for hiding this comment

Uh oh!

edwardneal Aug 12, 2025

Choose a reason for hiding this comment

Uh oh!

Wraith2 Aug 12, 2025

Choose a reason for hiding this comment

Uh oh!

paulmedynski commented Aug 27, 2025

Uh oh!

azure-pipelines bot commented Aug 27, 2025

Uh oh!

paulmedynski left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

paulmedynski commented Aug 28, 2025

Uh oh!

azure-pipelines bot commented Aug 28, 2025

Uh oh!

paulmedynski left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

edwardneal commented Aug 28, 2025

Uh oh!

paulmedynski left a comment

Choose a reason for hiding this comment

Uh oh!

edwardneal commented Aug 28, 2025

Uh oh!

paulmedynski commented Aug 29, 2025

Uh oh!

azure-pipelines bot commented Aug 29, 2025

Uh oh!

Uh oh!

`SqlColumnEncryptionCertificateStoreProvider`

`SqlColumnEncryptionCngProvider`

`SqlColumnEncryptionCspProvider`