perf: optimize EventHub cache mapping

[ReubenBond]

Problem

Issue #1742 identified unnecessary work when mapping Event Hubs EventData application properties into and out of the pooled stream cache. Orleans-produced messages normally contain only the internal StreamNamespace property, which is excluded from the cached application-property collection, but the adapter still serialized and deserialized an empty list for every message.

Solution

Represent an empty application-property collection as a zero-length cache segment and construct the mutable empty dictionary directly when reading it. For messages with application properties, populate the serialized list and reconstructed dictionary with allocation-conscious loops instead of LINQ.

The representation is internal to each silo's in-memory cache, so it introduces no durable wire-format compatibility requirement. Application properties, duplicate-key errors, and the mutable IDictionary contract are preserved.

Performance

BenchmarkDotNet on .NET 10 for the default no-application-properties path measured serialization improving from 117.6 ns / 128 B to 7.1 ns / 0 B, and deserialization improving from 74.6 ns / 112 B to 13.7 ns / 80 B. Messages with application properties retain equivalent throughput while the serialization-side temporary allocation drops from 280 B to 232 B for four properties.

Related to #1742.

Microsoft Reviewers: Open in CodeFlow

[Copilot]

Pull request overview

Optimizes Event Hubs application-property mapping in the in-memory stream cache to reduce allocations.

Changes:

• Uses zero-length cache segments for empty properties.
• Replaces LINQ mapping with allocation-conscious loops.
• Adds empty and round-trip mapping tests.

Show a summary per file

File	Description
test/Extensions/Orleans.Streaming.EventHubs.Tests/EventHubDataAdapterTests.cs	Tests property mapping behavior.
src/Azure/Orleans.Streaming.EventHubs/Providers/Streams/EventHub/EventDataExtensions.cs	Optimizes property serialization and deserialization.

Review details

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

• Files reviewed: 2/2 changed files
• Comments generated: 0
• Review effort level: Lite