perf(cache): remove per-operation allocations on the Redis and memory hot paths

[cosmin-staicu]

Stacked on #213. Review and merge that one first; this PR's base is feat/resolve-key-prefix.

Removes per-operation allocations on the Redis command path and the local memory path, in six changes:

1. Telemetry off still allocated. The caches call StartOperation(string, Type, string). NullTelemetryProvider never declared that overload, so the interface default built a TelemetryOperation, a Stopwatch and a scope string on every call. It now returns NullTelemetryOperation.Instance. TelemetryOperation also caches its metric names and times with Stopwatch timestamps.
2. Polly wrapper. Pipelines were keyed by (Type, object?), which boxed the default value on every command, and the GetOrAdd factory captured a closure. Now there is one typed map per result type, and the read path passes the callback to Polly as state.
3. Async lambdas. 43 single-command callbacks awaited the client task only to return it. They now wrap the task in a ValueTask.
4. Key composition. string.Join(char, params string[]) is replaced by interpolation in the prefix key strategies.
5. Comparers. TopicKey equality is now ordinal. Its name is already lowercased and it already hashes ordinally. The event-type sets and the hash field filter use OrdinalIgnoreCase, and the filter builds its result in a loop instead of LINQ with ToImmutableDictionary.
6. Small cleanups. Array.ConvertAll replaces Select().ToArray() over arrays, and the internal builders and setters are sealed.

Benchmark

A single-operation BenchmarkDotNet run with MemoryDiagnoser against a local Redis on net10.0. The baseline is #213's head. Time on the Redis rows is dominated by the round trip, so the allocation column is the signal.

Operation	Allocated before	Allocated after	Mean before	Mean after
Redis GetAsync	3724 B	3123 B	760 µs	882 µs
Redis SetAsync	3020 B	2376 B	864 µs	742 µs
Multilayer GetAsync, memory hit	32 B	32 B	288 ns	201 ns
Hash GetAsync(fields), memory hit	896 B	576 B	1827 ns	801 ns
Multilayer SetAsync	14523 B	14448 B	1.98 ms	1.71 ms

The multilayer SetAsync row is noise-bound. It also counts allocations from the background stream reader, and repeated runs of either build ranged from 13.2 KB to 16.0 KB.

Behavior to review

• Change 5. The known event types and hash field names now match with OrdinalIgnoreCase instead of InvariantCultureIgnoreCase. The event types are ASCII constants. Field names that differ only under culture rules no longer match.
• Change 3. A cancellation checked before the command now throws synchronously instead of returning a faulted task. Every call site awaits inside a try, and Polly produces the same outcome either way.

One shape for the single-command members

SonarQube flagged 10.8% duplicated lines on this PR's new code: the thirteen single-command members of RedisCache, RedisHashCache and RedisSetCache (ContainsAsync, RemoveAsync, TimeToLiveAsync, ExpireTimeAsync, the set's CountAsync, ContainsItemAsync, RemoveItemAsync, RemoveItemsAsync) shared one 20-line skeleton around the pipeline call, and this PR touched two lines inside each. They now forward to RedisCacheBase.RunAsync, a protected helper that runs the command under the pipeline, times it, tracks the hit and logs a failure. The TTL and expiration reads, with the pre-Redis-7 fallback from TTL to expiration, live in the base as KeyTimeToLiveAsync and KeyExpireTimeAsync over one SupportsExpireTime, and RedisHashCache reads a hash's fields in one ReadFields method instead of two copies of the loop. Behaviour is unchanged, including a cancelled token being logged and answered with the fallback, and the TTL and expiration reads still composing the key inside the command so a null key is logged and answered the same way (two tests per cache pin both). The one difference: on a server without EXPIRETIME the fallback runs under the ExpireTimeAsync operation alone rather than also under a TimeToLiveAsync one.

Verification

• dotnet build -c Release -warnaserror
• dotnet test: 4001 passed on net8.0 and net10.0, with the Redis integration tests on (RUN_REDIS_INTEGRATION_TESTS=1).

[github-actions]

🔎 Maintainer heads-up: automated triage flagged this PR as potentially material, so it may need a signed CLA in addition to the DCO sign-off.

Strong signals

• adds public API surface (PublicAPI.Unshipped.txt in src/UiPath.Caching.Abstractions)

This is advisory only — the bot does not decide. Please judge against the CLA criteria (material, product-critical, patent-sensitive, corporate contributor, broad commercial use). Note that thresholds can be gamed by splitting PRs, so use your judgement.

• If a CLA is needed → add the cla-required label (a contributor comment with signing steps is posted automatically).
• If it is not needed → replace needs-cla-review with cla-not-required so later pushes don't re-flag it.

[Copilot]

Copilot review overview

🟡 Changes recommended

String L1 keys can corrupt shared application cache entries, and telemetry caching can retain collectible consumer types indefinitely.

Get a fresh assessment by requesting another Copilot review.

Review effort: Balanced
Findings: 1 · 1

Open (2)

• Raw string cache keys can collide with application entries · New
• Type-keyed cache prevents collectible assemblies from unloading · New

What changed in this PR

Reduces allocations across Redis command execution, telemetry, key composition, and multilayer memory-cache hot paths.

Changes:

• Replaces async state machines and LINQ allocations with ValueTask wrappers and array conversions.
• Optimizes telemetry, resilience-pipeline caching, key composition, and comparisons.
• Uses string keys for local memory caching and updates corresponding tests.

File	Description
tests/​UiPath.Caching.Tests/​MultilayerHashCacheTests.cs	Updates memory-key expectations.
tests/​UiPath.Caching.Tests/​MultilayerCacheTryAddTests.cs	Updates TryAdd memory-key assertions.
tests/​UiPath.Caching.Tests/​MultilayerCacheTests.cs	Updates multilayer memory-key assertions.
tests/​UiPath.Caching.Tests/​MemoryCacheSetterTests.cs	Verifies string-key eviction behavior.
tests/​UiPath.Caching.Tests/​LocalCacheSetterTests.cs	Verifies string-key local caching.
src/​UiPath.Caching/​TaskObservation.cs	Adds task-to-ValueTask conversion.
src/​UiPath.Caching/​Redis/​ShardPrefixRedisKeyStrategy.cs	Avoids string.Join allocation.
src/​UiPath.Caching/​Redis/​RedisHashCache.cs	Removes redundant async state machines.
src/​UiPath.Caching/​Redis/​RedisCache.cs	Optimizes callbacks and array conversions.
src/​UiPath.Caching/​Redis/​PrefixRedisKeyStrategy.cs	Uses interpolation for key composition.
src/​UiPath.Caching/​PrefixCacheKeyStrategy.cs	Uses interpolation for prefixed keys.
src/​UiPath.Caching/​MultilayerHashCache.cs	Uses string L1 keys and ordinal filtering.
src/​UiPath.Caching/​MultilayerCache.cs	Optimizes arrays and string L1 keys.
src/​UiPath.Caching/​MemoryCacheSetter.cs	Stores entries under string keys.
src/​UiPath.Caching/​LocalMemorySetter.cs	Seals the local setter.
src/​UiPath.Caching/​HashLocalMemorySetter.cs	Seals the hash setter.
src/​UiPath.Caching/​HashCacheEntryBuilder.cs	Seals the hash builder.
src/​UiPath.Caching/​CacheEntryBuilder.cs	Seals the cache builder.
src/​UiPath.Caching/​Broadcast/​Redis/​RedisStreamsTopic.cs	Optimizes Redis stream callbacks.
src/​UiPath.Caching/​Broadcast/​Redis/​RedisPubSubTopic.cs	Optimizes Pub/Sub callbacks.
src/​UiPath.Caching/​Broadcast/​ChangeTokenFactory.cs	Uses ordinal event comparison.
src/​UiPath.Caching/​Broadcast/​CacheEventFactory.cs	Uses ordinal event comparison.
src/​UiPath.Caching.Queue/​TaskObservation.cs	Adds queue task conversion helper.
src/​UiPath.Caching.Queue/​RedisSetCache.cs	Optimizes Redis set callbacks.
src/​UiPath.Caching.Polly/​ResiliencePipelineWrapper.cs	Adds typed pipeline caches and state callbacks.
src/​UiPath.Caching.CloudEvents/​CloudCacheEventFactory.cs	Uses ordinal event comparison.
src/​UiPath.Caching.Abstractions/​Telemetry/​TelemetryOperation.cs	Caches metric names and timestamp state.
src/​UiPath.Caching.Abstractions/​Telemetry/​NullTelemetryProvider.cs	Adds allocation-free overload.
src/​UiPath.Caching.Abstractions/​PublicAPI.Unshipped.txt	Records the new public overload.
src/​UiPath.Caching.Abstractions/​CacheOfT.cs	Replaces LINQ array conversions.
src/​UiPath.Caching.Abstractions/​Broadcast/​TopicKey.cs	Makes equality ordinal.

---

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

[Copilot]

Copilot review overview

🟡 Changes recommended

The static telemetry metric-name cache can grow indefinitely from public high-cardinality inputs.

Get a fresh assessment by requesting another Copilot review.

Review effort: Balanced
Findings: 1

Open (1)

• Unbounded cache causes process-lifetime memory growth · New

Resolved since last review (2)

• Raw string cache keys can collide with application entries
• Type-keyed cache prevents collectible assemblies from unloading

[Copilot]

Copilot review overview

🟢 Approval recommended

The allocation optimizations preserve established behavior and are supported by comprehensive build and test verification.

Review effort: Balanced
Findings: None

Resolved since last review (1)

• Unbounded cache causes process-lifetime memory growth

[Copilot]

Copilot review overview

🟢 Approval recommended

The allocation optimizations are internally consistent, preserve the documented semantics, and introduce no unresolved correctness issues.

Review effort: Balanced
Findings: None

The base branch was changed.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
98.7% Coverage on New Code
2.6% Duplication on New Code

See analysis details on SonarQube Cloud

[Copilot]

Copilot review overview

🟡 Changes recommended

Base-class version detection now makes resolving RedisSetCache synchronously initiate and wait for a Redis connection.

Review effort: Balanced
Findings: 1

Open (1)

• Defer Redis version detection for caches that do not use expiration · New