[Gateway V1/V2]: Enable ReadConsistencyStrategy for Gateway V1 and Gateway V2

[jeet1995]

Issue

Fixes #48094

Summary

Enable ReadConsistencyStrategy across all connection modes -- Direct (already working), Gateway V1 (compute gateway), and Gateway V2 (thin client proxy).

What changed

ReadConsistencyStrategy overrides ConsistencyLevel -- When a non-DEFAULT ReadConsistencyStrategy is set (at client or request level), ConsistencyLevel is stripped from the request. This prevents dual-header rejection by the compute gateway and proxy. Request-level RCS overrides client-level RCS.

GLOBAL_STRONG validation -- GLOBAL_STRONG is only valid on accounts with Strong as the default consistency level. The SDK now validates this client-side and throws BadRequestException at request time (fail-fast), matching the existing Direct mode behavior.

Key design decisions

Decision	Rationale
Strip ConsistencyLevel when RCS is set	Compute gateway rejects requests with both headers. RCS takes precedence per design (matches .NET SDK behavior).
DEFAULT is transparent -- never written as a header	ReadConsistencyStrategy.DEFAULT means "use the account default". The SDK filters it out at the source (query/changefeed header-building) and defensively in downstream methods (isEffectiveSessionConsistency, resolveEffectiveConsistencyHeaders). This prevents "Default" from reaching the wire as a header string, which would break session-token logic (false negative on Session accounts) and cause dual-header rejection on Gateway V1.
GLOBAL_STRONG validation is client-side	Fail-fast at request time, not after round-trip to server.
RNTBD token ID 0x00FE, Byte type	Matches proxy RntbdConstants.cs L652 and proxy PR #2031635. Proxy expects C++ enum (int), not string.

Test coverage

RntbdReadConsistencyStrategyHeaderTests
Unit tests for the RNTBD ReadConsistencyStrategy token (ID 0x00FE, Byte type). Validates token metadata, byte encoding per strategy, encode/decode round-trip symmetry, and the full resolveEffectiveConsistencyHeaders -> wrapInHttpRequest pipeline for contention scenarios (both ConsistencyLevel and ReadConsistencyStrategy set). Uses RntbdRequest.decode() to verify the RNTBD frame produced by ThinClientStoreModel.

GatewayReadConsistencyStrategySpyWireTest
Spy wire tests that intercept the final HTTP request at the HTTP client layer for both Gateway V1 (HTTP/1) and Gateway V2 (HTTP/2 + thin client). V1 asserts HTTP headers; V2 decodes the RNTBD frame via RntbdRequest.decode() and validates :10250 endpoint routing. Covers no-contention, contention resolution, request-level overrides client-level, DEFAULT transparency, and write operations.

GatewayReadConsistencyStrategyE2ETest
End-to-end tests against a real account with two CosmosAsyncClient instances (Gateway V1 and Gateway V2). Validates that the server accepts and applies ReadConsistencyStrategy correctly across all request option surfaces: point reads, queries, readAll, changeFeed, readMany, client-level defaults, write operations, GLOBAL_STRONG validation, contention resolution, and operation policy overrides. V2 tests assert thin client endpoint usage via CosmosDiagnostics. Direct mode tests additionally assert quorumAckedLSN in diagnostics for LatestCommitted reads.

Tested against a Session consistency account (test4acct) on a thin-client-enabled federation (cdb-ms-test4-northeurope1-fe1).

Smoke test (Kusto validation) -- quorum reads for LatestCommitted

Ran 15 requests each for Read, Query, and ChangeFeed with ReadConsistencyStrategy.LATEST_COMMITTED across GW V1 and GW V2. Verified 1:2 client-to-backend ratio (quorum reads) via Kusto telemetry.

BackendEndRequest5M (proxy/compute → backend replicas):

BackendEndRequest5M
| where TIMESTAMP >= datetime(2026-04-29T16:20) and TIMESTAMP <= datetime(2026-04-29T16:30)
| where GlobalDatabaseAccountName == 'test4acct'
| where Res(ResourceType) == "Document"
| where Op(OperationType) in ("Read", "ReadFeed", "Query")
| extend Mode = iif(UserAgent contains "ThinProxy", "GW_V2", "GW_V1")
| summarize BackendReads = sum(SampleCount) by OperationType, Mode, StatusCode, ConsistencyLevel, DefaultConsistencyLevel

Op (OperationType)	Mode	StatusCode	BackendReads	ConsistencyLevel	DefaultConsistencyLevel
2 (Read)	GW_V1	200	30	0 (not set)	Session
2 (Read)	GW_V2	200	30	0 (not set)	Session
15 (Query)	GW_V1	200	30	0 (not set)	Session
15 (Query)	GW_V2	200	30	0 (not set)	Session
3 (ReadFeed)	GW_V1	200	30	0 (not set)	Session
3 (ReadFeed)	GW_V1	304	30	0 (not set)	Session
3 (ReadFeed)	GW_V2	200	30	0 (not set)	Session
3 (ReadFeed)	GW_V2	304	30	0 (not set)	Session

ThinClientProxyRequest5M (client → proxy, GW V2):

ThinClientProxyRequest5M
| where TIMESTAMP >= datetime(2026-04-29T16:20) and TIMESTAMP <= datetime(2026-04-29T16:30)
| where globalDatabaseAccountName == 'test4acct'
| where resourceType == "Document"
| where operationType in ("Read", "ReadFeed", "Query")
| summarize ClientReads = sum(SampleCount) by operationType, statusCode

operationType	statusCode	ClientReads
Read	200	15
Query	200	15
ReadFeed	200	15
ReadFeed	304	15

ComputeRequest5M (client → compute, GW V1):

ComputeRequest5M
| where TIMESTAMP >= datetime(2026-04-29T16:20) and TIMESTAMP <= datetime(2026-04-29T16:30)
| where GlobalDatabaseAccountName == 'test4acct'
| where ResourceName == "Document"
| where OperationName in ("Read", "Query", "ChangeFeed/Incremental")
| summarize ClientReads = sum(SampleCount) by OperationName, StatusCode

OperationName	StatusCode	ClientReads
Read	200	15
Query	200	15
ChangeFeed/Incremental	200	15
ChangeFeed/Incremental	304	15

Quorum validation summary:

Operation	Client requests (per mode)	Backend reads (per mode)	Ratio
Read	15	30	1:2 ✅
Query	15	30	1:2 ✅
ChangeFeed	30 (15×200 + 15×304)	60 (30×200 + 30×304)	1:2 ✅

ConsistencyLevel=0 (not set) on all backend entries confirms x-ms-consistency-level is correctly stripped when ReadConsistencyStrategy is active.

.NET SDK reference

PR #5685 -- .NET SDK ReadConsistencyStrategy implementation. HTTP header name and enum values match.

Proxy coordination

ADO PR #2031635 -- "Add ReadConsistencyStrategy for Proxy". Merged and deployed to test4 federations.

---

SDK Contribution checklist:

• The pull request does not introduce [breaking changes]
• CHANGELOG is updated for new features, bug fixes or other significant changes.
• I have read the contribution guidelines.

Testing Guidelines

• Pull request includes test coverage for the included changes.

[FabianMeiswinkel]

LGTM

[jeet1995]

/azp run java - cosmos - tests

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jeet1995]

/azp run java - cosmos - tests

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jeet1995]

/azp run java - cosmos - tests

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jeet1995]

@sdkReviewAgent

[xinlian12]

✅ Review complete (43:19)

Posted 6 inline comment(s).

_{Steps: ✓ context, correctness, cross-sdk, design, history, past-prs, synthesis, test-coverage}