Azure/azure-sdk-for-java com.azure+azure-cosmos_4.82.0

on GitHub

4.82.0 (2026-08-26)

Features Added

• Enabled Gateway V2 (thin-client) data-plane routing by default for Cosmos(Async)Client instances configured with gatewayMode and HTTP/2, gated by an HTTP/2 connectivity probe with automatic fallback to Gateway V1. - See PR 49437
• Added support for QueryPlan and Execute Stored Procedure requests to be routed to Gateway V2. - See PR 47759

Bugs Fixed

• Fixed a possible delay of PPCB failback when the SDK has never received a successful address response for the affected partition before failover, which could impact tail latency in certain workloads - See PR 50182.
• Fixed an intermittent IndexOutOfBoundsException in cross-partition hybrid search queries caused by multiple subscriptions to the coalesced component query results. - See PR 49831
• Fixed document requests failing when Gateway V2 is enabled with resource-token or permission-feed authentication by routing those requests through Compute Gateway. - See PR 50084.
• Unified request-level consistency override behavior across transports: invalid attempts to upgrade the request consistency level above the account default are now silently ignored instead of returning BadRequest in some gateway paths. - See PR 49606.
• Fixed partitionLevelCircuitBreakerCfg missing from the clientCfgs section of CosmosDiagnostics when Per-Partition Circuit Breaker is explicitly enabled. - See PR 49734.
• Fixed thin-client (Gateway V2) queries with a prefix (partial) hierarchical partition key returning co-located documents from other logical partitions. - See PR 49688.
• Fixed hedged requests losing request-scoped routing, timeout, authorization, throughput-control, and metadata state when cloning the original request. - See PR 50069.

Other Changes

• Fixed an issue in direct connectivity mode where 429 (Too Many Requests) responses during read/write barrier requests could cause excessive retries instead of yielding early. When all contacted replicas return 429 on consecutive barrier attempts, the SDK now propagates the throttle response (for reads) or returns a 408 with substatus 21013 (for writes) to allow the built-in ResourceThrottleRetryPolicy to handle backoff. This behavior is enabled by default; opt out by setting system property COSMOS.ENABLE_BARRIER_EARLY_YIELD_ON_429 or environment variable COSMOS_ENABLE_BARRIER_EARLY_YIELD_ON_429 to false. - See PR 48914
• Added per-region Per-Partition Circuit Breaker health and last failback outcome snapshots to CosmosDiagnostics, including structured failure reasons, and WARN logging for failback failures. - See PR 50158.
• Reduced memory footprint of deserialized PartitionKeyRange instances by stripping unused fields in the PartitionKeyRange(ObjectNode) constructor - See PR 49513.
• Added bounded retries for transient "collection routing map / partition key range metadata not available" responses (HTTP 404 with sub-status 0, 1003, or 1013) that can briefly occur right after a container is (re)created, improving the robustness of data-plane operations against the post-creation metadata-propagation race. As part of this change, when the routing map remains unavailable after retries an operation now fails with a CosmosException (HTTP 404, sub-status 1024 / INCORRECT_CONTAINER_RID) instead of an internal IllegalStateException. - See PR 49639.
• Reduced memory footprint and redundant /pkranges reads when multiple CosmosClient / CosmosAsyncClient instances in the same JVM are configured with the same service endpoint. Disable with system property COSMOS.SHARED_PARTITION_KEY_RANGE_CACHE_ENABLED=false if needed. - See PR 49560.