Bail out from barriers when barriers hit 410 Lease Not Found.

sdk/cosmos/azure-cosmos-tests/pom.xml

@@ -906,10 +906,10 @@ Licensed under the MIT License.
       </build>
     </profile>
     <profile>
-      <!-- integration tests, requires Cosmos DB endpoint -->
-      <id>fault-injection-barrier</id>
+      <!-- tests which target a multi-region strong Cosmos DB account -->
+      <id>multi-region-strong</id>
       <properties>
-        <test.groups>fault-injection-barrier</test.groups>
+        <test.groups>multi-region-strong</test.groups>
       </properties>
       <build>
         <plugins>
@@ -919,9 +919,14 @@ Licensed under the MIT License.
             <version>3.5.3</version> <!-- {x-version-update;org.apache.maven.plugins:maven-failsafe-plugin;external_dependency} -->
             <configuration>
               <suiteXmlFiles>
-                <suiteXmlFile>src/test/resources/fault-injection-barrier-testng.xml</suiteXmlFile>
+                <suiteXmlFile>src/test/resources/multi-region-strong.xml</suiteXmlFile>
               </suiteXmlFiles>
-
+              <systemPropertyVariables>
+                <COSMOS.CLIENT_LEAK_DETECTION_ENABLED>true</COSMOS.CLIENT_LEAK_DETECTION_ENABLED>
+                <io.netty.leakDetection.samplingInterval>1</io.netty.leakDetection.samplingInterval>
+                <io.netty.leakDetection.targetRecords>256</io.netty.leakDetection.targetRecords>
+                <io.netty.leakDetection.level>paranoid</io.netty.leakDetection.level>
+              </systemPropertyVariables>
             </configuration>
           </plugin>
         </plugins>

sdk/cosmos/azure-cosmos-tests/src/test/java/com/azure/cosmos/faultinjection/FaultInjectionServerErrorRuleOnDirectTests.java

@@ -84,7 +84,7 @@ public FaultInjectionServerErrorRuleOnDirectTests(CosmosClientBuilder clientBuil
         this.subscriberValidationTimeout = TIMEOUT;
     }
 
-    @BeforeClass(groups = {"multi-region", "long", "fast", "fi-multi-master", "fault-injection-barrier"}, timeOut = TIMEOUT)
+    @BeforeClass(groups = {"multi-region", "long", "fast", "fi-multi-master", "multi-region-strong"}, timeOut = TIMEOUT)
     public void beforeClass() {
         clientWithoutPreferredRegions = getClientBuilder()
             .preferredRegions(new ArrayList<>())
@@ -1057,7 +1057,7 @@ public void faultInjectionServerErrorRuleTests_HitLimit() throws JsonProcessingE
         }
     }
 
-    @AfterClass(groups = {"multi-region", "long", "fast", "fi-multi-master", "fault-injection-barrier"}, timeOut = SHUTDOWN_TIMEOUT, alwaysRun = true)
+    @AfterClass(groups = {"multi-region", "long", "fast", "fi-multi-master", "multi-region-strong"}, timeOut = SHUTDOWN_TIMEOUT, alwaysRun = true)
     public void afterClass() {
         safeClose(clientWithoutPreferredRegions);
     }
@@ -1475,7 +1475,7 @@ public void faultInjectionInjectTcpResponseDelay() throws JsonProcessingExceptio
         }
     }
 
-    @Test(groups = {"fault-injection-barrier"}, dataProvider = "barrierRequestServerErrorResponseProvider", timeOut = 2 * TIMEOUT)
+    @Test(groups = {"multi-region-strong"}, dataProvider = "barrierRequestServerErrorResponseProvider", timeOut = 2 * TIMEOUT)
     public void faultInjection_serverError_barrierRequest(
         OperationType operationType,
         FaultInjectionServerErrorType serverErrorType,

sdk/cosmos/azure-cosmos-tests/src/test/java/com/azure/cosmos/implementation/directconnectivity/ReflectionUtils.java

@@ -309,6 +309,10 @@ public static StoreReader getStoreReader(ConsistencyReader consistencyReader) {
         return get(StoreReader.class, consistencyReader, "storeReader");
     }
 
+    public static StoreReader getStoreReader(ConsistencyWriter consistencyWriter) {
+        return get(StoreReader.class, consistencyWriter, "storeReader");
+    }
+
     public static void setStoreReader(ConsistencyReader consistencyReader, StoreReader storeReader) {
         set(consistencyReader, storeReader, "storeReader");
     }

sdk/cosmos/azure-cosmos-tests/src/test/java/com/azure/cosmos/implementation/directconnectivity/StoreResponseInterceptorUtils.java

@@ -0,0 +1,149 @@
+// Copyright (c) Microsoft Corporation. All rights reserved.
+// Licensed under the MIT License.
+
+package com.azure.cosmos.implementation.directconnectivity;
+
+import com.azure.cosmos.ConsistencyLevel;
+import com.azure.cosmos.implementation.OperationType;
+import com.azure.cosmos.implementation.RxDocumentServiceRequest;
+import com.azure.cosmos.implementation.Utils;
+
+import java.util.concurrent.atomic.AtomicInteger;
+import java.util.function.BiFunction;
+
+public class StoreResponseInterceptorUtils {
+
+    public static BiFunction<RxDocumentServiceRequest, StoreResponse, StoreResponse> forceBarrierFollowedByBarrierFailure(
+        ConsistencyLevel operationConsistencyLevel,
+        String regionName,
+        int maxAllowedFailureCount,
+        AtomicInteger failureCount,
+        int statusCode,
+        int subStatusCode) {
+
+        return (request, storeResponse) -> {
+
+            if (OperationType.Create.equals(request.getOperationType()) && regionName.equals(request.requestContext.regionalRoutingContextToRoute.getGatewayRegionalEndpoint().toString())) {
+
+                long localLsn = Long.parseLong(storeResponse.getHeaderValue(WFConstants.BackendHeaders.LOCAL_LSN));
+                long manipulatedGCLSN = localLsn - 1;
+
+                storeResponse.setHeaderValue(WFConstants.BackendHeaders.GLOBAL_COMMITTED_LSN, String.valueOf(manipulatedGCLSN));
+
+                return storeResponse;
+            }
+
+            if (OperationType.Read.equals(request.getOperationType()) && regionName.equals(request.requestContext.regionalRoutingContextToRoute.getGatewayRegionalEndpoint().toString())) {
+
+                if (ConsistencyLevel.STRONG.equals(operationConsistencyLevel)) {
+
+                    long globalLsn = Long.parseLong(storeResponse.getHeaderValue(WFConstants.BackendHeaders.LSN));
+                    long itemLsn = Long.parseLong(storeResponse.getHeaderValue(WFConstants.BackendHeaders.ITEM_LSN));
+                    long manipulatedGlobalCommittedLSN = Math.min(globalLsn, itemLsn) - 1;
+
+                    storeResponse.setHeaderValue(WFConstants.BackendHeaders.GLOBAL_COMMITTED_LSN, String.valueOf(manipulatedGlobalCommittedLSN));
+
+                    return storeResponse;
+                } else if (ConsistencyLevel.BOUNDED_STALENESS.equals(operationConsistencyLevel)) {
+
+                    long manipulatedItemLSN = -1;
+                    long manipulatedGlobalLSN = 0;
+
+                    storeResponse.setHeaderValue(WFConstants.BackendHeaders.LSN, String.valueOf(manipulatedGlobalLSN));
+                    storeResponse.setHeaderValue(WFConstants.BackendHeaders.LOCAL_LSN, String.valueOf(manipulatedGlobalLSN));
+                    storeResponse.setHeaderValue(WFConstants.BackendHeaders.ITEM_LSN, String.valueOf(manipulatedItemLSN));
+                    storeResponse.setHeaderValue(WFConstants.BackendHeaders.ITEM_LOCAL_LSN, String.valueOf(manipulatedItemLSN));
+
+                    return storeResponse;
+                }
+
+                return storeResponse;
+            }
+
+            if (OperationType.Head.equals(request.getOperationType()) && regionName.equals(request.requestContext.regionalRoutingContextToRoute.getGatewayRegionalEndpoint().toString())) {
+                if (failureCount.incrementAndGet() <= maxAllowedFailureCount) {
+                    throw Utils.createCosmosException(statusCode, subStatusCode, new Exception("An intercepted exception occurred. Check status and substatus code for details."), null);
+                }
+            }
+
+            return storeResponse;
+        };
+    }
+
+    public static BiFunction<RxDocumentServiceRequest, StoreResponse, StoreResponse> forceSuccessfulBarriersOnReadUntilQuorumSelectionThenForceBarrierFailures(
+        ConsistencyLevel operationConsistencyLevel,
+        String regionName,
+        int allowedSuccessfulHeadRequestsWithoutBarrierBeingMet,
+        AtomicInteger successfulHeadRequestCount,
+        int maxAllowedFailureCount,
+        AtomicInteger failureCount,
+        int statusCode,
+        int subStatusCode) {
+
+        return (request, storeResponse) -> {
+
+            if (OperationType.Read.equals(request.getOperationType()) && regionName.equals(request.requestContext.regionalRoutingContextToRoute.getGatewayRegionalEndpoint().toString())) {
+
+                if (ConsistencyLevel.STRONG.equals(operationConsistencyLevel)) {
+
+                    long globalLsn = Long.parseLong(storeResponse.getHeaderValue(WFConstants.BackendHeaders.LSN));
+                    long itemLsn = Long.parseLong(storeResponse.getHeaderValue(WFConstants.BackendHeaders.ITEM_LSN));
+                    long manipulatedGlobalCommittedLSN = Math.min(globalLsn, itemLsn) - 1;
+
+                    storeResponse.setHeaderValue(WFConstants.BackendHeaders.GLOBAL_COMMITTED_LSN, String.valueOf(manipulatedGlobalCommittedLSN));
+
+                    return storeResponse;
+                } else if (ConsistencyLevel.BOUNDED_STALENESS.equals(operationConsistencyLevel)) {
+
+                    long manipulatedItemLSN = -1;
+                    long manipulatedGlobalLSN = 0;
+
+                    storeResponse.setHeaderValue(WFConstants.BackendHeaders.LSN, String.valueOf(manipulatedGlobalLSN));
+                    storeResponse.setHeaderValue(WFConstants.BackendHeaders.LOCAL_LSN, String.valueOf(manipulatedGlobalLSN));
+                    storeResponse.setHeaderValue(WFConstants.BackendHeaders.ITEM_LSN, String.valueOf(manipulatedItemLSN));
+                    storeResponse.setHeaderValue(WFConstants.BackendHeaders.ITEM_LOCAL_LSN, String.valueOf(manipulatedItemLSN));
+
+                    return storeResponse;
+                }
+
+                return storeResponse;
+            }
+
+            if (OperationType.Head.equals(request.getOperationType()) && regionName.equals(request.requestContext.regionalRoutingContextToRoute.getGatewayRegionalEndpoint().toString())) {
+
+                if (successfulHeadRequestCount.incrementAndGet() <= allowedSuccessfulHeadRequestsWithoutBarrierBeingMet) {
+
+                    if (ConsistencyLevel.STRONG.equals(operationConsistencyLevel)) {
+
+                        long localLsn = Long.parseLong(storeResponse.getHeaderValue(WFConstants.BackendHeaders.LOCAL_LSN));
+                        long itemLsn = Long.parseLong(storeResponse.getHeaderValue(WFConstants.BackendHeaders.ITEM_LSN));
+                        long manipulatedGCLSN = Math.min(localLsn, itemLsn) - 1;
+
+                        storeResponse.setHeaderValue(WFConstants.BackendHeaders.GLOBAL_COMMITTED_LSN, String.valueOf(manipulatedGCLSN));
+
+                        return storeResponse;
+                    } else if (ConsistencyLevel.BOUNDED_STALENESS.equals(operationConsistencyLevel)) {
+
+                        long manipulatedItemLSN = -1;
+                        long manipulatedGlobalLSN = -1;
+
+                        storeResponse.setHeaderValue(WFConstants.BackendHeaders.LSN, String.valueOf(manipulatedGlobalLSN));
+                        storeResponse.setHeaderValue(WFConstants.BackendHeaders.LOCAL_LSN, String.valueOf(manipulatedGlobalLSN));
+                        storeResponse.setHeaderValue(WFConstants.BackendHeaders.ITEM_LSN, String.valueOf(manipulatedItemLSN));
+                        storeResponse.setHeaderValue(WFConstants.BackendHeaders.ITEM_LOCAL_LSN, String.valueOf(manipulatedItemLSN));
+
+                        return storeResponse;
+                    }
+
+                    return storeResponse;
+                }
+
+                if (failureCount.incrementAndGet() <= maxAllowedFailureCount) {
+                    throw Utils.createCosmosException(statusCode, subStatusCode, new Exception("An intercepted exception occurred. Check status and substatus code for details."), null);
+                }
+            }
+
+            return storeResponse;
+        };
+    }
+}

sdk/cosmos/azure-cosmos-tests/src/test/java/com/azure/cosmos/rx/TestSuiteBase.java

@@ -205,7 +205,7 @@ public CosmosAsyncDatabase getDatabase(String id) {
 
     @BeforeSuite(groups = {"thinclient", "fast", "long", "direct", "multi-region", "multi-master", "flaky-multi-master", "emulator",
         "emulator-vnext", "split", "query", "cfp-split", "circuit-breaker-misc-gateway", "circuit-breaker-misc-direct",
-        "circuit-breaker-read-all-read-many", "fi-multi-master", "long-emulator", "fi-thinclient-multi-region", "fi-thinclient-multi-master", "fault-injection-barrier"}, timeOut = SUITE_SETUP_TIMEOUT)
+        "circuit-breaker-read-all-read-many", "fi-multi-master", "long-emulator", "fi-thinclient-multi-region", "fi-thinclient-multi-master", "multi-region-strong"}, timeOut = SUITE_SETUP_TIMEOUT)
     public void beforeSuite() {
 
         logger.info("beforeSuite Started");
@@ -223,7 +223,7 @@ public void beforeSuite() {
 
     @AfterSuite(groups = {"thinclient", "fast", "long", "direct", "multi-region", "multi-master", "flaky-multi-master",
         "emulator", "split", "query", "cfp-split", "circuit-breaker-misc-gateway", "circuit-breaker-misc-direct",
-        "circuit-breaker-read-all-read-many", "fi-multi-master", "long-emulator", "fi-thinclient-multi-region", "fi-thinclient-multi-master", "fault-injection-barrier"}, timeOut = SUITE_SHUTDOWN_TIMEOUT)
+        "circuit-breaker-read-all-read-many", "fi-multi-master", "long-emulator", "fi-thinclient-multi-region", "fi-thinclient-multi-master", "multi-region-strong"}, timeOut = SUITE_SHUTDOWN_TIMEOUT)
     public void afterSuite() {
 
         logger.info("afterSuite Started");

sdk/cosmos/azure-cosmos-tests/src/test/resources/fault-injection-barrier-testng.xml → sdk/cosmos/azure-cosmos-tests/src/test/resources/multi-region-strong.xml

@@ -21,14 +21,14 @@
   ~ SOFTWARE.
   -->
 <!DOCTYPE suite SYSTEM "https://testng.org/testng-1.0.dtd">
-<suite name="fault-injection-barrier">
+<suite name="multi-region-strong">
   <listeners>
     <listener class-name="com.azure.cosmos.CosmosNettyLeakDetectorFactory"/>
   </listeners>
-  <test name="fault-injection-barrier" group-by-instances="true">
+  <test name="multi-region-strong" group-by-instances="true">
     <groups>
       <run>
-        <include name="fault-injection-barrier"/>
+        <include name="multi-region-strong"/>
       </run>
     </groups>
     <packages>

sdk/cosmos/azure-cosmos/CHANGELOG.md

@@ -10,7 +10,8 @@
 * Fixed a possible memory leak (Netty buffers) in Gateway mode caused by a race condition when timeouts are happening. - [47228](https://github.com/Azure/azure-sdk-for-java/pull/47228) and [47251](https://github.com/Azure/azure-sdk-for-java/pull/47251)
 
 #### Other Changes
-* Changed to use incremental change feed to get partition key ranges. - [46810](https://github.com/Azure/azure-sdk-for-java/pull/46810)
+* Changed to use incremental change feed to get partition key ranges. - [PR 46810](https://github.com/Azure/azure-sdk-for-java/pull/46810)
+* Optimized 410 `Lease Not Found` handling for Strong Consistency account by avoiding unnecessary retries in the barrier attainment flow. - [PR 47232](https://github.com/Azure/azure-sdk-for-java/pull/47232)
 
 ### 4.75.0 (2025-10-21)
 > [!IMPORTANT]