Fix JVM <clinit> deadlock by removing static final accessor fields by jeet1995 · Pull Request #48689 · Azure/azure-sdk-for-java

jeet1995 · 2026-04-03T23:20:40Z

Summary

Fixes a JVM <clinit> deadlock (#48622, #48585) that permanently hangs threads when multiple threads concurrently trigger Cosmos SDK class loading. Also fixes a latent CosmosItemSerializer.DEFAULT_SERIALIZER null bug.

Fixes: #48622, #48585

Root Cause

Deadlock: Consuming classes cached accessors in private static final fields. During <clinit>, the getter calls initializeAllAccessors(), eagerly loading 40+ classes. Concurrent <clinit> of different classes creates circular init-lock waits — permanent deadlock per JLS §12.4.2.

DEFAULT_SERIALIZER null: CosmosItemSerializer.DEFAULT_SERIALIZER cross-referenced DefaultCosmosItemSerializer.DEFAULT_SERIALIZER. When DefaultCosmosItemSerializer.<clinit> ran first, recursive same-thread <clinit> of the parent read the child's field before it was set.

Parent-child <clinit> deadlock: DefaultCosmosItemSerializer.INTERNAL_DEFAULT_SERIALIZER was accessed independently by implementation code, triggering child <clinit> on a different thread than the parent — creating an AB/BA init-lock deadlock between parent and child.

Fix

1. Uniform static getter pattern

// Before — triggers initializeAllAccessors() during <clinit>
private static final FeedResponseAccessor feedResponseAccessor =
    ImplementationBridgeHelpers.FeedResponseHelper.getFeedResponseAccessor();

// After — no <clinit> involvement
private static FeedResponseAccessor feedResponseAccessor() {
    return ImplementationBridgeHelpers.FeedResponseHelper.getFeedResponseAccessor();
}

2. Break CosmosItemSerializer ↔ DefaultCosmosItemSerializer cycle

CosmosItemSerializer.DEFAULT_SERIALIZER creates instance directly via new DefaultCosmosItemSerializer(...) — no cross-class <clinit> dependency
INTERNAL_DEFAULT_SERIALIZER moved from DefaultCosmosItemSerializer to CosmosItemSerializer (private, exposed via CosmosItemSerializerAccessor.getInternalDefaultSerializer()) — so that accessing it no longer triggers child <clinit> from a different thread, eliminating the AB/BA init-lock between parent and child
static { initialize(); } placed before DEFAULT_SERIALIZER so the accessor is registered before construction — eliminates initializeAllAccessors() fallback during <clinit>
DefaultCosmosItemSerializer.DEFAULT_SERIALIZER and its serializationInclusionModeAwareObjectMapper removed (dead code)

Scope

Category	Description
Static final/instance accessor fields removed	66 fields across consuming classes
Static getter methods added	~120 `private static XxxAccessor xxx()` methods
Missing `static { initialize(); }` added	`CosmosRequestContext`, `CosmosOperationDetails`, `CosmosDiagnosticsContext`
Accessor rename fix	`getCosmosAsyncClientAccessor()` → `getCosmosDiagnosticsThresholdsAccessor()` in `CosmosDiagnosticsThresholdsHelper`
`checkNotNull` bug fix	`DefaultCosmosItemSerializer` constructor passed string literal instead of parameter
Files changed	85

Exceptions (not converted to static getters)

HttpClient.java — Java 8 interface, no private static methods

Tests

Test	What it proves
`concurrentAccessorInitializationShouldNotDeadlock` (×5 invocations)	Forked JVMs with 12 concurrent threads triggering `<clinit>` — catches deadlock via 30s timeout
`allAccessorClassesMustHaveStaticInitializerBlock`	Forked JVM verifies every accessor is non-null after `<clinit>`
`noStaticOrInstanceAccessorFieldsInConsumingClasses`	Reflection scan: fails if any class has a `static` or `final` Accessor field
`accessorInitialization`	Validates `initializeAllAccessors()` bootstrap path

jeet1995 · 2026-04-03T23:23:45Z

Closing — bridge classes don't allow adding new methods. Proceeding with #48667 (Class.forName with explicit classloader).

Copilot

Pull request overview

This PR addresses a JVM <clinit> deadlock in the Cosmos Java SDK by removing many static final ...Accessor caches in consuming classes and switching call sites to resolve accessors lazily via ImplementationBridgeHelpers.*Helper.get*Accessor() on demand, reducing class-initialization-time cross-dependencies.

Changes:

Replaced numerous private static final XxxAccessor ... = getXxxAccessor() fields with inline (lazy) getter calls at usage sites.
Added/adjusted static { initialize(); } blocks and <clinit> ordering to ensure accessors are registered safely during class initialization where required.
Added forked-JVM regression/enforcement tests around concurrent <clinit> behavior and accessor registration.

Reviewed changes

Copilot reviewed 54 out of 54 changed files in this pull request and generated 9 comments.

Show a summary per file

File	Description
sdk/spring/azure-spring-data-cosmos/README.md	Trailing whitespace/newline adjustment.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/util/CosmosPagedFluxStaticListImpl.java	Removed static accessor cache; inline FeedResponse accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/util/CosmosPagedFluxDefaultImpl.java	Removed static accessor cache; inline diagnostics context accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/models/FeedResponse.java	Removed static diagnostics accessor cache; inline accessor calls.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/models/CosmosOperationDetails.java	Added `static { initialize(); }` registration block.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/models/CosmosItemRequestOptions.java	Removed static thresholds accessor cache; inline thresholds accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/StaleResourceRetryPolicy.java	Removed static exception accessor cache.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/SessionTokenMismatchRetryPolicy.java	Removed static accessor cache; inline session retry options accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/RxDocumentClientImpl.java	Removed multiple static accessor caches; inline accessor usage across implementation.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/QueryPlanRetriever.java	Removed static accessor caches; inline accessors for options/exception handling.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/PipelinedQueryExecutionContext.java	Removed static accessor cache; inline accessor usage when cloning options.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/PipelinedDocumentQueryExecutionContext.java	Removed static accessor caches; inline options and serializer accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/ParallelDocumentQueryExecutionContext.java	Removed static accessor caches; inline options/diagnostics accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/OrderByUtils.java	Removed static diagnostics accessor cache; inline diagnostics accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/OrderByDocumentQueryExecutionContext.java	Removed static accessor caches; inline feed/diagnostics accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/OrderByDocumentProducer.java	Removed static feed accessor cache; inline feed accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/NonStreamingOrderByUtils.java	Removed static diagnostics accessor cache; inline diagnostics accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/NonStreamingOrderByDocumentQueryExecutionContext.java	Removed static accessor caches; inline feed/diagnostics accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/HybridSearchDocumentQueryExecutionContext.java	Removed static accessor caches; inline feed/diagnostics accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/GroupByDocumentQueryExecutionContext.java	Removed static diagnostics accessor cache; inline diagnostics accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/Fetcher.java	Removed static diagnostics accessor cache; inline diagnostics accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/DocumentQueryExecutionContextFactory.java	Inline options accessor usage in creation flow.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/DocumentQueryExecutionContextBase.java	Removed static accessor caches; inline accessor usage for request creation and cloning.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/DocumentProducer.java	Removed static accessor cache; inline accessor usage when cloning options.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/DefaultDocumentQueryExecutionContext.java	Removed static accessor cache; inline accessor usage for partition key definition/properties.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/DCountDocumentQueryExecutionContext.java	Removed static diagnostics accessor cache; inline diagnostics accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/ChangeFeedFetcher.java	Removed static feed accessor cache; inline feed accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/query/AggregateDocumentQueryExecutionContext.java	Removed static accessor caches; inline feed/diagnostics accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/JsonSerializable.java	Removed static serializer accessor cache; inline serializer accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/ImplementationBridgeHelpers.java	Renamed thresholds accessor getter (`getCosmosAsyncClientAccessor` → `getCosmosDiagnosticsThresholdsAccessor`).
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/http/HttpClientConfig.java	Removed static HTTP2 config accessor cache; inline accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/Document.java	Removed static serializer accessor cache; inline serializer accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/directconnectivity/GoneAndRetryWithRetryPolicy.java	Removed static exception accessor cache; inline exception accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/DiagnosticsProvider.java	Removed multiple static accessor caches; inline accessors throughout tracing/metrics paths.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/CosmosQueryRequestOptionsImpl.java	Updated thresholds accessor getter name usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/CosmosQueryRequestOptionsBase.java	Updated thresholds accessor getter name usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/ConnectionPolicy.java	Removed static HTTP2 config accessor cache; inline accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/clienttelemetry/ClientTelemetryMetrics.java	Removed static accessor caches; inline accessors for telemetry metrics recording.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/clienttelemetry/ClientMetricsDiagnosticsHandler.java	Removed static telemetry config accessor cache.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/ChangeFeedQueryImpl.java	Removed static accessor caches; inline accessors for change feed request/response.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/caches/RxCollectionCache.java	Removed static exception accessor cache; inline exception accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/batch/TransactionalBulkExecutor.java	Removed static batch request options accessor cache; inline accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/implementation/batch/BulkExecutor.java	Removed static accessor caches; inline accessors for batch response and diagnostics provider.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/CosmosRequestContext.java	Added `static { initialize(); }` registration block.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/CosmosItemSerializer.java	Reordered `<clinit>` to register accessor before static fields.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/CosmosDiagnosticsContext.java	Added `static { initialize(); }` registration block.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/CosmosContainerProactiveInitConfig.java	Removed static container identity accessor cache; inline accessor usage.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/CosmosAsyncUser.java	Removed static accessor caches; inline accessors for query naming/feed response creation.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/CosmosAsyncScripts.java	Removed static accessor caches; inline accessors for query naming/feed response creation.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/CosmosAsyncDatabase.java	Removed static accessor caches; inline accessors for query naming/feed response creation.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/CosmosAsyncContainer.java	Removed many static accessor caches; inline accessors across request/response, policies, and telemetry.
sdk/cosmos/azure-cosmos/src/main/java/com/azure/cosmos/CosmosAsyncClient.java	Removed static accessor caches; inline telemetry/options/feed response accessors.
sdk/cosmos/azure-cosmos/CHANGELOG.md	Added changelog bullet for `<clinit>` deadlock fix.
sdk/cosmos/azure-cosmos-tests/src/test/java/com/azure/cosmos/implementation/ImplementationBridgeHelpersTest.java	Added forked-JVM deadlock regression test and accessor registration enforcement test.

… stale docs - Removed remaining static final accessor fields in DocumentQueryExecutionContextFactory, CosmosQueryRequestOptionsBase, CosmosQueryRequestOptionsImpl - Extracted local variables for long inline accessor chains in SessionTokenMismatchRetryPolicy, RxDocumentClientImpl, CosmosPagedFluxDefaultImpl - Updated test Javadoc to reflect lazy accessor approach (not Class.forName) - Reduced child JVM runs from 3 to 1 (invocationCount=5 provides repetition) - Fixed CHANGELOG PR link to Azure#48689 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

xinlian12 · 2026-04-11T05:59:28Z

✅ Review complete (36:01)

Posted 2 inline comment(s).

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

… null Fixes Azure#48622, Azure#48585 Replace all static final accessor fields and inline ImplementationBridgeHelpers calls with uniform private static getter methods. This eliminates <clinit>-time class loading that caused permanent deadlocks under concurrent class initialization (JLS 12.4.2). Fix CosmosItemSerializer.DEFAULT_SERIALIZER circular <clinit> — create instance directly and move INTERNAL_DEFAULT_SERIALIZER to parent class to prevent concurrent <clinit> between parent and child. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

- ModelBridgeInternal.java: Remove 26 duplicate ImplementationBridgeHelpers imports - ItemBulkOperation.java: Remove 2 duplicate ImplementationBridgeHelpers imports - SqlQuerySpecWithEncryption.java: Add private static internalDefaultSerializer() getter (matching uniform pattern), replace inline accessor calls, remove unused DefaultCosmosItemSerializer import Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

jeet1995 · 2026-04-11T21:23:05Z

/azp run java - cosmos - ci

jeet1995 · 2026-04-11T21:23:14Z

/azp run java - cosmos - spark

jeet1995 · 2026-04-11T21:23:15Z

/azp run java - cosmos - tests

jeet1995 · 2026-04-11T21:23:15Z

/azp run java - cosmos - kafka

azure-pipelines · 2026-04-11T21:23:20Z

Azure Pipelines successfully started running 1 pipeline(s).

azure-pipelines · 2026-04-11T21:23:25Z

Azure Pipelines successfully started running 1 pipeline(s).

azure-pipelines · 2026-04-11T21:23:33Z

Azure Pipelines successfully started running 1 pipeline(s).

azure-pipelines · 2026-04-11T21:23:38Z

Azure Pipelines successfully started running 1 pipeline(s).

…Block test Clarify that the test verifies accessor resolvability (via <clinit> or initializeAllAccessors fallback), not that each class independently registers its accessor. Structural enforcement is done by the companion noStaticOrInstanceAccessorFieldsInConsumingClasses test. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

jeet1995 · 2026-04-13T16:37:28Z

/azp run java - cosmos - ci

jeet1995 · 2026-04-13T16:37:30Z

/azp run java - cosmos - tests

jeet1995 · 2026-04-13T16:37:31Z

/azp run java - cosmos - kafka

jeet1995 · 2026-04-13T16:37:33Z

/azp run java - cosmos - spark

jeet1995 · 2026-04-13T16:37:34Z

/azp run java - spring - ci

azure-pipelines · 2026-04-13T16:37:43Z

Azure Pipelines successfully started running 1 pipeline(s).

azure-pipelines · 2026-04-13T16:37:45Z

Azure Pipelines successfully started running 1 pipeline(s).

azure-pipelines · 2026-04-13T16:37:47Z

Azure Pipelines successfully started running 1 pipeline(s).

azure-pipelines · 2026-04-13T16:37:48Z

Azure Pipelines successfully started running 1 pipeline(s).

azure-pipelines · 2026-04-13T16:37:57Z

Azure Pipelines successfully started running 1 pipeline(s).

FabianMeiswinkel

LGTM

xinlian12

LGTM, thanks

jeet1995 · 2026-04-14T20:42:23Z

Thin-client test failures are due to service side config updates.

jeet1995 · 2026-04-14T20:48:09Z

/check-enforcer override

github-actions bot added azure-spring All azure-spring related issues Cosmos labels Apr 3, 2026

jeet1995 closed this Apr 3, 2026

jeet1995 reopened this Apr 3, 2026

jeet1995 force-pushed the fix/clinit-deadlock-bridge-methods branch from e57066d to 66afd43 Compare April 4, 2026 00:05

jeet1995 changed the title ~~Fix JVM <clinit> deadlock using targeted bridge methods (alternative to #48667)~~ Fix JVM <clinit> deadlock by removing static final accessor fields (alternative to #48667) Apr 4, 2026

jeet1995 marked this pull request as ready for review April 5, 2026 00:30

jeet1995 requested review from a team, Netyyyy, kirankumarkolli, moarychan, rujche and saragluna as code owners April 5, 2026 00:30

Copilot AI review requested due to automatic review settings April 5, 2026 00:30

jeet1995 mentioned this pull request Apr 5, 2026

Fix JVM <clinit> deadlock in Cosmos SDK accessor initialization #48667

Closed

jeet1995 changed the title ~~Fix JVM <clinit> deadlock by removing static final accessor fields (alternative to #48667)~~ Fix JVM <clinit> deadlock by removing static final accessor fields Apr 5, 2026

Copilot started reviewing on behalf of jeet1995 April 5, 2026 00:31 View session

Copilot AI reviewed Apr 5, 2026

View reviewed changes

jeet1995 mentioned this pull request Apr 6, 2026

Experiment: static block initialization in ImplementationBridgeHelpers #48697

Closed

jeet1995 force-pushed the fix/clinit-deadlock-bridge-methods branch from 4e41bc9 to 101efda Compare April 7, 2026 22:08

jeet1995 force-pushed the fix/clinit-deadlock-bridge-methods branch 5 times, most recently from d8edfe4 to 7332533 Compare April 8, 2026 00:43

rujche requested a review from Copilot April 8, 2026 00:57

xinlian12 reviewed Apr 11, 2026

View reviewed changes

Comment thread ...-encryption/src/main/java/com/azure/cosmos/encryption/models/SqlQuerySpecWithEncryption.java

jeet1995 and others added 2 commits April 11, 2026 17:21

jeet1995 and others added 3 commits April 13, 2026 11:02

Merge branch 'Azure:main' into fix/clinit-deadlock-bridge-methods

fc90755

Refactoring.

bc54e90

FabianMeiswinkel approved these changes Apr 14, 2026

View reviewed changes

xinlian12 approved these changes Apr 14, 2026

View reviewed changes

Merge branch 'main' into fix/clinit-deadlock-bridge-methods

f136b78

Conversation

jeet1995 commented Apr 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Root Cause

Fix

1. Uniform static getter pattern

2. Break CosmosItemSerializer ↔ DefaultCosmosItemSerializer cycle

Scope

Exceptions (not converted to static getters)

Tests

Uh oh!

jeet1995 commented Apr 3, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

xinlian12 commented Apr 11, 2026

Uh oh!

jeet1995 commented Apr 11, 2026

Uh oh!

jeet1995 commented Apr 11, 2026

Uh oh!

jeet1995 commented Apr 11, 2026

Uh oh!

jeet1995 commented Apr 11, 2026

Uh oh!

azure-pipelines bot commented Apr 11, 2026

Uh oh!

azure-pipelines bot commented Apr 11, 2026

Uh oh!

azure-pipelines bot commented Apr 11, 2026

Uh oh!

azure-pipelines bot commented Apr 11, 2026

Uh oh!

jeet1995 commented Apr 13, 2026

Uh oh!

jeet1995 commented Apr 13, 2026

Uh oh!

jeet1995 commented Apr 13, 2026

Uh oh!

jeet1995 commented Apr 13, 2026

Uh oh!

jeet1995 commented Apr 13, 2026

Uh oh!

azure-pipelines bot commented Apr 13, 2026

Uh oh!

azure-pipelines bot commented Apr 13, 2026

Uh oh!

azure-pipelines bot commented Apr 13, 2026

Uh oh!

azure-pipelines bot commented Apr 13, 2026

Uh oh!

azure-pipelines bot commented Apr 13, 2026

Uh oh!

FabianMeiswinkel left a comment

Choose a reason for hiding this comment

Uh oh!

xinlian12 left a comment

Choose a reason for hiding this comment

Uh oh!

jeet1995 commented Apr 14, 2026

Uh oh!

jeet1995 commented Apr 14, 2026

Uh oh!

Reviewers

Assignees

Labels

jeet1995 commented Apr 3, 2026 •

edited

Loading