Feat/child context by maschnetwork · Pull Request #100 · aws/aws-durable-execution-sdk-java

maschnetwork · 2026-02-16T16:28:16Z

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

Issue Link, if available

Description

Adds runInChildContext and runInChildContextAsync to the Java Durable Execution SDK. Each child context gets its own operation counter and checkpoint log, enabling concurrent branches of work with per-context determinism.

Implementation details:

ChildContextOperation extends BaseDurableOperation<T> and manages the child context lifecycle: START (fire-and-forget), execute user function in a separate thread, SUCCEED/FAIL (blocking checkpoint).
parentId propagated through BaseDurableOperation to all operation subclasses, replacing the hardcoded null.
Per-context replay state tracked via isReplaying on each DurableContext instance, since a child may be replaying while the parent is already executing.
Operation IDs are prefixed with the parent context's ID using - as separator (e.g., "3-1", "3-2" for operations inside parent "3"). This matches the JS SDK's stepPrefix convention and ensures global uniqueness — the backend validates type consistency by operation ID alone. Nested contexts chain naturally (e.g., "3-2-1").
Large results (≥256KB) trigger the ReplayChildren flow — SUCCEED checkpoint with empty payload + ContextOptions { replayChildren: true }, reconstructed via re-execution on replay.
ChildContextFailedException follows the same pattern as StepFailedException.

Deferred:

Orphan detection in CheckpointBatcher (preventing stale checkpoints from in-flight child operations after parent completes)
summaryGenerator for large-result observability

See docs/design-run-in-child-context.md for the full design.

Demo/Screenshots

Checklist

I have filled out every section of the PR template
I have thoroughly tested this change

Testing

Unit Tests

ChildContextOperationTest — covers first execution, replay SUCCEEDED, replay FAILED, replay STARTED, replayChildren path, non-deterministic detection, and error handling.
DurableContextTest, ReplayValidationTest — updated for createRootContext factory method.

Integration Tests

ChildContextIntegrationTest — 15 different cases including waits and nesting.

Examples

ChildContextExample — demonstrates three concurrent child contexts (two with step+wait, one with a nested child context), collected via DurableFuture.allOf.
Cloud test added to CloudBasedIntegrationTest.

…ite key lookups - Create OperationKey record in execution/ package with parentId/operationId fields and static factory methods of() and fromOperation() - Refactor ExecutionManager.operations map from Map<String, Operation> to Map<OperationKey, Operation> for scoped operation lookups - Update fetchAllPages collector to use OperationKey.fromOperation(op) - Change getOperationAndUpdateReplayState to accept parentId parameter - Refactor openPhasers map to Map<OperationKey, Phaser> - Update startPhaser to accept parentId parameter - Update BaseDurableOperation to pass null as parentId (temporary until parentId propagation is wired in task 3) - Update all test files to match new method signatures

- Add parentId field to DurableContext to track parent-child relationships - Pass parentId through all operation constructors (StepOperation, WaitOperation, InvokeOperation, CallbackOperation) - Update BaseDurableOperation to accept and store parentId parameter - Add convenience constructor in BaseDurableOperation for root-context operations where parentId is null - Update ExecutionManager calls to use parentId for scoped operation lookups instead of null - Add getContextId() method to DurableContext to retrieve parentId - Add getParentId() protected method to BaseDurableOperation for accessing parent context ID - Update operation update builder to include parentId instead of hardcoded null value - Enables proper operation tracking and isolation within child execution contexts

- Add isReplaying field to track per-context replay mode state - Initialize replay state based on cached operations in ExecutionManager - Add setExecutionMode() to transition context from replay to execution - Refactor DurableContext constructor to private shared initialization - Add createRootContext() static factory methods for root context creation - Add createChildContext() static factory method for child context creation - Add hasOperationsForContext() to ExecutionManager for replay state detection - Update DurableExecutor to use createRootContext() factory method - Update test fixtures to use new factory methods - Enables per-context replay tracking for improved operation handling

…nd unit tests - Add ChildContextOperation extending BaseDurableOperation with execute/get lifecycle - Handle first execution (START fire-and-forget, then run child context) - Handle replay: SUCCEEDED (cached result), FAILED (re-throw), STARTED (re-execute) - Implement large result handling (>=256KB) via ReplayChildren flow - Add ChildContextFailedException for non-reconstructable exception fallback - Add DurableContext.createChildContext() factory (no thread registration) - Add per-context replay state (isReplaying field) to DurableContext - Unit tests covering replay scenarios, failure preservation, and non-determinism detection

examples/template-child-context.yaml

sdk/src/main/java/com/amazonaws/lambda/durable/DurableContext.java

docs/design-run-in-child-context.md

sdk/src/main/java/com/amazonaws/lambda/durable/operation/CallbackOperation.java

sdk/src/main/java/com/amazonaws/lambda/durable/operation/ChildContextOperation.java

zhongkechen · 2026-02-19T20:25:31Z

sdk/src/main/java/com/amazonaws/lambda/durable/operation/ChildContextOperation.java

+        var contextId = getOperationId();
+
+        // Register child context thread before executor runs (prevents suspension)
+        registerActiveThread(contextId, ThreadType.CONTEXT);


I think it's registered twice and also setCurrentContext is called twice here

This is the same pattern as StepOperation.executeStepLogic. registerActiveThread runs on the parent thread to ensure the child is tracked before the parent can deregister (preventing a false "no active threads" suspension), while setCurrentContext sets the ThreadLocal on the actual child execution thread.

I will add a more detailed comment for this

zhongkechen · 2026-02-19T23:28:15Z

sdk/src/main/java/com/amazonaws/lambda/durable/DurableContext.java


-        // Register root context thread as active
+    /** Creates a root context with the given contextId and registers the current thread. */
+    static DurableContext createRootContext(


Is this still a Root context if a contextId is specified?

I can actually merge the two constructors to make it more concrete. Good call.

maschnetwork added 16 commits February 16, 2026 10:01

fix: testing

5a779b4

feat: add public methods

707646e

feat: added integration tests

7777543

feat: added edge case tests

c0c9023

feat: cloud based tests

b035edc

feat: adapted readme and design to match implementation

490014b

feat: remove debug line from testing commit

df73635

fix: naming and null handling

b8c8bed

fix: use run instead of complete

75fbe8a

fix: run for PENDING

49cf1e9

feat: simplify operation id

75a6da4

fix: revert test changes for operation id

8a2b3ab

maschnetwork commented Feb 16, 2026

View reviewed changes

examples/template-child-context.yaml Outdated Show resolved Hide resolved

fix: simplify testing

3ac3b9a

maschnetwork requested review from phipag and zhongkechen February 16, 2026 22:56

zhongkechen previously approved these changes Feb 16, 2026

View reviewed changes

sdk/src/main/java/com/amazonaws/lambda/durable/DurableContext.java Outdated Show resolved Hide resolved

zhongkechen reviewed Feb 18, 2026

View reviewed changes

docs/design-run-in-child-context.md Outdated Show resolved Hide resolved

zhongkechen reviewed Feb 18, 2026

View reviewed changes

sdk/src/main/java/com/amazonaws/lambda/durable/operation/CallbackOperation.java Show resolved Hide resolved

zhongkechen reviewed Feb 18, 2026

View reviewed changes

sdk/src/main/java/com/amazonaws/lambda/durable/operation/ChildContextOperation.java Outdated Show resolved Hide resolved

fix: addressed review comments

842a106

maschnetwork dismissed zhongkechen’s stale review via 842a106 February 19, 2026 16:39

Merge remote-tracking branch 'origin/main' into child-context

06e5233

zhongkechen previously approved these changes Feb 19, 2026

View reviewed changes

zhongkechen reviewed Feb 19, 2026

View reviewed changes

fix: review comments

1a4c8c4

maschnetwork dismissed zhongkechen’s stale review via 1a4c8c4 February 20, 2026 14:47

maschnetwork marked this pull request as ready for review February 20, 2026 14:48

fix: debug tests

765fbd5

zhongkechen previously approved these changes Feb 20, 2026

View reviewed changes

fix: used runUntilComplete for slow CI worker

c1b30a5

maschnetwork dismissed zhongkechen’s stale review via c1b30a5 February 20, 2026 15:48

fix: ci comment

45ccbff

zhongkechen approved these changes Feb 20, 2026

View reviewed changes

zhongkechen merged commit f40431b into main Feb 20, 2026
7 of 10 checks passed

zhongkechen deleted the feat/child-context branch February 20, 2026 20:56

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Feat/child context#100

Feat/child context#100
zhongkechen merged 23 commits intomainfrom
feat/child-context

maschnetwork commented Feb 16, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

zhongkechen Feb 19, 2026

Uh oh!

maschnetwork Feb 20, 2026

Uh oh!

maschnetwork Feb 20, 2026

Uh oh!

zhongkechen Feb 19, 2026

Uh oh!

maschnetwork Feb 20, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

maschnetwork commented Feb 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Issue Link, if available

Description

Demo/Screenshots

Checklist

Testing

Unit Tests

Integration Tests

Examples

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

zhongkechen Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

maschnetwork Feb 20, 2026

Choose a reason for hiding this comment

Uh oh!

maschnetwork Feb 20, 2026

Choose a reason for hiding this comment

Uh oh!

zhongkechen Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

maschnetwork Feb 20, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

maschnetwork commented Feb 16, 2026 •

edited

Loading