Fix Airbyte deferrable execution timeout deadline evaluation and cleanup by SameerMesiah97 · Pull Request #67816 · apache/airflow

SameerMesiah97 · 2026-05-31T21:04:23Z

Description

This change is a follow-up to PR #67382 and addresses the same execution timeout handling gap that was subsequently fixed for the dbt Cloud provider in PR #67626, but for the Airbyte provider.

It also adopts the deferred timeout buffering approach introduced in PR #67229 for the Kubernetes provider to ensure timeout cleanup logic can execute before Airflow's framework-level deferred timeout handling terminates the task.

The implementation now:

Uses wall-clock timestamps (time.time()) for serialized deadline calculations instead of monotonic time.
Enforces execution timeout checks before Airbyte job state evaluation within the trigger.
Adds a defer timeout buffer to allow timeout events to be processed and Airbyte job cancellation to occur before framework-level deferred timeout handling terminates the task.

Rationale

In deferrable mode, Airbyte job cancellation occurs in execute_complete() when the trigger emits a timeout event. However, Airflow's framework-level deferred timeout handling bypasses execute_complete(), which can leave Airbyte jobs running after task timeout.

Additionally, serialized deadlines must use wall-clock time rather than monotonic time, as they are persisted across defer/resume boundaries. Timeout evaluation must also take precedence over job state evaluation to ensure consistent handling once a deadline has elapsed.

This change keeps timeout handling within the trigger/operator flow and provides sufficient time for Airbyte job cancellation before framework-level task termination occurs.

Tests

Added/modified unit tests verifying that:

Wall-clock time is used for serialized deadline handling.
Execution timeout checks take precedence over Airbyte job state evaluation.
Deferred timeout buffering is applied when execution_timeout is configured.

Backwards Compatibility

This change does not modify public APIs or method signatures.

Was generative AI tooling used to co-author this PR?

Yes (please specify the tool below)

Generated-by: [GPT 5.5] following the guidelines

Use wall-clock deadlines instead of monotonic time, enforce timeout checks before job state evaluation, and add a defer timeout buffer to allow execute_complete() to cancel timed-out jobs before task termination.

SameerMesiah97 · 2026-06-01T19:06:29Z

@jscheffl

This PR applies the same fix as in PR #67626.

jscheffl

Thanks for the fix!

SameerMesiah97 · 2026-05-31T21:04:23Z

Description

This change is a follow-up to PR #67382 and addresses the same execution timeout handling gap that was subsequently fixed for the dbt Cloud provider in PR #67626, but for the Airbyte provider.

It also adopts the deferred timeout buffering approach introduced in PR #67229 for the Kubernetes provider to ensure timeout cleanup logic can execute before Airflow's framework-level deferred timeout handling terminates the task.

The implementation now:

Uses wall-clock timestamps (time.time()) for serialized deadline calculations instead of monotonic time.
Enforces execution timeout checks before Airbyte job state evaluation within the trigger.
Adds a defer timeout buffer to allow timeout events to be processed and Airbyte job cancellation to occur before framework-level deferred timeout handling terminates the task.

Rationale

In deferrable mode, Airbyte job cancellation occurs in execute_complete() when the trigger emits a timeout event. However, Airflow's framework-level deferred timeout handling bypasses execute_complete(), which can leave Airbyte jobs running after task timeout.

Additionally, serialized deadlines must use wall-clock time rather than monotonic time, as they are persisted across defer/resume boundaries. Timeout evaluation must also take precedence over job state evaluation to ensure consistent handling once a deadline has elapsed.

This change keeps timeout handling within the trigger/operator flow and provides sufficient time for Airbyte job cancellation before framework-level task termination occurs.

Tests

Added/modified unit tests verifying that:

Wall-clock time is used for serialized deadline handling.
Execution timeout checks take precedence over Airbyte job state evaluation.
Deferred timeout buffering is applied when execution_timeout is configured.

Backwards Compatibility

This change does not modify public APIs or method signatures.

Was generative AI tooling used to co-author this PR?

Yes (please specify the tool below)

Generated-by: [GPT 5.5] following the guidelines

Use wall-clock deadlines instead of monotonic time, enforce timeout checks before job state evaluation, and add a defer timeout buffer to allow execute_complete() to cancel timed-out jobs before task termination.

SameerMesiah97 · 2026-06-01T19:06:29Z

@jscheffl

This PR applies the same fix as in PR #67626.

jscheffl

Thanks for the fix!

SameerMesiah97 · 2026-05-31T21:04:23Z

Description

This change is a follow-up to PR #67382 and addresses the same execution timeout handling gap that was subsequently fixed for the dbt Cloud provider in PR #67626, but for the Airbyte provider.

It also adopts the deferred timeout buffering approach introduced in PR #67229 for the Kubernetes provider to ensure timeout cleanup logic can execute before Airflow's framework-level deferred timeout handling terminates the task.

The implementation now:

Uses wall-clock timestamps (time.time()) for serialized deadline calculations instead of monotonic time.
Enforces execution timeout checks before Airbyte job state evaluation within the trigger.
Adds a defer timeout buffer to allow timeout events to be processed and Airbyte job cancellation to occur before framework-level deferred timeout handling terminates the task.

Rationale

In deferrable mode, Airbyte job cancellation occurs in execute_complete() when the trigger emits a timeout event. However, Airflow's framework-level deferred timeout handling bypasses execute_complete(), which can leave Airbyte jobs running after task timeout.

Additionally, serialized deadlines must use wall-clock time rather than monotonic time, as they are persisted across defer/resume boundaries. Timeout evaluation must also take precedence over job state evaluation to ensure consistent handling once a deadline has elapsed.

This change keeps timeout handling within the trigger/operator flow and provides sufficient time for Airbyte job cancellation before framework-level task termination occurs.

Tests

Added/modified unit tests verifying that:

Wall-clock time is used for serialized deadline handling.
Execution timeout checks take precedence over Airbyte job state evaluation.
Deferred timeout buffering is applied when execution_timeout is configured.

Backwards Compatibility

This change does not modify public APIs or method signatures.

Was generative AI tooling used to co-author this PR?

Yes (please specify the tool below)

Generated-by: [GPT 5.5] following the guidelines

Use wall-clock deadlines instead of monotonic time, enforce timeout checks before job state evaluation, and add a defer timeout buffer to allow execute_complete() to cancel timed-out jobs before task termination.

SameerMesiah97 · 2026-06-01T19:06:29Z

@jscheffl

This PR applies the same fix as in PR #67626.

jscheffl

Thanks for the fix!

SameerMesiah97 · 2026-05-31T21:04:23Z

Description

This change is a follow-up to PR #67382 and addresses the same execution timeout handling gap that was subsequently fixed for the dbt Cloud provider in PR #67626, but for the Airbyte provider.

It also adopts the deferred timeout buffering approach introduced in PR #67229 for the Kubernetes provider to ensure timeout cleanup logic can execute before Airflow's framework-level deferred timeout handling terminates the task.

The implementation now:

Uses wall-clock timestamps (time.time()) for serialized deadline calculations instead of monotonic time.
Enforces execution timeout checks before Airbyte job state evaluation within the trigger.
Adds a defer timeout buffer to allow timeout events to be processed and Airbyte job cancellation to occur before framework-level deferred timeout handling terminates the task.

Rationale

In deferrable mode, Airbyte job cancellation occurs in execute_complete() when the trigger emits a timeout event. However, Airflow's framework-level deferred timeout handling bypasses execute_complete(), which can leave Airbyte jobs running after task timeout.

Additionally, serialized deadlines must use wall-clock time rather than monotonic time, as they are persisted across defer/resume boundaries. Timeout evaluation must also take precedence over job state evaluation to ensure consistent handling once a deadline has elapsed.

This change keeps timeout handling within the trigger/operator flow and provides sufficient time for Airbyte job cancellation before framework-level task termination occurs.

Tests

Added/modified unit tests verifying that:

Wall-clock time is used for serialized deadline handling.
Execution timeout checks take precedence over Airbyte job state evaluation.
Deferred timeout buffering is applied when execution_timeout is configured.

Backwards Compatibility

This change does not modify public APIs or method signatures.

Was generative AI tooling used to co-author this PR?

Yes (please specify the tool below)

Generated-by: [GPT 5.5] following the guidelines

Use wall-clock deadlines instead of monotonic time, enforce timeout checks before job state evaluation, and add a defer timeout buffer to allow execute_complete() to cancel timed-out jobs before task termination.

SameerMesiah97 · 2026-06-01T19:06:29Z

@jscheffl

This PR applies the same fix as in PR #67626.

jscheffl

Thanks for the fix!

SameerMesiah97 · 2026-05-31T21:04:23Z

Description

This change is a follow-up to PR #67382 and addresses the same execution timeout handling gap that was subsequently fixed for the dbt Cloud provider in PR #67626, but for the Airbyte provider.

It also adopts the deferred timeout buffering approach introduced in PR #67229 for the Kubernetes provider to ensure timeout cleanup logic can execute before Airflow's framework-level deferred timeout handling terminates the task.

The implementation now:

Uses wall-clock timestamps (time.time()) for serialized deadline calculations instead of monotonic time.
Enforces execution timeout checks before Airbyte job state evaluation within the trigger.
Adds a defer timeout buffer to allow timeout events to be processed and Airbyte job cancellation to occur before framework-level deferred timeout handling terminates the task.

Rationale

In deferrable mode, Airbyte job cancellation occurs in execute_complete() when the trigger emits a timeout event. However, Airflow's framework-level deferred timeout handling bypasses execute_complete(), which can leave Airbyte jobs running after task timeout.

Additionally, serialized deadlines must use wall-clock time rather than monotonic time, as they are persisted across defer/resume boundaries. Timeout evaluation must also take precedence over job state evaluation to ensure consistent handling once a deadline has elapsed.

This change keeps timeout handling within the trigger/operator flow and provides sufficient time for Airbyte job cancellation before framework-level task termination occurs.

Tests

Added/modified unit tests verifying that:

Wall-clock time is used for serialized deadline handling.
Execution timeout checks take precedence over Airbyte job state evaluation.
Deferred timeout buffering is applied when execution_timeout is configured.

Backwards Compatibility

This change does not modify public APIs or method signatures.

Was generative AI tooling used to co-author this PR?

Yes (please specify the tool below)

Generated-by: [GPT 5.5] following the guidelines

Use wall-clock deadlines instead of monotonic time, enforce timeout checks before job state evaluation, and add a defer timeout buffer to allow execute_complete() to cancel timed-out jobs before task termination.

SameerMesiah97 · 2026-06-01T19:06:29Z

@jscheffl

This PR applies the same fix as in PR #67626.

jscheffl

Thanks for the fix!

SameerMesiah97 · 2026-05-31T21:04:23Z

Description

This change is a follow-up to PR #67382 and addresses the same execution timeout handling gap that was subsequently fixed for the dbt Cloud provider in PR #67626, but for the Airbyte provider.

It also adopts the deferred timeout buffering approach introduced in PR #67229 for the Kubernetes provider to ensure timeout cleanup logic can execute before Airflow's framework-level deferred timeout handling terminates the task.

The implementation now:

Uses wall-clock timestamps (time.time()) for serialized deadline calculations instead of monotonic time.
Enforces execution timeout checks before Airbyte job state evaluation within the trigger.
Adds a defer timeout buffer to allow timeout events to be processed and Airbyte job cancellation to occur before framework-level deferred timeout handling terminates the task.

Rationale

In deferrable mode, Airbyte job cancellation occurs in execute_complete() when the trigger emits a timeout event. However, Airflow's framework-level deferred timeout handling bypasses execute_complete(), which can leave Airbyte jobs running after task timeout.

Additionally, serialized deadlines must use wall-clock time rather than monotonic time, as they are persisted across defer/resume boundaries. Timeout evaluation must also take precedence over job state evaluation to ensure consistent handling once a deadline has elapsed.

This change keeps timeout handling within the trigger/operator flow and provides sufficient time for Airbyte job cancellation before framework-level task termination occurs.

Tests

Added/modified unit tests verifying that:

Wall-clock time is used for serialized deadline handling.
Execution timeout checks take precedence over Airbyte job state evaluation.
Deferred timeout buffering is applied when execution_timeout is configured.

Backwards Compatibility

This change does not modify public APIs or method signatures.

Was generative AI tooling used to co-author this PR?

Yes (please specify the tool below)

Generated-by: [GPT 5.5] following the guidelines

Use wall-clock deadlines instead of monotonic time, enforce timeout checks before job state evaluation, and add a defer timeout buffer to allow execute_complete() to cancel timed-out jobs before task termination.

SameerMesiah97 · 2026-06-01T19:06:29Z

@jscheffl

This PR applies the same fix as in PR #67626.

jscheffl

Thanks for the fix!

SameerMesiah97 · 2026-05-31T21:04:23Z

Description

This change is a follow-up to PR #67382 and addresses the same execution timeout handling gap that was subsequently fixed for the dbt Cloud provider in PR #67626, but for the Airbyte provider.

It also adopts the deferred timeout buffering approach introduced in PR #67229 for the Kubernetes provider to ensure timeout cleanup logic can execute before Airflow's framework-level deferred timeout handling terminates the task.

The implementation now:

Uses wall-clock timestamps (time.time()) for serialized deadline calculations instead of monotonic time.
Enforces execution timeout checks before Airbyte job state evaluation within the trigger.
Adds a defer timeout buffer to allow timeout events to be processed and Airbyte job cancellation to occur before framework-level deferred timeout handling terminates the task.

Rationale

In deferrable mode, Airbyte job cancellation occurs in execute_complete() when the trigger emits a timeout event. However, Airflow's framework-level deferred timeout handling bypasses execute_complete(), which can leave Airbyte jobs running after task timeout.

Additionally, serialized deadlines must use wall-clock time rather than monotonic time, as they are persisted across defer/resume boundaries. Timeout evaluation must also take precedence over job state evaluation to ensure consistent handling once a deadline has elapsed.

This change keeps timeout handling within the trigger/operator flow and provides sufficient time for Airbyte job cancellation before framework-level task termination occurs.

Tests

Added/modified unit tests verifying that:

Wall-clock time is used for serialized deadline handling.
Execution timeout checks take precedence over Airbyte job state evaluation.
Deferred timeout buffering is applied when execution_timeout is configured.

Backwards Compatibility

This change does not modify public APIs or method signatures.

Was generative AI tooling used to co-author this PR?

Yes (please specify the tool below)

Generated-by: [GPT 5.5] following the guidelines

Use wall-clock deadlines instead of monotonic time, enforce timeout checks before job state evaluation, and add a defer timeout buffer to allow execute_complete() to cancel timed-out jobs before task termination.

SameerMesiah97 · 2026-06-01T19:06:29Z

@jscheffl

This PR applies the same fix as in PR #67626.

jscheffl

Thanks for the fix!

SameerMesiah97 · 2026-05-31T21:04:23Z

Description

This change is a follow-up to PR #67382 and addresses the same execution timeout handling gap that was subsequently fixed for the dbt Cloud provider in PR #67626, but for the Airbyte provider.

It also adopts the deferred timeout buffering approach introduced in PR #67229 for the Kubernetes provider to ensure timeout cleanup logic can execute before Airflow's framework-level deferred timeout handling terminates the task.

The implementation now:

Uses wall-clock timestamps (time.time()) for serialized deadline calculations instead of monotonic time.
Enforces execution timeout checks before Airbyte job state evaluation within the trigger.
Adds a defer timeout buffer to allow timeout events to be processed and Airbyte job cancellation to occur before framework-level deferred timeout handling terminates the task.

Rationale

In deferrable mode, Airbyte job cancellation occurs in execute_complete() when the trigger emits a timeout event. However, Airflow's framework-level deferred timeout handling bypasses execute_complete(), which can leave Airbyte jobs running after task timeout.

Additionally, serialized deadlines must use wall-clock time rather than monotonic time, as they are persisted across defer/resume boundaries. Timeout evaluation must also take precedence over job state evaluation to ensure consistent handling once a deadline has elapsed.

This change keeps timeout handling within the trigger/operator flow and provides sufficient time for Airbyte job cancellation before framework-level task termination occurs.

Tests

Added/modified unit tests verifying that:

Wall-clock time is used for serialized deadline handling.
Execution timeout checks take precedence over Airbyte job state evaluation.
Deferred timeout buffering is applied when execution_timeout is configured.

Backwards Compatibility

This change does not modify public APIs or method signatures.

Was generative AI tooling used to co-author this PR?

Yes (please specify the tool below)

Generated-by: [GPT 5.5] following the guidelines

Use wall-clock deadlines instead of monotonic time, enforce timeout checks before job state evaluation, and add a defer timeout buffer to allow execute_complete() to cancel timed-out jobs before task termination.

SameerMesiah97 · 2026-06-01T19:06:29Z

@jscheffl

This PR applies the same fix as in PR #67626.

jscheffl

Thanks for the fix!

boring-cyborgBot added area:providers provider:airbyte labels May 31, 2026

Fix deferrable execution timeout handling for Airbyte jobs.
8849425
Use wall-clock deadlines instead of monotonic time, enforce timeout checks before job state evaluation, and add a defer timeout buffer to allow execute_complete() to cancel timed-out jobs before task termination.

SameerMesiah97 force-pushed the AirbyteTriggerSyncOperator-Misc-Fixes branch from 084e02e to 8849425Compare May 31, 2026 21:09

SameerMesiah97 marked this pull request as ready for review June 1, 2026 08:45

potiuk added the ready for maintainer review Set after triaging when all criteria pass. label Jun 1, 2026

jscheffl approved these changes Jun 1, 2026
View reviewed changes

jscheffl merged commit 31dded5 into apache:mainJun 1, 2026
175 checks passed

potiuk mentioned this pull request Jun 3, 2026
Status of testing Providers that were prepared on June 02, 2026 #67957
Closed