kopf

Enhancements:

• Renamed @kopf.on.this to @kopf.subhandler for readability. #618

Bugfixes:

• Fixed the resource discovery in case of parallel requests (improper locks & checks). #615

Internals:

• Fixed the usage example of kopf.execute() to be asynchronous. #618
• CI functional tests fixed for ResourceWarnings due to pykube-ng's session left-overs. #619 #608

Bugfixes:

• Fixed superseding of one cause by another (deletion-during-creation/-update/-resume, update-during-resume, etc). #607 #606

Bugfixes:

• Fixed superseding of one cause by another (deletion-during-creation/-update/-resume, update-during-resume, etc). #607 #606

Bugfixes:

• No more false "patching failed with inconsistencies" warning for CRDs with status as a subresource. #588

Internal:

• Back-ported GitHub Actions for CI tests. #592, originally #582

General topic: stability, resilience, recoverability, better logging and error handling.

New features:

• JSON logging with --log-format=json. #544
• Plain-text logging --log-format=plain. #544

Bugfixes:

• On-field handlers were triggered for non-existent fields with None→None change. #523
• Sub-handlers' state was not purged, preventing their execution in the future. #517
• Timers were ticking forever after the resource was deleted (if there were no finalizers). #548
• Persisted states of executed and then filtered-out handlers were not purged. #557

Resilience improvements:

• Annotations from other Kopf-based operators are fully ignored to avoid ping-pong effects. #539 #538
• The operator will force-stop itself by SIGKILL after 10 minutes of attempts to stop gracefully. #543
• Crash the whole operator on unrecoverable errors in watchers/workers. #509
• Throttle individual resource processing on unexpected errors. #510
• Extra exceptions are intercepted and ignored for k8s-event posting. #507
• Treat client timeouts during watches similar to other API errors. #506
• Handle all connection errors from the initial streaming request. #512

Experience improvements:

• Annotations restrictions on key length are reworked to allow longer names (v2/v1 keys). #529
• Annotations values are now compacted (no spaces) for the better visual outlook. #516
• Log/warn when the resulting patched object does not match the intended patch. #527
• Log when the patched object unexpectedly disappeared (e.g. was deleted). #559
• Log K8s API error explanations from K8s API itself, not from the HTTP client library. #558
• Filtered-out resources do not create phantom logs & do not get annotated anymore. #545
• Cluster-scoped objects are logged by their name, without "None" as the namespace. #544
• Better phrasing for some log messages to avoid confusion. #540
• Documentation fixes. #530

Internal changes:

• The first release since the fork. #502 #500 #505 #546
• CI feedback is sped up from ≈40-50 mins to ≈5 mins per push. #536 #537
• K3d/K3s is used for CI functional/integration testing; Minikube is for nightly tests. #536 #547
• K8s versions are updated for CI builds. #501
• Imports are sorted with isort. #525 #531
• Added tests for operators' peering. #519
• Added tests for logging & loggers. #544
• Get rid of mocks for causes & registries, use normal classes. #521
• Time-based tests are stabilised (though not fully fixed). #534 #528 #520 #522
• Asyncio tasks are now named in Python 3.8+, which makes the debugging easier. #542
• Python 3.9 support. #561

Bugfixes:

• Support Python 3.9 with Kopf's 0.27 codebase and features. #562

Experience improvements:

• Log when the patched object disappears (e.g. is deleted). Omit inconsistency warnings for this. #559

Experience improvements:

• K8s API errors contain the explanations from K8s API itself, not from an HTTP client library. #558

Bugfixes:

• Persisted states of executed and then filtered-out handlers were not purged. #557

Bugfixes:

• Timers were ticking forever after the resource was deleted (if there were no finalizers). #548

General topic: stability, resilience, recoverability, better logging and error handling.

New features:

• JSON logging with --log-format=json. #544
• Plain-text logging --log-format=plain. #544

Bugfixes:

• On-field handlers were triggered for non-existent fields with None→None change. #523
• Sub-handlers' state was not purged, preventing their execution in the future. #517

Resilience improvements:

• Annotations from other Kopf-based operators are fully ignored to avoid ping-pong effects. #539 #538
• The operator will force-stop itself by SIGKILL after 10 minutes of attempts to stop gracefully. #543
• Crash the whole operator on unrecoverable errors in watchers/workers. #509
• Throttle individual resource processing on unexpected errors. #510
• Extra exceptions are intercepted and ignored for k8s-event posting. #507
• Treat client timeouts during watches similar to other API errors. #506
• Handle all connection errors from the initial streaming request. #512

Experience improvements:

• Annotations restrictions on key length are reworked to allow longer names (v2/v1 keys). #529
• Annotations values are now compacted (no spaces) for the better visual outlook. #516
• Log/warn when the resulting patched object does not match the intended patch. #527
• Filtered-out resources do not create phantom logs & do not get annotated anymore. #545
• Cluster-scoped objects are logged by their name, without "None" as the namespace. #544
• Better phrasing for some log messages to avoid confusion. #540
• Documentation fixes. #530

Internal changes:

• The first release since the fork. #502 #500 #505 #546
• CI feedback is sped up from ≈40-50 mins to ≈5 mins per push. #536 #537
• K3d/K3s is used for CI functional/integration testing; Minikube is for nightly tests. #536 #547
• K8s versions are updated for CI builds. #501
• Imports are sorted with isort. #525 #531
• Added tests for operators' peering. #519
• Added tests for logging & loggers. #544
• Get rid of mocks for causes & registries, use normal classes. #521
• Time-based tests are stabilised (though not fully fixed). #534 #528 #520 #522
• Asyncio tasks are now named in Python 3.8+, which makes the debugging easier. #542

Originally released on 2019-03-27 12:54:00+00:00 (link).

Initial import of the squashed repository.

Originally released on 2019-03-27 12:03:34+00:00 (link).

Originally released on 2019-03-27 13:58:12+00:00 (link).

Originally released on 2019-04-16 11:53:26+00:00 (link).

Originally released on 2019-03-28 13:12:39+00:00 (link).

Originally released on 2019-04-18 13:40:04+00:00 (link).

Originally released on 2019-04-24 15:56:00+00:00 (link).

• Peering mode auto-detection by default, falling back to standalone if not configured. #33 #38

Originally released on 2019-04-25 09:22:10+00:00 (link).

• Better PyPI's README page #41

Originally released on 2019-04-26 15:15:00+00:00 (link).

• Follow the Kubernetes "eventual consistency" principle and react only to the latest events, ignore the past replayed events #43 #42
• Tests for the diff calculations #40 #48