Skip to content

Conversation

@fgiorgetti
Copy link
Member

@fgiorgetti fgiorgetti commented Nov 20, 2025

Proper control to the lifecycle of a PodWatcher used by AttachedConnectors when:

  • Site is deleted and re-created
  • AttachedConnector.spec.siteNamespace changes
  • AttachedConnectorBinding.spec.connectorNamespace changes
  • AttachedConnector is deleted
  • AttachedConnectorBinding is deleted
  • Avoid unnecessary updates to the skupper-router configmap on podwatcher events (for attached connectors)
  • Unit tests

Fixes #2320.

@fgiorgetti fgiorgetti marked this pull request as ready for review November 21, 2025 21:32
@fgiorgetti fgiorgetti requested a review from ajssmith November 24, 2025 12:55
@fgiorgetti fgiorgetti merged commit 24720f8 into skupperproject:main Dec 10, 2025
2 checks passed
fgiorgetti added a commit to fgiorgetti/skupper that referenced this pull request Dec 11, 2025
…pperproject#2321)

* Fixes the lifecycle of the PodWatcher used by AttachedConnectors

Fixes skupperproject#2320.

* Add unit tests

* Stopping watcher when binding is deleted

* Unit tests to validate podwatcher stopped when attached connector or binding deleted
fgiorgetti added a commit that referenced this pull request Dec 12, 2025
* Updated versions

* upgrade go version to 1.24.9

Signed-off-by: Christian Kruse <christian@c-kruse.com>

* Fix rendering of system site with LinkAccess specified

* Change kube-adaptor Leader Election Loss Error Handling (#2296)

* Change kube-adaptor Leader Election Loss Error Handling

Updates the kube-adaptor so that when the skupper-site-leader Lease is
lost the kube-adaptor will retry instead of exiting.

Signed-off-by: Christian Kruse <christian@c-kruse.com>

* remove harmful lease owner assignment

Signed-off-by: Christian Kruse <christian@c-kruse.com>

* fix kube flow controller go routine leak

Signed-off-by: Christian Kruse <christian@c-kruse.com>

---------

Signed-off-by: Christian Kruse <christian@c-kruse.com>

* Prevents constant changes to HA connector (#2297)

* Prevents constant changes to HA connector

* Not mapped Ordinal causing indefinite sslProfile updates
* Ensure HA connector is not considered different by setting cost to 1

* Mapping oldestValidOrdinal as well as it is needed when HA is enabled and disabled multiple times

* Updated routeraccess test to match expected connector

* Refactor internal/kube/watchers implementation (#2304)

* Refactor internal/kube/watchers implementation

Replaces the ~20 duplicated watchers.Watcher implementations with a
single generic implementation. Also removes the unused
watchers.Callback structures.

Signed-off-by: Christian Kruse <christian@c-kruse.com>

* spell

Signed-off-by: Christian Kruse <christian@c-kruse.com>

---------

Signed-off-by: Christian Kruse <christian@c-kruse.com>

* Adds nginx.command template value to network observer chart (#2308)

Adds nginx.command template value to the network observer chart for the
network-observer deployment's nginx proxy container. Allows for
alternate images and configurations to be used where the command needs
to be specified.

Signed-off-by: Christian Kruse <christian@c-kruse.com>

* Fixes network observer traffic metrics bug (#2310)

Previously traffic metrics (skupper_sent_bytes_total) was only updated
when skupper_received_bytes_total was also incremented. This is
especially visible in asymetric applications when data flows in one
direction like iperf.

Signed-off-by: Christian Kruse <christian@c-kruse.com>

* Fix CI Image Build docker dependency (#2318)

Removes superfluous docker install step from build-oci-images CI Job.
Installed version was conflicting with containerd running in CircleCI VM
image.

Signed-off-by: Christian Kruse <christian@c-kruse.com>

* Use a valid context for sending first flush (#2330)

In case the wait for initial message times out, the initial flush
message was sent using an expired context.

* System controller properly accepts multicast message (#2332)

Fixes #2331

* remove unhelpful logs (#2329)

* Error if a Site is created but old configuration is still present (#2324)

In case a Site is deleted and recreated quickly (automated), the
skupper-router ConfigMap owned by the previous site, may still be
present (owned recource not yet deleted).

The controller now fails, if it finds a router configuration that
is not owned by the currently active site.

Fixes #2323.

* Fixes the lifecycle of the PodWatcher used by AttachedConnectors (#2321)

* Fixes the lifecycle of the PodWatcher used by AttachedConnectors

Fixes #2320.

* Add unit tests

* Stopping watcher when binding is deleted

* Unit tests to validate podwatcher stopped when attached connector or binding deleted

---------

Signed-off-by: Christian Kruse <christian@c-kruse.com>
Co-authored-by: Christian Kruse <christian@c-kruse.com>
Co-authored-by: ajssmith <ansmith@redhat.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Tcp Connectors gets deleted and created indefinitely

2 participants