kubeovn / kube-ovn / 23327398190

23%

Build Type

push

github

Committed by web-flow

Commit Message

fix(daemon): create listen sockets before starting controller to avoid race (#6478)

* fix(daemon): create listen sockets before starting controller to avoid race

The daemon starts a metrics/health server and the controller
concurrently. The controller's configProviderNic() transfers IP
addresses from the physical NIC to an OVS bridge. If the metrics
server's bind() call races with this transfer, it fails with
"cannot assign requested address" and calls LogFatalAndExit,
crashing the daemon.

On restart, the new daemon instance finds the IP address already
on br-provider but the IPv4 default route permanently lost (the
kernel removed it when the address was deleted from eth0, and the
first instance crashed before transferring routes). This leaves
the node unable to reach external endpoints, causing image pull
failures for all subsequent pod creations.

Fix by creating all listen sockets (metrics, health, connectivity
check) synchronously before starting the controller. Once a socket
is bound, it survives the address transfer from eth0 to br-provider.

Additionally, refactor StartMetricsOrHealthServer to create the TCP
listener synchronously (outside the goroutine) and introduce
ServeWithListener() that accepts a pre-created net.Listener, avoiding
the internal createListener() call in controller-runtime's server.Start().

Signed-off-by: Mengxin Liu <liumengxinfly@gmail.com>

* fix(daemon): address review feedback

- Use distinct variable names for TCP/UDP listen addresses
- Apply auth filter to pprof endpoints when secureServing is enabled,
  matching the controller-runtime FilterProvider behavior
- Reuse existing package constants for HTTP server timeouts

Signed-off-by: Mengxin Liu <liumengxinfly@gmail.com>

* fix(daemon): rewrite Run() to use ServeWithListener for all components

Replace the controller-runtime based Run() implementation with a thin
wrapper that creates a net.Listener synchronously then delegates to
ServeWithListener(). This ensures ... (continued)

Run Details

0 of 108 new or added lines in 2 files covered. (0.0%)

3 existing lines in 1 file now uncovered.

12899 of 55215 relevant lines covered (23.36%)

0.27 hits per line