mirror of
https://github.com/openshift/installer.git
synced 2026-02-05 06:46:36 +01:00
fc462861f8
This is some quick docs to allow the use of `rr` to troubleshoot and trace difficult cluster api provider problems. Assisted-by: cursor
119 lines
3.4 KiB
Diff
119 lines
3.4 KiB
Diff
diff --git a/pkg/clusterapi/internal/process/process.go b/pkg/clusterapi/internal/process/process.go
|
|
index 29d4c37619..ce24efd886 100644
|
|
--- a/pkg/clusterapi/internal/process/process.go
|
|
+++ b/pkg/clusterapi/internal/process/process.go
|
|
@@ -231,26 +231,41 @@ func (ps *State) Stop() error {
|
|
if ps.Cmd == nil {
|
|
return nil
|
|
}
|
|
+
|
|
if done, err := ps.Exited(); done {
|
|
+
|
|
+ logrus.Infof("Process %s exited *withOUT* error", ps.Path)
|
|
+
|
|
if err != nil {
|
|
logrus.Warnf("process %s exited with error: %v", path.Base(ps.Path), err)
|
|
}
|
|
return nil
|
|
}
|
|
- if err := ps.Cmd.Process.Signal(syscall.SIGTERM); err != nil {
|
|
- return fmt.Errorf("unable to signal for process %s to stop: %w", ps.Path, err)
|
|
- }
|
|
|
|
- timedOut := time.After(ps.StopTimeout)
|
|
- select {
|
|
- case <-ps.waitDone:
|
|
- break
|
|
- case <-timedOut:
|
|
- if err := ps.Cmd.Process.Signal(syscall.SIGKILL); err != nil {
|
|
- return fmt.Errorf("unable to signal for process %s to stop: %w", ps.Path, err)
|
|
- }
|
|
- return fmt.Errorf("timeout waiting for process %s to stop, sent SIGKILL", path.Base(ps.Path))
|
|
+ if err := syscall.Kill(-ps.Cmd.Process.Pid, syscall.SIGTERM); err != nil {
|
|
+ return fmt.Errorf("unable to signal for group process %s to stop: %w", ps.Path, err)
|
|
}
|
|
+
|
|
+ /*
|
|
+ NOTE: rr doesn't like to be SIGKILL'ed and have a non-incomplete trace
|
|
+
|
|
+
|
|
+ if err := ps.Cmd.Process.Signal(syscall.SIGTERM); err != nil {
|
|
+ return fmt.Errorf("unable to signal for process %s to stop: %w", ps.Path, err)
|
|
+ }
|
|
+
|
|
+ timedOut := time.After(ps.StopTimeout)
|
|
+ select {
|
|
+ case <-ps.waitDone:
|
|
+ break
|
|
+ case <-timedOut:
|
|
+ if err := ps.Cmd.Process.Signal(syscall.SIGTERM); err != nil {
|
|
+ return fmt.Errorf("unable to signal for process %s to stop: %w", ps.Path, err)
|
|
+ }
|
|
+ return fmt.Errorf("timeout waiting for process %s to stop, sent SIGKILL", path.Base(ps.Path))
|
|
+ }
|
|
+
|
|
+ */
|
|
ps.ready = false
|
|
return nil
|
|
}
|
|
diff --git a/pkg/clusterapi/system.go b/pkg/clusterapi/system.go
|
|
index 9fd5fcb2bb..80336a91d7 100644
|
|
--- a/pkg/clusterapi/system.go
|
|
+++ b/pkg/clusterapi/system.go
|
|
@@ -427,11 +427,15 @@ func (c *system) Run(ctx context.Context) error { //nolint:gocyclo
|
|
logrus.Info("Shutting down local Cluster API controllers...")
|
|
for _, ct := range controllers {
|
|
if ct.state != nil {
|
|
- if err := ct.state.Stop(); err != nil {
|
|
- logrus.Warnf("Failed to stop controller: %s: %v", ct.Name, err)
|
|
- continue
|
|
- }
|
|
- logrus.Infof("Stopped controller: %s", ct.Name)
|
|
+ logrus.Warn("STOP CONTROLLER")
|
|
+ /*
|
|
+ if err := ct.state.Stop(); err != nil {
|
|
+ logrus.Warnf("Failed to stop controller: %s: %v", ct.Name, err)
|
|
+ continue
|
|
+ }
|
|
+ logrus.Infof("Stopped controller: %s", ct.Name)
|
|
+
|
|
+ */
|
|
}
|
|
}
|
|
}()
|
|
@@ -456,6 +460,10 @@ func (c *system) Client() client.Client {
|
|
|
|
// Teardown shuts down the local capi control plane and all its controllers.
|
|
func (c *system) Teardown() {
|
|
+
|
|
+ // We don't want to teardown any controllers while using rr
|
|
+ return
|
|
+
|
|
c.Lock()
|
|
defer c.Unlock()
|
|
|
|
@@ -651,6 +659,23 @@ func (c *system) runController(ctx context.Context, ct *controller) error {
|
|
}
|
|
}
|
|
|
|
+ if ct.Provider != nil {
|
|
+ capvPath := ct.Path
|
|
+ capvArgs := ct.Args
|
|
+
|
|
+ ct.Path = "/usr/bin/rr"
|
|
+ rrArgs := []string{
|
|
+ "record",
|
|
+ "--wait",
|
|
+ "--disable-avx-512",
|
|
+ "--bind-to-cpu=0",
|
|
+ }
|
|
+
|
|
+ rrArgs = append(rrArgs, capvPath)
|
|
+ rrArgs = append(rrArgs, capvArgs...)
|
|
+ ct.Args = rrArgs
|
|
+ }
|
|
+
|
|
// Create the process state.
|
|
pr := &process.State{
|
|
Path: ct.Path,
|