From 93951f10d3f80b9af9b23a23658b97e9765dcb5d Mon Sep 17 00:00:00 2001
From: Marco Hofstetter <marco.hofstetter@isovalent.com>
Date: Mon, 4 Dec 2023 10:56:23 +0100
Subject: [PATCH] connectivity test: introduce connectivity test suite timeout
 flag

Currently, the connectivity test suite command doesn't have a timeout.
Its context only gets cancelled due to interrupt of SIGTERM signal.

For most connectivity test scenarios and actions, this isn't a problem
as they have individual fine-grained timeouts (e.g. request timeout).

But there are test scenarios (e.g. `health ` - `Cilium Health Probe`)
that might run infinitely on failiure (e.g. if a Cilium Agent Pod no
longer is available).

Therefore, this commit introduces a new flag `--timeout` that provides
the possibility to configure an overall timeout for the connectivity
test suite. It defaults to `0` - meaning no timeout by default (current
behaviour).

Signed-off-by: Marco Hofstetter <marco.hofstetter@isovalent.com>
---
 connectivity/check/check.go      |  2 ++
 defaults/defaults.go             |  7 ++++++-
 internal/cli/cmd/connectivity.go | 13 +++++++++++--
 3 files changed, 19 insertions(+), 3 deletions(-)

diff --git a/connectivity/check/check.go b/connectivity/check/check.go
index e358d6a7ee..25bdefc66c 100644
--- a/connectivity/check/check.go
+++ b/connectivity/check/check.go
@@ -95,6 +95,8 @@ type Parameters struct {
 
 	ExternalTargetCANamespace string
 	ExternalTargetCAName      string
+
+	Timeout time.Duration
 }
 
 type podCIDRs struct {
diff --git a/defaults/defaults.go b/defaults/defaults.go
index af7650363b..10cde1bcb6 100644
--- a/defaults/defaults.go
+++ b/defaults/defaults.go
@@ -3,7 +3,9 @@
 
 package defaults
 
-import "time"
+import (
+	"time"
+)
 
 const (
 	// renovate: datasource=github-releases depName=cilium/cilium
@@ -135,6 +137,9 @@ const (
 	// ClustermeshMaxConnectedClusters is the default number of the maximum
 	// number of clusters that should be allowed to connect to the Clustermesh.
 	ClustermeshMaxConnectedClusters = 255
+
+	// Default timeout for Connectivity Test Suite (disabled by default)
+	ConnectivityTestSuiteTimeout = 0 * time.Minute
 )
 
 var (
diff --git a/internal/cli/cmd/connectivity.go b/internal/cli/cmd/connectivity.go
index 3578d23293..b0a78a8fa6 100644
--- a/internal/cli/cmd/connectivity.go
+++ b/internal/cli/cmd/connectivity.go
@@ -44,6 +44,7 @@ var params = check.Parameters{
 		Writer:                   os.Stdout,
 	},
 }
+
 var tests []string
 
 func newCmdConnectivityTest(hooks Hooks) *cobra.Command {
@@ -76,11 +77,17 @@ func newCmdConnectivityTest(hooks Hooks) *cobra.Command {
 				return err
 			}
 
-			ctx, _ := signal.NotifyContext(context.Background(), os.Interrupt, syscall.SIGTERM)
+			ctx, _ := signal.NotifyContext(cmd.Context(), os.Interrupt, syscall.SIGTERM)
+
+			if params.Timeout > 0 {
+				timeoutCtx, cancelCtx := context.WithTimeoutCause(ctx, params.Timeout, fmt.Errorf("connectivity test suite timeout (%s) reached", params.Timeout))
+				defer cancelCtx()
+				ctx = timeoutCtx
+			}
 
 			go func() {
 				<-ctx.Done()
-				cc.Log("Interrupt received, cancelling tests...")
+				cc.Logf("Cancellation request (%s) received, cancelling tests...", context.Cause(ctx))
 			}()
 
 			done := make(chan struct{})
@@ -187,6 +194,8 @@ func newCmdConnectivityTest(hooks Hooks) *cobra.Command {
 	cmd.Flags().MarkHidden("flush-ct")
 	cmd.Flags().StringVar(&params.SecondaryNetworkIface, "secondary-network-iface", "", "Secondary network iface name (e.g., to test NodePort BPF on multiple networks)")
 
+	cmd.Flags().DurationVar(&params.Timeout, "timeout", defaults.ConnectivityTestSuiteTimeout, "Maximum time to allow the connectivity test suite to take")
+
 	hooks.AddConnectivityTestFlags(cmd.Flags())
 
 	return cmd