From 2b1a9025469fc5a9915d22bae0c3c886f946737f Mon Sep 17 00:00:00 2001
From: jmontesi <jmontesi@redhat.com>
Date: Fri, 27 Oct 2023 12:12:05 +0200
Subject: [PATCH] performance: avoid failing test case if a process cannot be
 found

It applies to rt-apps-no-exec-probes.

The reason a process cannot be found after it has been properly
retrieved previously could be that the process has finished. So, in this
case it seems more adequate to just ignore it than to produce an error.
If the error is of another kind the container is added to the non
compliant list but the processing for that container continues instead
of breaking, to have more info on the processes.

The commit also changes the verification order in that test case, making
the faster check of exec probes at the beginning, skipping the slower
part in which the processes are list.

Finally, the compliant containers are added to its list with the reason
for compliance.
---
 cnf-certification-test/performance/suite.go | 25 +++++++++++++++++++--
 pkg/scheduling/scheduling.go                |  3 ++-
 2 files changed, 25 insertions(+), 3 deletions(-)

diff --git a/cnf-certification-test/performance/suite.go b/cnf-certification-test/performance/suite.go
index bef86ab94..792307a84 100644
--- a/cnf-certification-test/performance/suite.go
+++ b/cnf-certification-test/performance/suite.go
@@ -19,6 +19,7 @@ package performance
 import (
 	"fmt"
 	"strconv"
+	"strings"
 
 	"github.com/onsi/ginkgo/v2"
 	"github.com/sirupsen/logrus"
@@ -232,30 +233,50 @@ func testSchedulingPolicyInCPUPool(env *provider.TestEnvironment,
 	testhelper.AddTestResultReason(compliantContainersPids, nonCompliantContainersPids, tnf.ClaimFilePrintf, ginkgo.Fail)
 }
 
+const noProcessFoundErrMsg = "No such process"
+
 func testRtAppsNoExecProbes(env *provider.TestEnvironment, cuts []*provider.Container) {
 	var compliantObjects []*testhelper.ReportObject
 	var nonCompliantObjects []*testhelper.ReportObject
 	for _, cut := range cuts {
+		if !cut.HasExecProbes() {
+			compliantObjects = append(compliantObjects, testhelper.NewContainerReportObject(cut.Namespace, cut.Podname, cut.Name, "Container does not define exec probes", true))
+			continue
+		}
+
 		processes, err := crclient.GetContainerProcesses(cut, env)
 		if err != nil {
 			tnf.ClaimFilePrintf("Could not determine the processes pids for container %s, err: %v", cut, err)
 			nonCompliantObjects = append(nonCompliantObjects, testhelper.NewContainerReportObject(cut.Namespace, cut.Podname, cut.Name, "Could not determine the processes pids for container", false))
 			break
 		}
+
+		allProcessesCompliant := true
 		for _, p := range processes {
 			schedPolicy, _, err := scheduling.GetProcessCPUScheduling(p.Pid, cut)
 			if err != nil {
+				// If the process does not exist anymore it means that it has finished since the time the process list
+				// was retrieved. In this case, just ignore the error and continue processing the rest of the processes.
+				if strings.Contains(err.Error(), noProcessFoundErrMsg) {
+					continue
+				}
 				tnf.ClaimFilePrintf("Could not determine the scheduling policy for container %s (pid=%v), err: %v", cut, p.Pid, err)
 				nonCompliantObjects = append(nonCompliantObjects, testhelper.NewContainerReportObject(cut.Namespace, cut.Podname, cut.Name, "Could not determine the scheduling policy for container", false).
 					AddField(testhelper.ProcessID, strconv.Itoa(p.Pid)))
-				break
+				allProcessesCompliant = false
+				continue
 			}
-			if scheduling.PolicyIsRT(schedPolicy) && cut.HasExecProbes() {
+			if scheduling.PolicyIsRT(schedPolicy) {
 				tnf.ClaimFilePrintf("Pod %s/Container %s defines exec probes while having a RT scheduling policy for pid %d", cut.Podname, cut, p.Pid)
 				nonCompliantObjects = append(nonCompliantObjects, testhelper.NewContainerReportObject(cut.Namespace, cut.Podname, cut.Name, "Container defines exec probes while having a RT scheduling policy", false).
 					AddField(testhelper.ProcessID, strconv.Itoa(p.Pid)))
+				allProcessesCompliant = false
 			}
 		}
+
+		if allProcessesCompliant {
+			compliantObjects = append(compliantObjects, testhelper.NewContainerReportObject(cut.Namespace, cut.Podname, cut.Name, "Container defines exec probes but does not have a RT scheduling policy", true))
+		}
 	}
 	testhelper.AddTestResultReason(compliantObjects, nonCompliantObjects, tnf.ClaimFilePrintf, ginkgo.Fail)
 }
diff --git a/pkg/scheduling/scheduling.go b/pkg/scheduling/scheduling.go
index bf7ae338b..1e09eab8d 100644
--- a/pkg/scheduling/scheduling.go
+++ b/pkg/scheduling/scheduling.go
@@ -132,7 +132,8 @@ func GetProcessCPUScheduling(pid int, testContainer *provider.Container) (schedu
 
 	stdout, stderr, err := ch.ExecCommandContainer(ctx, command)
 	if err != nil || stderr != "" {
-		return schedulePolicy, InvalidPriority, fmt.Errorf("command %q failed to run in debug pod %s (node %s): %v", command, ctx.GetPodName(), testContainer.NodeName, err)
+		return schedulePolicy, InvalidPriority, fmt.Errorf("command %q failed to run in debug pod %s (node %s): %v (stderr: %v)",
+			command, ctx.GetPodName(), testContainer.NodeName, err, stderr)
 	}
 
 	schedulePolicy, schedulePriority, err = parseSchedulingPolicyAndPriority(stdout)