Add granular termination reason in container termination message

Related with tektoncd#7223.

To report specific Steps termination reasons we need to know why its continer finished; we use the termination message to store a new "state" with this information. We evaluated changing the container `reason` directly, but looks like k8s doesn't allow this.

cmd/entrypoint/main.go

@@ -176,7 +176,7 @@ func main() {
 	if err := e.Go(); err != nil {
 		breakpointExitPostFile := e.PostFile + breakpointExitSuffix
 		switch t := err.(type) { //nolint:errorlint // checking for multiple types with errors.As is ugly.
-		case skipError:
+		case entrypoint.SkipError:
 			log.Print("Skipping step because a previous step failed")
 			os.Exit(1)
 		case termination.MessageLengthError:

cmd/entrypoint/waiter.go

@@ -71,7 +71,7 @@ func (rw *realWaiter) Wait(ctx context.Context, file string, expectContent bool,
 			if breakpointOnFailure {
 				return nil
 			}
-			return skipError("error file present, bail and skip the step")
+			return entrypoint.SkipErroPreviousStepFailed
 		}
 		select {
 		case <-ctx.Done():
@@ -86,9 +86,3 @@ func (rw *realWaiter) Wait(ctx context.Context, file string, expectContent bool,
 		}
 	}
 }
-
-type skipError string
-
-func (e skipError) Error() string {
-	return string(e)
-}

pkg/entrypoint/entrypointer.go

@@ -53,11 +53,19 @@ func (e ContextError) Error() string {
 	return string(e)
 }
 
+type SkipError string
+
+func (e SkipError) Error() string {
+	return string(e)
+}
+
 var (
 	// ErrContextDeadlineExceeded is the error returned when the context deadline is exceeded
 	ErrContextDeadlineExceeded = ContextError(context.DeadlineExceeded.Error())
 	// ErrContextCanceled is the error returned when the context is canceled
 	ErrContextCanceled = ContextError(context.Canceled.Error())
+	// SkipErroPreviousStepFailed is the error returned when the step is skipped due to previous step error
+	SkipErroPreviousStepFailed = SkipError("error file present, bail and skip the step")
 )
 
 // IsContextDeadlineError determine whether the error is context deadline
@@ -160,6 +168,11 @@ func (e Entrypointer) Go() error {
 				Value:      time.Now().Format(timeFormat),
 				ResultType: result.InternalTektonResultType,
 			})
+
+			if errors.Is(err, SkipErroPreviousStepFailed) {
+				output = append(output, e.getTerminationReason(result.TerminationReasonSkipped))
+			}
+
 			return err
 		}
 	}
@@ -189,28 +202,22 @@ func (e Entrypointer) Go() error {
 			}
 		}()
 		err = e.Runner.Run(ctx, e.Command...)
-		if errors.Is(err, ErrContextDeadlineExceeded) {
-			output = append(output, result.RunResult{
-				Key:        "Reason",
-				Value:      "TimeoutExceeded",
-				ResultType: result.InternalTektonResultType,
-			})
-		}
 	}
 
 	var ee *exec.ExitError
 	switch {
 	case err != nil && errors.Is(err, ErrContextCanceled):
 		logger.Info("Step was canceling")
-		output = append(output, result.RunResult{
-			Key:        "Reason",
-			Value:      "Cancelled",
-			ResultType: result.InternalTektonResultType,
-		})
 		e.WritePostFile(e.PostFile, ErrContextCanceled)
 		e.WriteExitCodeFile(e.StepMetadataDir, syscall.SIGKILL.String())
+		output = append(output, e.getTerminationReason(result.TerminationReasonCancelled))
+
+	case errors.Is(err, ErrContextDeadlineExceeded):
+		output = append(output, e.getTerminationReason(result.TerminationReasonTimeoutExceeded))
+
 	case err != nil && e.BreakpointOnFailure:
 		logger.Info("Skipping writing to PostFile")
+
 	case e.OnError == ContinueOnError && errors.As(err, &ee):
 		// with continue on error and an ExitError, write non-zero exit code and a post file
 		exitCode := strconv.Itoa(ee.ExitCode())
@@ -221,13 +228,18 @@ func (e Entrypointer) Go() error {
 		})
 		e.WritePostFile(e.PostFile, nil)
 		e.WriteExitCodeFile(e.StepMetadataDir, exitCode)
+		output = append(output, e.getTerminationReason(result.TerminationReasonContinued))
+
 	case err == nil:
 		// if err is nil, write zero exit code and a post file
 		e.WritePostFile(e.PostFile, nil)
 		e.WriteExitCodeFile(e.StepMetadataDir, "0")
+		output = append(output, e.getTerminationReason(result.TerminationReasonCompleted))
+
 	default:
 		// for a step without continue on error and any error, write a post file with .err
 		e.WritePostFile(e.PostFile, err)
+		output = append(output, e.getTerminationReason(result.TerminationReasonError))
 	}
 
 	// strings.Split(..) with an empty string returns an array that contains one element, an empty string.
@@ -317,3 +329,11 @@ func (e Entrypointer) waitingCancellation(ctx context.Context, cancel context.Ca
 	cancel()
 	return nil
 }
+
+func (e Entrypointer) getTerminationReason(terminationReason string) result.RunResult {
+	return result.RunResult{
+		Key:        "Reason",
+		Value:      terminationReason,
+		ResultType: result.InternalTektonResultType,
+	}
+}

pkg/result/result.go

@@ -33,6 +33,19 @@ const (
 	InternalTektonResultType = 3
 	// UnknownResultType default unknown result type value
 	UnknownResultType = 10
+
+	// TerminationReasonCompleted indicates a step finished successfully.
+	TerminationReasonCompleted = "Completed"
+	// TerminationReasonContinued indicates a step errored but was ignored since onError was set to continue.
+	TerminationReasonContinued = "Continued"
+	// TerminationReasonError indicates a step failed with a non-zero exit code.
+	TerminationReasonError = "Error"
+	// TerminationReasonTimeoutExceeded indicates a step execution timed out.
+	TerminationReasonTimeoutExceeded = "TimeoutExceeded"
+	// TerminationReasonSkipped indicates a step execution was skipped due to previous step failed.
+	TerminationReasonSkipped = "Skipped"
+	// TerminationReasonCancelled indicates a step was cancelled by user.
+	TerminationReasonCancelled = "Cancelled"
 )
 
 // RunResult is used to write key/value pairs to TaskRun pod termination messages.