PromTool: Support Scrape Interval Lint Checking

Signed-off-by: zhaowang <zhaowang@apac.freewheel.com>

cmd/promtool/main.go

@@ -68,11 +68,15 @@ const (
 	// Exit code 3 is used for "one or more lint issues detected".
 	lintErrExitCode = 3
 
-	lintOptionAll            = "all"
-	lintOptionDuplicateRules = "duplicate-rules"
-	lintOptionNone           = "none"
-	checkHealth              = "/-/healthy"
-	checkReadiness           = "/-/ready"
+	lintOptionAll                = "all"
+	lintOptionDuplicateRules     = "duplicate-rules"
+	lintOptionLongScrapeInterval = "long-scrape-inerval"
+	lintOptionNone               = "none"
+	checkHealth                  = "/-/healthy"
+	checkReadiness               = "/-/ready"
+
+	// Remove when the lookback delta can be set as a command line flag on PromTool.
+	defaultLookbackDelta = 5 * time.Minute
 )
 
 var lintOptions = []string{lintOptionAll, lintOptionDuplicateRules, lintOptionNone}
@@ -443,9 +447,10 @@ func checkExperimental(f bool) {
 var errLint = fmt.Errorf("lint error")
 
 type lintConfig struct {
-	all            bool
-	duplicateRules bool
-	fatal          bool
+	all                bool
+	duplicateRules     bool
+	longScrapeInterval bool
+	fatal              bool
 }
 
 func newLintConfig(stringVal string, fatal bool) lintConfig {
@@ -459,6 +464,8 @@ func newLintConfig(stringVal string, fatal bool) lintConfig {
 			ls.all = true
 		case lintOptionDuplicateRules:
 			ls.duplicateRules = true
+		case lintOptionLongScrapeInterval:
+			ls.longScrapeInterval = true
 		case lintOptionNone:
 		default:
 			fmt.Printf("WARNING: unknown lint option %s\n", setting)
@@ -471,6 +478,10 @@ func (ls lintConfig) lintDuplicateRules() bool {
 	return ls.all || ls.duplicateRules
 }
 
+func (ls lintConfig) lintLongScrapeInterval() bool {
+	return ls.all || ls.longScrapeInterval
+}
+
 // CheckServerStatus - healthy & ready.
 func CheckServerStatus(serverURL *url.URL, checkEndpoint string, roundTripper http.RoundTripper) error {
 	if serverURL.Scheme == "" {
@@ -514,10 +525,10 @@ func CheckConfig(agentMode, checkSyntaxOnly bool, lintSettings lintConfig, files
 	hasErrors := false
 
 	for _, f := range files {
-		ruleFiles, err := checkConfig(agentMode, f, checkSyntaxOnly)
+		ruleFiles, err := checkConfig(agentMode, f, checkSyntaxOnly, lintSettings)
 		if err != nil {
 			fmt.Fprintln(os.Stderr, "  FAILED:", err)
-			hasErrors = true
+			hasErrors = !errors.Is(err, errLint)
 			failed = true
 		} else {
 			if len(ruleFiles) > 0 {
@@ -571,7 +582,7 @@ func checkFileExists(fn string) error {
 	return err
 }
 
-func checkConfig(agentMode bool, filename string, checkSyntaxOnly bool) ([]string, error) {
+func checkConfig(agentMode bool, filename string, checkSyntaxOnly bool, lintSettings lintConfig) ([]string, error) {
 	fmt.Println("Checking", filename)
 
 	cfg, err := config.LoadFile(filename, agentMode, false, log.NewNopLogger())
@@ -611,6 +622,13 @@ func checkConfig(agentMode bool, filename string, checkSyntaxOnly bool) ([]strin
 	}
 
 	for _, scfg := range scfgs {
+		if lintSettings.lintLongScrapeInterval() {
+			if scfg.ScrapeInterval >= model.Duration(defaultLookbackDelta) {
+				errMessage := fmt.Sprintf("Long Scrape Interval found. Data point will be marked as stale. Job: %s. Interval: %s", scfg.JobName, scfg.ScrapeInterval.String())
+				return nil, fmt.Errorf("%w %s", errLint, errMessage)
+			}
+		}
+
 		if !checkSyntaxOnly && scfg.HTTPClientConfig.Authorization != nil {
 			if err := checkFileExists(scfg.HTTPClientConfig.Authorization.CredentialsFile); err != nil {
 				return nil, fmt.Errorf("error checking authorization credentials or bearer token file %q: %w", scfg.HTTPClientConfig.Authorization.CredentialsFile, err)

cmd/promtool/main_test.go

@@ -219,7 +219,7 @@ func TestCheckTargetConfig(t *testing.T) {
 	}
 	for _, test := range cases {
 		t.Run(test.name, func(t *testing.T) {
-			_, err := checkConfig(false, "testdata/"+test.file, false)
+			_, err := checkConfig(false, "testdata/"+test.file, false, newLintConfig(lintOptionNone, false))
 			if test.err != "" {
 				require.Equalf(t, test.err, err.Error(), "Expected error %q, got %q", test.err, err.Error())
 				return
@@ -302,7 +302,7 @@ func TestCheckConfigSyntax(t *testing.T) {
 	}
 	for _, test := range cases {
 		t.Run(test.name, func(t *testing.T) {
-			_, err := checkConfig(false, "testdata/"+test.file, test.syntaxOnly)
+			_, err := checkConfig(false, "testdata/"+test.file, test.syntaxOnly, newLintConfig(lintOptionNone, false))
 			expectedErrMsg := test.err
 			if strings.Contains(runtime.GOOS, "windows") {
 				expectedErrMsg = test.errWindows
@@ -336,7 +336,7 @@ func TestAuthorizationConfig(t *testing.T) {
 
 	for _, test := range cases {
 		t.Run(test.name, func(t *testing.T) {
-			_, err := checkConfig(false, "testdata/"+test.file, false)
+			_, err := checkConfig(false, "testdata/"+test.file, false, newLintConfig(lintOptionNone, false))
 			if test.err != "" {
 				require.Contains(t, err.Error(), test.err, "Expected error to contain %q, got %q", test.err, err.Error())
 				return
@@ -414,7 +414,7 @@ func TestExitCodes(t *testing.T) {
 				t.Run(strconv.FormatBool(lintFatal), func(t *testing.T) {
 					args := []string{"-test.main", "check", "config", "testdata/" + c.file}
 					if lintFatal {
-						args = append(args, "--lint-fatal")
+						args = append(args, "--lint-fatal", "--lint=all")
 					}
 					tool := exec.Command(promtoolPath, args...)
 					err := tool.Run()
@@ -531,6 +531,13 @@ func TestCheckRules(t *testing.T) {
 		exitCode := CheckRules(newLintConfig(lintOptionDuplicateRules, true))
 		require.Equal(t, lintErrExitCode, exitCode, "")
 	})
+
+	t.Run("config-lint-fatal", func(t *testing.T) {
+		_, err := checkConfig(false, "./testdata/prometheus-config.lint.yml", false, newLintConfig(lintOptionLongScrapeInterval, false))
+		expectedErrMsg := "lint error Long Scrape Interval found. Data point will be marked as stale. Job: long_scrape_interval_test. Interval: 1h"
+		require.Equalf(t, expectedErrMsg, err.Error(), "Expected error %q, got %q", expectedErrMsg, err.Error())
+		return
+	})
 }
 
 func TestCheckRulesWithRuleFiles(t *testing.T) {

cmd/promtool/testdata/prometheus-config.lint.yml

@@ -1,2 +1,3 @@
-rule_files:
-  - prometheus-rules.lint.yml
+scrape_configs:
+  - job_name: long_scrape_interval_test
+    scrape_interval: 1h