An empty case in a Go switch: scrape mode changes on reload in sql_exporter

A reload that switched sql_exporter from one target to jobs returned success and kept the old target. I fixed it in PR #1119, merged 2026-10-17, 24 minutes after I opened it. A case with no body ended a Go switch, and Reload returned nil.

When it happens

The exporter runs with a single target:. The config is edited to use jobs:, and the exporter gets SIGHUP or a request to /reload. The endpoint exists only with -web.enable-reload.

Both paths log only two warnings: Reloading collectors has started... and Connections will not be changed upon the restart of the exporter. No error is logged or returned. /reload answers 200, and the scrapes keep serving the old target.

The other direction, jobs: to a single target:, was refused. The log shows changing scrape mode is not allowed. Please restart the exporter and /reload answers 500.

How Reload picks a case

Reload in reload.go loads the new config, then runs a switch on the old and new scrape mode. Target to target calls reloadTarget. Jobs to jobs calls reloadJobs. The two mode changes look meant to share one error. Anything else logs a warning.

A Go switch runs one case and does not fall through. A case with no body ends the switch, and the function goes on to its final return nil.

Why the gap existed

The empty case came with the first version of reload.go, in PR #548, merged in June 2024. Two stacked cases read as if they share a body. In Go they do not.

The only test of Reload I found, TestReloadMemoryLeak in reload_test.go, reloads one jobs config 500 times behind the integration build tag. It never changes the scrape mode.

The fix

Put both conditions in one case. This is the hunk from reload.go:

@@ -38,8 +38,8 @@ func Reload(e Exporter, configFile *string) error {
 		if err = reloadJobs(e, configNext, configCurrent); err != nil {
 			return err
 		}
-	case configCurrent.Target != nil && len(configNext.Jobs) > 0:
-	case len(configCurrent.Jobs) > 0 && configNext.Target != nil:
+	case configCurrent.Target != nil && len(configNext.Jobs) > 0,
+		len(configCurrent.Jobs) > 0 && configNext.Target != nil:
 		return errors.New("changing scrape mode is not allowed. Please restart the exporter")
 	default:
 		slog.Warn("No target or jobs have been found - nothing to reload")

The error is the one the jobs-to-target case already returned. Nothing else in reload.go changes.

The test

TestReloadScrapeModeChange in exporter_test.go writes two configs, one with a single target and one with jobs. It builds an exporter from the first and reloads the second, then does the opposite. Each direction must return an error that contains changing scrape mode. It needs no database.

Before the fix, TargetToJobs failed with Reload = <nil>, want a scrape mode change error, and JobsToTarget passed. Over 10 runs, TargetToJobs failed 10 times and JobsToTarget passed 10 times. After the fix both passed in all 10 runs.

What the maintainer did

@burningalchemist approved with “🚀” and merged 22 seconds later. It was my fifth merged sql_exporter PR, after #1115, #1116, #1117 and #1118.

The lesson

• An empty case in Go is a silent exit, not a fall-through. To share a body, list both conditions in one case, separated by a comma.
• Test both directions of a symmetric rule. Here one direction already worked, so a test of only that direction would have passed.
• Return an error for a request you cannot honour. A success reply says the new config took effect.

If a switch in your own code has a case with no body, reply on X.