br: fix the issue that config not paused with ttl correctly (#31725)

close #31733

br/pkg/pdutil/pd.go

@@ -38,6 +38,7 @@ const (
 	schedulerPrefix      = "pd/api/v1/schedulers"
 	maxMsgSize           = int(128 * units.MiB) // pd.ScanRegion may return a large response
 	scheduleConfigPrefix = "pd/api/v1/config/schedule"
+	configPrefix         = "pd/api/v1/config"
 	pauseTimeout         = 5 * time.Minute
 
 	// pd request retry time when connection fail
@@ -541,12 +542,20 @@ func (p *PdController) UpdatePDScheduleConfig(ctx context.Context) error {
 func (p *PdController) doUpdatePDScheduleConfig(
 	ctx context.Context, cfg map[string]interface{}, post pdHTTPRequest, prefixs ...string,
 ) error {
-	prefix := scheduleConfigPrefix
+	prefix := configPrefix
 	if len(prefixs) != 0 {
 		prefix = prefixs[0]
 	}
+	newCfg := make(map[string]interface{})
+	for k, v := range cfg {
+		// if we want use ttl, we need use config prefix first.
+		// which means cfg should transfer from "max-merge-region-keys" to "schedule.max-merge-region-keys".
+		sc := fmt.Sprintf("schedule.%s", k)
+		newCfg[sc] = v
+	}
+
 	for _, addr := range p.addrs {
-		reqData, err := json.Marshal(cfg)
+		reqData, err := json.Marshal(newCfg)
 		if err != nil {
 			return errors.Trace(err)
 		}
@@ -562,7 +571,7 @@ func (p *PdController) doUpdatePDScheduleConfig(
 
 func (p *PdController) doPauseConfigs(ctx context.Context, cfg map[string]interface{}, post pdHTTPRequest) error {
 	// pause this scheduler with 300 seconds
-	prefix := fmt.Sprintf("%s?ttlSecond=%.0f", scheduleConfigPrefix, pauseTimeout.Seconds())
+	prefix := fmt.Sprintf("%s?ttlSecond=%.0f", configPrefix, pauseTimeout.Seconds())
 	return p.doUpdatePDScheduleConfig(ctx, cfg, post, prefix)
 }
 
@@ -584,7 +593,7 @@ func restoreSchedulers(ctx context.Context, pd *PdController, clusterCfg Cluster
 	prefix := make([]string, 0, 1)
 	if pd.isPauseConfigEnabled() {
 		// set config's ttl to zero, make temporary config invalid immediately.
-		prefix = append(prefix, fmt.Sprintf("%s?ttlSecond=%d", scheduleConfigPrefix, 0))
+		prefix = append(prefix, fmt.Sprintf("%s?ttlSecond=%d", configPrefix, 0))
 	}
 	// reset config with previous value.
 	if err := pd.doUpdatePDScheduleConfig(ctx, mergeCfg, pdRequest, prefix...); err != nil {

br/tests/br_other/run.sh

@@ -89,10 +89,23 @@ run_curl "https://localhost:$PPROF_PORT/debug/pprof/trace?seconds=1" &>/dev/null
 echo "pprof started..."
 
 run_curl https://$PD_ADDR/pd/api/v1/config/schedule | grep '"disable": false'
-run_curl https://$PD_ADDR/pd/api/v1/config/schedule | jq '."enable-location-replacement"' | grep "false"
-run_curl https://$PD_ADDR/pd/api/v1/config/schedule | jq '."max-pending-peer-count"' | grep "2147483647"
-run_curl https://$PD_ADDR/pd/api/v1/config/schedule | jq '."max-merge-region-size"' | grep -E "^0$"
-run_curl https://$PD_ADDR/pd/api/v1/config/schedule | jq '."max-merge-region-keys"' | grep -E "^0$"
+
+# after apply the pause api. these configs won't change any more.
+# run_curl https://$PD_ADDR/pd/api/v1/config/schedule | jq '."enable-location-replacement"' | grep "false"
+# run_curl https://$PD_ADDR/pd/api/v1/config/schedule | jq '."max-pending-peer-count"' | grep "2147483647"
+# run_curl https://$PD_ADDR/pd/api/v1/config/schedule | jq '."max-merge-region-size"' | grep -E "^0$"
+# run_curl https://$PD_ADDR/pd/api/v1/config/schedule | jq '."max-merge-region-keys"' | grep -E "^0$"
+
+# after https://github.com/tikv/pd/pull/4781 merged.
+# we can use a hack way to check whether we pause config succeed.
+# By setting a paused config again and expect to fail with a clear message.
+run_pd_ctl -u https://$PD_ADDR config set max-merge-region-size 0 | grep -q "need to clean up TTL first for schedule.max-merge-region-size"
+run_pd_ctl -u https://$PD_ADDR config set max-merge-region-keys 0 | grep -q "need to clean up TTL first for schedule.max-merge-region-keys"
+run_pd_ctl -u https://$PD_ADDR config set max-pending-peer-count 0 | grep -q "need to clean up TTL first for schedule.max-pending-peer-count"
+run_pd_ctl -u https://$PD_ADDR config set enable-location-replacement false | grep -q "need to clean up TTL first for schedule.enable-location-replacement"
+run_pd_ctl -u https://$PD_ADDR config set leader-schedule-limit 0 | grep -q "need to clean up TTL first for schedule.leader-schedule-limit"
+run_pd_ctl -u https://$PD_ADDR config set region-schedule-limit 0 | grep -q "need to clean up TTL first for schedule.region-schedule-limit"
+run_pd_ctl -u https://$PD_ADDR config set max-snapshot-count 0 | grep -q "need to clean up TTL first for schedule.max-snapshot-count"
 
 backup_fail=0
 # generate 1.sst to make another backup failed.