diff --git a/app/vmauth/auth_config.go b/app/vmauth/auth_config.go
index e2e38edf0f1c..b4afeee31278 100644
--- a/app/vmauth/auth_config.go
+++ b/app/vmauth/auth_config.go
@@ -693,6 +693,14 @@ func loadAuthConfig() (bool, error) {
 	authConfig.Store(ac)
 	authConfigData.Store(&data)
 	authUsers.Store(&m)
+	if prevAc != nil {
+		// explicilty unregister metrics, since all summary type metrics
+		// are registered at global state of metrics package
+		// and must be removed from it to release memory.
+		// Metrics must be unregistered only after atomic.Value.Store calls above
+		// Otherwise it may lead to metric gaps, since UnregisterAllMetrics is slow operation
+		prevAc.ms.UnregisterAllMetrics()
+	}
 
 	return true, nil
 }
diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md
index d14ab38ec0a8..eea50f1b098a 100644
--- a/docs/CHANGELOG.md
+++ b/docs/CHANGELOG.md
@@ -50,6 +50,7 @@ See also [LTS releases](https://docs.victoriametrics.com/lts-releases/).
 * BUGFIX: [stream aggregation](https://docs.victoriametrics.com/stream-aggregation/): set correct suffix `<output>_prometheus` for aggregation outputs [increase_prometheus](https://docs.victoriametrics.com/stream-aggregation/#increase_prometheus) and [total_prometheus](https://docs.victoriametrics.com/stream-aggregation/#total_prometheus). Before, outputs `total` and `total_prometheus` or `increase` and `increase_prometheus` had the same suffix.
 * BUGFIX: properly estimate the needed memory for query execution if it has the format [`aggr_func`](https://docs.victoriametrics.com/metricsql/#aggregate-functions)([`rollup_func[d]`](https://docs.victoriametrics.com/metricsql/#rollup-functions) (for example, `sum(rate(request_duration_seconds_bucket[5m]))`). This should allow performing aggregations over bigger number of time series when VictoriaMetrics runs in environments with small amounts of available memory. The issue has been introduced in [this commit](https://github.com/VictoriaMetrics/VictoriaMetrics/commit/5138eaeea0791caa34bcfab410e0ca9cd253cd8f) in [v1.83.0](https://docs.victoriametrics.com/changelog_2022/#v1830).
 * BUGFIX: [Single-node VictoriaMetrics](https://docs.victoriametrics.com/) and `vmstorage` in [VictoriaMetrics cluster](https://docs.victoriametrics.com/cluster-victoriametrics/): correctly apply `-inmemoryDataFlushInterval` when it's set to minimum supported value 1s.
+* BUGFIX: [vmauth](https://docs.victoriametrics.com/vmauth/): properly release memory used for metrics during config reload. See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/6247).
 
 * DEPRECATION: [vmagent](https://docs.victoriametrics.com/vmagent/): removed deprecated `-remoteWrite.multitenantURL` flag from vmagent. This flag was deprecated since [v1.96.0](https://github.com/VictoriaMetrics/VictoriaMetrics/releases/tag/v1.96.0). Use `-enableMultitenantHandlers` instead, as it is easier to use and combine with [multitenant URL at vminsert](https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html#multitenancy-via-labels). See these [docs for details](https://docs.victoriametrics.com/vmagent.html#multitenancy).