VictoriaMetrics v1.148.2

Released at 2026-08-14

v1.148.x is a line of LTS releases. It contains important up-to-date bugfixes for VictoriaMetrics enterprise. All these fixes are also included in the latest community release. The v1.148.x line will be supported for at least 12 months since v1.148.0 release

• SECURITY: upgrade Go builder from Go1.26.5 to Go1.26.6. See the list of issues addressed in Go1.26.6.

• BUGFIX: vmsingle, vmselect in VictoriaMetrics cluster and vmctl: properly parse small fractional Unix timestamps in timestamp args such as start and end in /api/v1/query_range and --vm-native-filter-time-start and --vm-native-filter-time-end in vmctl. Previously, fractional Unix timestamps with the integer part below 9223372 were interpreted with the wrong unit, for example 12.0 was parsed as 12000 seconds instead of 12 seconds. See #11324.

• BUGFIX: vmsingle, vmstorage and vmselect in VictoriaMetrics cluster: persist the previous working set cache during graceful shutdown when it is likely to contain the active working set. This prevents saving an empty or cold current cache right after split-mode cache rotation, which could otherwise slow down ingestion or queries after restart until the cache warms up again. See #11299.

• BUGFIX: vmagent and vmsingle: properly assign scrape target IP address at IPv6-only networks for docker_sd_configs. See #10965.

• BUGFIX: vmagent and vmsingle: avoid suggesting the unrelated -enableTCP6 command-line flag when scraping a target over a Unix domain socket fails.

• BUGFIX: vmui: respect the custom query step specified via g0.step_input when opening a URL. Previously, it could be reset to the automatically calculated step and potentially cause dashboards to freeze. See #11137.

• BUGFIX: vmsingle and vmselect in VictoriaMetrics cluster: change the HTTP response code for Prometheus querying API requests from 422 Unprocessable Entity to 400 Bad Request when request parameters are missing or incorrect. See #11330.

• BUGFIX: vmsingle and vminsert in VictoriaMetrics cluster: skip labels with empty name at /api/v1/import. Previously such a label replaced the metric name, so a series sent with "metric":{"__name__":"foo","":"bar"} was stored under the name bar. Other ingestion protocols already skip such labels. See #4962.

• BUGFIX: vmalert: rename vmalert_rule_group_results_limit back to vmalert_group_rule_results_limit. The metric was introduced in v1.147.0 but was accidentally given the wrong name. See #11179.

• BUGFIX: vmalert: properly update group-level eval_delay and eval_alignment for existing groups during runtime when config reload is triggered periodically or manually via /-/reload. Previously, these settings weren't updated after config reload during runtime. See #11374.

• BUGFIX: vmselect in VictoriaMetrics cluster: scale the default -search.maxConcurrentRequests with the number of available CPU cores instead of capping it at 16. See #11191. Thanks to @Dhru1Tanna for contribution.