v1.82.0

版本发布时间: 2022-10-07 09:20:29

VictoriaMetrics/VictoriaMetrics最新发布版本:v1.101.0(2024-04-26 17:06:08)

Changes

v1.82.0

Released at 07-10-2022

It isn't recommended to use VictoriaMetrics and vmagent v1.82.0 because of the bug, which may result in incorrect query results and relabeling results. Upgrade to v1.82.1 instead.

Update note 1: this release changes data format for /api/v1/export/native in incompatible way, so it cannot be imported into older version of VictoriaMetrics via /api/v1/import/native.

Update note 2: vmalert changes default value for command-line flag -datasource.queryStep from 0s to 5m. The change supposed to improve reliability of the rules evaluation when evaluation interval is lower than scraping interval.

Update note 3: vm_account_id and vm_project_id labels must be passed to tcp-based Graphite, InfluxDB and OpenTSDB endpoints at VictoriaMetrics cluster instead of undocumented VictoriaMetrics_AccountID and VictoriaMetrics_ProjectID labels when writing samples to the needed tenant. See these docs for details.

FEATURE: VictoriaMetrics cluster: support specifying tenant ids via vm_account_id and vm_project_id labels. See these docs and this feature request.
FEATURE: vmagent: improve relabeling performance by up to 3x for non-trivial regex values such as ([^:]+):.+, which can be used for extracting a host part from host:port label value.
FEATURE: MetricsQL: improve performance by up to 4x for queries containing non-trivial regex filters such as {path=~"/foo/.+|/bar"}.
FEATURE: improve performance scalability on systems with many CPU cores for /federate and /api/v1/export/... endpoints.
FEATURE: sanitize metric names for data ingested via DataDog protocol according to DataDog metric naming. The behaviour can be disabled by passing -datadog.sanitizeMetricName=false command-line flag. Thanks to @PerGon for the pull request.
FEATURE: add -usePromCompatibleNaming command-line flag to vmagent, to single-node VictoriaMetrics and to vminsert component of VictoriaMetrics cluster. This flag can be used for normalizing the ingested metric names and label names to Prometheus-compatible form. If this flag is set, then all the chars unsupported by Prometheus are replaced with _ chars in metric names and labels of the ingested samples. See this feature request.
FEATURE: accept whitespace in metric names and tags ingested via Graphite plaintext protocol according to the specs. See this issue.
FEATURE: check the correctess of raw sample timestamps stored on disk when reading them. This reduces the probability of possible silent corruption of the data stored on disk. This should help this and this issue.
FEATURE: atomically delete directories with snapshots, parts and partitions at storage level. Previously such directories can be left in partially deleted state when the deletion operation was interrupted by unclean shutdown. This may result in cannot open file ...: no such file or directory error on the next start. The probability of this error was quite high when NFS or EFS was used as persistent storage for VictoriaMetrics data. See this issue.
FEATURE: set the start arg to end - 5 minutes if isn't passed explicitly to /api/v1/labels and /api/v1/label/.../values. See this pull request.
FEATURE: allow to define the minimum TLS version to use when accepting https requests to VictoriaMetrics components if -tls command-line flag is set. The minimum TLS version can be set via -tlsMinVersion command-line flag. See this feature request.
FEATURE: vmctl: add vm-native-step-interval command line flag for vm-native mode. New option allows splitting the import process into chunks by time interval. This helps migrating data sets with high churn rate and provides better control over the process. See feature request.
FEATURE: vmui: add top queries tab, which shows various stats for recently executed queries. See these docs and this feature request.
FEATURE: vmui: move the "Execute Query" and "Add Query" buttons below the query fields, change icon for remove query. See this issue.
FEATURE: vmui: set the maximum number of queries to 4, remove multi Y-axes, left one for all queries and dotted lines to indicate queries in the graph. See this issue.
FEATURE: vmalert: add debug mode to the alerting rule settings for printing additional information into logs during evaluation. See debug param in alerting rule config.
FEATURE: vmalert: add experimental feature for displaying last 10 states of the rule (recording or alerting) evaluation. The state is available on the Rule page, which can be opened by clicking on Details link next to Rule's name on the /groups page.
FEATURE: vmalert: allow using extra labels in annotiations. See this feature request.
FEATURE: vmalert: allow configuring authorization params per list of targets in vmalert's notifier config for static_configs. See this issue.
FEATURE: vmalert: allow using {{$labels}} for templating in command-line flag -external.alert.source. The change supposed to provide additional flexibility for generating alert's source link based on labels values.
FEATURE: vmalert: add vm_account_id and vm_project_id labels to results of alerting and recording rules if -clusterMode is enabled. This improves multitenant support in vmalert.
FEATURE: vmagent: minimize the time needed for reading large responses from scrape targets in stream parsing mode. This should reduce scrape durations for such targets as kube-state-metrics running in a big Kubernetes cluster.
FEATURE: MetricsQL: add sort_by_label_numeric and sort_by_label_numeric_desc functions for numeric sort of input time series by the specified labels. See this feature request.
FEATURE: vmbackup and vmrestore: retry GCS operations for up to 3 minutes on temporary failures. See this issue.
FEATURE: vmbackup: add support for saving / restoring backups to / from Azure blob storage. See this feature request.
FEATURE: vmbackupmanager: expose vm_backup_in_flight metric, which can be used for determining which backup types - latest, hourly, daily, weekly or monthly - are currently executed.
FEATURE: vmgateway: add ability to extract JWT authorization token from non-standard HTTP header by passing it via -auth.httpHeader command-line flag. See this feature request.
FEATURE: vmagent: expose __meta_ec2_region label for ec2_sd_config in the same way as Prometheus 2.39 does.
FEATURE: vmagent: accept data ingestion requests via paths starting from /prometheus prefix in the same way as VictoriaMetrics does. For example, vmagent now accepts Prometheus remote_write data via both /api/v1/write and /prometheus/api/v1/write. This simplifies switching between single-node VictoriaMetrics and vmagent.
FEATURE: vmagent: add external_labels from global section at -promscrape.config after the relabeling is applied to scraped metrics. This aligns with Prometheus behaviour. Previously the external_labels were added to scrape targets, so they could be modified during relabeling. See this issue.
FEATURE: vmagent: allow specifying per--remoteWrite.url limits for on-disk size for pending data via -remoteWrite.maxDiskUsagePerURL command-line flag. Thanks to @rbizos for the pull request.
FEATURE: VictoriaMetrics cluster: log clear error when multiple identical -storageNode command-line flags are passed to vmselect or to vminsert. Previously these components were crashed with cryptic panic metric ... is already registered in this case. See this issue.
BUGFIX: do not export stale metrics via /federate api after the staleness markers. Previously such metrics were exported with NaN values. this could break some setups. See this issue.
BUGFIX: export ininity numbers as "Infinity" strings at /api/v1/export, so they can be parsed by standard JSON parsers. Previously infinity numbers were exported as Inf values, which couldn't be parsed by standard JSON parsers. See this issue.
BUGFIX: vmauth: properly handle request paths ending with / such as /vmui/. Previously vmui was dropping the traling /, which could prevent from using vmui via vmauth. See this issue.
BUGFIX: vmagent: properly encode query params for aws signed requests, use %20 instead of + as api requires. See this issue.
BUGFIX: vmagent: properly parse relabel config when regex ending with escaped $. See this issue.
BUGFIX: MetricsQL: properly calculate rate_over_sum(m[d]) as sum_over_time(m[d])/d. Previously the sum_over_time(m[d]) could be improperly divided by smaller than d time range. See rate_over_sum() docs and this issue.
BUGFIX: MetricsQL: properly calculate increase(m[d]) over slow-changing counters with values smaller than 100. Previously increase could return unexpectedly big results in this case. See the related issue and this pull request.
BUGFIX: MetricsQL: ignore empty series when applying limit_offset. It should improve queries with additional filters by value in expressions like limit_offset(1,1, foo > 1).
BUGFIX: MetricsQL: properly calculate quantiles_over_time when the lookbehind window contains only a single sample. Previously an empty result was incorrectly returned in this case.
BUGFIX: vmui: fix RangeError: Maximum call stack size exceeded error when the query returns too many data points at Table view. See this pull request.
BUGFIX: vmui: fix workaround for adding more queries via URL. See this issue.
BUGFIX: vmalert: re-evaluate annotations per each alert evaluation. Previously, annotations were evaluated only on alert's value change. This could result in stale annotations in some cases described in this pull request.
BUGFIX: prevent from excessive CPU usage when the storage enters read-only mode. The previous fix in v1.81.0 wasn't complete.
BUGFIX: vmalert: change default value for command-line flag -datasource.queryStep from 0s to 5m. Param step is added by vmalert to every rule evaluation request sent to datasource. Before this change, step was equal to group's evaluation interval by default. Param step for instant queries defines how far VM can look back for the last written data point. The change supposed to improve reliability of the rules evaluation when evaluation interval is lower than scraping interval.
BUGFIX: properly calculate vm_rows_scanned_per_query histogram exported at /metrics page of vmselect and single-node VictoriaMetrics. Previously it could return misleadingly high numbers for rollup functions, which scan only a few samples on the provided lookbehind window in square brackets. For example, increase(m[1d]) always scans only 2 rows (aka raw samples) per each returned time series.