lib/streamaggr: huge pile of changes

- Reduce memory usage by up to 5x when de-duplicating samples across big number of time series.
- Reduce memory usage by up to 5x when aggregating across big number of output time series.
- Add lib/promutils.LabelsCompressor, which is going to be used by other VictoriaMetrics components
  for reducing memory usage for marshaled []prompbmarshal.Label.
- Add `dedup_interval` option at aggregation config, which allows setting individual
  deduplication intervals per each aggregation.
- Add `keep_metric_names` option at aggregation config, which allows keeping the original
  metric names in the output samples.
- Add `unique_samples` output, which counts the number of unique sample values.
- Add `increase_prometheus` and `total_prometheus` outputs, which ignore the first sample
  per each newly encountered time series.
- Use 64-bit hashes instead of marshaled labels as map keys when calculating `count_series` output.
  This makes obsolete https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5579
- Expose various metrics, which may help debugging stream aggregation:
  - vm_streamaggr_dedup_state_size_bytes - the size of data structures responsible for deduplication
  - vm_streamaggr_dedup_state_items_count - the number of items in the deduplication data structures
  - vm_streamaggr_labels_compressor_size_bytes - the size of labels compressor data structures
  - vm_streamaggr_labels_compressor_items_count - the number of entries in the labels compressor
  - vm_streamaggr_flush_duration_seconds - a histogram, which shows the duration of stream aggregation flushes
  - vm_streamaggr_dedup_flush_duration_seconds - a histogram, which shows the duration of deduplication flushes
  - vm_streamaggr_flush_timeouts_total - counter for timed out stream aggregation flushes,
    which took longer than the configured interval
  - vm_streamaggr_dedup_flush_timeouts_total - counter for timed out deduplication flushes,
    which took longer than the configured dedup_interval
- Actualize docs/stream-aggregation.md

The memory usage reduction increases CPU usage during stream aggregation by up to 30%.

This commit is based on https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5850
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5898

docs/CHANGELOG.md

@@ -30,6 +30,12 @@ See also [LTS releases](https://docs.victoriametrics.com/lts-releases/).
 
 ## tip
 
+* FEATURE: [stream aggregation](https://docs.victoriametrics.com/stream-aggregation/): reduce memory usage by up to 5x when aggregating over big number of unique [time series](https://docs.victoriametrics.com/keyconcepts/#time-series). This is expecially visible when [stream deduplication](https://docs.victoriametrics.com/stream-aggregation/#deduplication) is enabled. The downside is increased CPU usage by up to 30%.
+* FEATURE: [stream aggregation](https://docs.victoriametrics.com/stream-aggregation/): add support for `dedup_interval` option, which allows configuring individual [deduplication intervals](https://docs.victoriametrics.com/stream-aggregation/#deduplication) per each [stream aggregation config](https://docs.victoriametrics.com/stream-aggregation/#stream-aggregation-config).
+* FEATURE: [stream aggregation](https://docs.victoriametrics.com/stream-aggregation/): add support for `keep_metric_names` option, which can be set at [stream aggregation config](https://docs.victoriametrics.com/stream-aggregation/#stream-aggregation-config) in order to keep the original metric names in the output aggregated samples instead of using [the default output metric naming scheme](https://docs.victoriametrics.com/stream-aggregation/#output-metric-names).
+* FEATURE: [stream aggregation](https://docs.victoriametrics.com/stream-aggregation/): add [unique_samples](https://docs.victoriametrics.com/stream-aggregation/#unique_samples) output, which can be used for calculating the number of unique sample values over the given `interval`.
+* FEATURE: [stream aggregation](https://docs.victoriametrics.com/stream-aggregation/): add [increase_prometheus](https://docs.victoriametrics.com/stream-aggregation/#increase_prometheus) and [total_prometheus](https://docs.victoriametrics.com/stream-aggregation/#total_prometheus) outputs, which can be used for `increase` and `total` aggregations when the first sample of every new [time series](https://docs.victoriametrics.com/keyconcepts/#time-series) must be ignored.
+* FEATURE: [stream aggregation](https://docs.victoriametrics.com/stream-aggregation/): expose `vm_streamaggr_flush_timeouts_total` and `vm_streamaggr_dedup_flush_timeouts_total` [counters](https://docs.victoriametrics.com/keyconcepts/#counter) at [`/metrics` page](https://docs.victoriametrics.com/#monitoring), which can be used for detecting flush timeouts for stream aggregation states. Expose also `vm_streamaggr_flush_duration_seconds` and `vm_streamaggr_dedup_flush_duration_seconds` [histograms](https://docs.victoriametrics.com/keyconcepts/#histogram) for monitoring the real flush durations of stream aggregation states.
 
 ## [v1.99.0](https://github.com/VictoriaMetrics/VictoriaMetrics/releases/tag/v1.99.0)
 

docs/stream-aggregation.md

@@ -13,7 +13,8 @@ aliases:
 # Streaming aggregation
 
 [vmagent](https://docs.victoriametrics.com/vmagent.html) and [single-node VictoriaMetrics](https://docs.victoriametrics.com/Single-server-VictoriaMetrics.html)
-can aggregate incoming [samples](https://docs.victoriametrics.com/keyConcepts.html#raw-samples) in streaming mode by time and by labels before data is written to remote storage.
+can aggregate incoming [samples](https://docs.victoriametrics.com/keyConcepts.html#raw-samples) in streaming mode by time and by labels before data is written to remote storage
+(or local storage for single-node VictoriaMetrics).
 The aggregation is applied to all the metrics received via any [supported data ingestion protocol](https://docs.victoriametrics.com/#how-to-import-time-series-data)
 and/or scraped from [Prometheus-compatible targets](https://docs.victoriametrics.com/#how-to-scrape-prometheus-exporters-such-as-node-exporter)
 after applying all the configured [relabeling stages](https://docs.victoriametrics.com/vmagent.html#relabeling).
@@ -40,22 +41,26 @@ This behaviour can be changed via the following command-line flags:
 
 - `-remoteWrite.streamAggr.keepInput` at [vmagent](https://docs.victoriametrics.com/vmagent.html) and `-streamAggr.keepInput`
   at [single-node VictoriaMetrics](https://docs.victoriametrics.com/Single-server-VictoriaMetrics.html).
-  If one of these flags are set, then all the input samples are written to the storage alongside the aggregated samples.
+  If one of these flags is set, then all the input samples are written to the storage alongside the aggregated samples.
   The `-remoteWrite.streamAggr.keepInput` flag can be specified individually per each `-remoteWrite.url`.
 - `-remoteWrite.streamAggr.dropInput` at [vmagent](https://docs.victoriametrics.com/vmagent.html) and `-streamAggr.dropInput`
   at [single-node VictoriaMetrics](https://docs.victoriametrics.com/Single-server-VictoriaMetrics.html).
   If one of these flags are set, then all the input samples are dropped, while only the aggregated samples are written to the storage.
   The `-remoteWrite.streamAggr.dropInput` flag can be specified individually per each `-remoteWrite.url`.
 
-By default, all the input samples are aggregated. Sometimes it is needed to de-duplicate samples before the aggregation.
+## Deduplication
-For example, if the samples are received from replicated sources.
-The following command-line flag can be used for enabling the [de-duplication](https://docs.victoriametrics.com/#deduplication)
-before aggregation in this case:
 
-- `-remoteWrite.streamAggr.dedupInterval` at [vmagent](https://docs.victoriametrics.com/vmagent.html).
+By default, all the input samples are aggregated. Sometimes it is needed to de-duplicate samples for the same [time series](https://docs.victoriametrics.com/keyconcepts/#time-series)
+before the aggregation. For example, if the samples are received from replicated sources.
+The [de-duplication](https://docs.victoriametrics.com/#deduplication) can be enabled via the following options:
+
+- `-remoteWrite.streamAggr.dedupInterval` command-line flag at [vmagent](https://docs.victoriametrics.com/vmagent.html).
   This flag can be specified individually per each `-remoteWrite.url`.
   This allows setting different de-duplication intervals per each configured remote storage.
-- `-streamAggr.dedupInterval` at [single-node VictoriaMetrics](https://docs.victoriametrics.com/Single-server-VictoriaMetrics.html).
+- `-streamAggr.dedupInterval` command-line flag at [single-node VictoriaMetrics](https://docs.victoriametrics.com/Single-server-VictoriaMetrics.html).
+- `dedup_interval` option per each [aggregate config](#stream-aggregation-config).
+
+De-duplicatation is performed after performing the input relabeling with `input_relabel_configs` - see [these docs](#relabeling).
 
 ## Use cases
 
@@ -387,6 +392,8 @@ Output metric names for stream aggregation are constructed according to the foll
 
 Both input and output metric names can be modified if needed via relabeling according to [these docs](#relabeling).
 
+It is possible to leave the original metric name after the aggregation by specifying `keep_metric_names: true` option at [stream aggregation config](#stream-aggregation-config).
+The `keep_metric_names` option can be used only if only a single output is set in [`outputs` list](#aggregation-outputs).
 
 ## Relabeling
 
@@ -410,7 +417,7 @@ For example, the following config removes the `:1m_sum_samples` suffix added [to
 ## Aggregation outputs
 
 The aggregations are calculated during the `interval` specified in the [config](#stream-aggregation-config)
-and then sent to the storage once per `interval`.
+and then sent to the storage once per `interval`. The aggregated samples are named according to [output metric naming](#output-metric-names).
 
 If `by` and `without` lists are specified in the [config](#stream-aggregation-config),
 then the [aggregation by labels](#aggregating-by-labels) is performed additionally to aggregation by `interval`.
@@ -419,14 +426,241 @@ On vmagent shutdown or [configuration reload](#configuration-update) unfinished
 as they might produce lower values than user expects. It is possible to specify `flush_on_shutdown: true` setting in 
 aggregation config to make vmagent to send unfinished states to the remote storage.
 
-Below are aggregation functions that can be put in the `outputs` list at [stream aggregation config](#stream-aggregation-config).
+Below are aggregation functions that can be put in the `outputs` list at [stream aggregation config](#stream-aggregation-config):
+
+* [avg](#avg)
+* [count_samples](#count_samples)
+* [count_series](#count_series)
+* [increase](#increase)
+* [increase_prometheus](#increase_prometheus)
+* [histogram_bucket](#histogram_bucket)
+* [last](#last)
+* [max](#max)
+* [min](#min)
+* [stddev](#stddev)
+* [stdvar](#stdvar)
+* [sum_samples](#sum_samples)
+* [total](#total)
+* [total_prometheus](#total_prometheus)
+* [unique_samples](#unique_samples)
+* [quantiles](#quantiles)
+
+### avg
+
+`avg` returns the average over input [sample values](https://docs.victoriametrics.com/keyConcepts.html#raw-samples).
+`avg` makes sense only for aggregating [gauges](https://docs.victoriametrics.com/keyConcepts.html#gauge).
+
+The results of `avg` is equal to the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
+
+```metricsql
+sum(sum_over_time(some_metric[interval])) / sum(count_over_time(some_metric[interval]))
+```
+
+For example, see below time series produced by config with aggregation interval `1m` and `by: ["instance"]` and  the regular query:
+
+<img alt="avg aggregation" src="stream-aggregation-check-avg.webp">
+
+See also [min](#min), [max](#max), [sum_samples](#sum_samples) and [count_samples](#count_samples).
+
+### count_samples
+
+`count_samples` counts the number of input [samples](https://docs.victoriametrics.com/keyConcepts.html#raw-samples) over the given `interval`.
+
+The results of `count_samples` is equal to the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
+
+```metricsql
+sum(count_over_time(some_metric[interval]))
+```
+
+See also [count_series](#count_series) and [sum_samples](#sum_samples).
+
+### count_series
+
+`count_series` counts the number of unique [time series](https://docs.victoriametrics.com/keyConcepts.html#time-series) over the given `interval`.
+
+The results of `count_series` is equal to the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
+
+```metricsql
+count(last_over_time(some_metric[interval]))
+```
+
+See also [count_samples](#count_samples) and [unique_samples](#unique_samples).
+
+### increase
+
+`increase` returns the increase of input [time series](https://docs.victoriametrics.com/keyconcepts/#time-series) over the given 'interval'.
+`increase` makes sense only for aggregating [counters](https://docs.victoriametrics.com/keyConcepts.html#counter).
+
+The results of `increase` is equal to the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
+
+```metricsql
+sum(increase_pure(some_counter[interval]))
+```
+
+`increase` assumes that all the counters start from 0. For example, if the fist seen sample for new [time series](https://docs.victoriametrics.com/keyconcepts/#time-series)
+is `10`, then `increase` assumes that the time series has been increased by `10`. If you need ignoring the first sample for new time series,
+then take a look at [increase_prometheus](#increase_prometheus).
+
+For example, see below time series produced by config with aggregation interval `1m` and `by: ["instance"]` and the regular query:
+
+<img alt="increase aggregation" src="stream-aggregation-check-increase.webp">
+
+`increase` can be used as an alternative for [rate](https://docs.victoriametrics.com/MetricsQL.html#rate) function.
+For example, if `increase` is calculated for `some_counter` with `interval: 5m`, then `rate` can be calculated
+by dividing the resulting aggregation by `5m`:
+
+```metricsql
+some_counter:5m_increase / 5m
+```
+
+This is similar to `rate(some_counter[5m])`.
+
+Please note, opposite to [rate](https://docs.victoriametrics.com/MetricsQL.html#rate), `increase` aggregations can be 
+combined safely afterwards. This is helpful when the aggregation is calculated by more than one vmagent.
+
+Aggregating irregular and sporadic metrics (received from [Lambdas](https://aws.amazon.com/lambda/)
+or [Cloud Functions](https://cloud.google.com/functions)) can be controlled via [staleness_inteval](#stream-aggregation-config).
+
+See also [increase_prometheus](#increase_prometheus) and [total](#total).
+
+### increase_prometheus
+
+`increase_prometheus` returns the increase of input [time series](https://docs.victoriametrics.com/keyconcepts/#time-series) over the given `interval`.
+`increase_prometheus` makes sense only for aggregating [counters](https://docs.victoriametrics.com/keyConcepts.html#counter).
+
+The results of `increase_prometheus` is equal to the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
+
+```metricsql
+sum(increase_prometheus(some_counter[interval]))
+```
+
+`increase_prometheus` skips the first seen sample value per each [time series](https://docs.victoriametrics.com/keyconcepts/#time-series).
+If you need taking into account the first sample per time series, then take a look at [increase](#increase).
+
+See also [increase](#increase), [total](#total) and [total_prometheus](#total_prometheus).
+
+### histogram_bucket
+
+`histogram_bucket` returns [VictoriaMetrics histogram buckets](https://valyala.medium.com/improving-histogram-usability-for-prometheus-and-grafana-bc7e5df0e350)
+  for the input [sample values](https://docs.victoriametrics.com/keyConcepts.html#raw-samples) over the given `interval`.
+`histogram_bucket` makes sense only for aggregating [gauges](https://docs.victoriametrics.com/keyConcepts.html#gauge).
+See how to aggregate regular histograms [here](#aggregating-histograms).
+
+The results of `histogram_bucket` is equal to the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
+
+```metricsql
+sum(histogram_over_time(some_histogram_bucket[interval])) by (vmrange)
+```
+
+See also [quantiles](#quantiles), [min](#min), [max](#max) and [avg](#avg).
+
+### last
+
+`last` returns the last input [sample value](https://docs.victoriametrics.com/keyConcepts.html#raw-samples) over the given `interval`.
+
+The results of `last` is roughly equal to the the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
+
+```metricsql
+last_over_time(some_metric[interval])
+```
+
+See also [min](#min), [max](#max) and [avg](#avg).
+
+### max
+
+`max` returns the maximum input [sample value](https://docs.victoriametrics.com/keyConcepts.html#raw-samples) over the given `interval`.
+
+The results of `max` is equal to the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
+
+```metricsql
+max(max_over_time(some_metric[interval]))
+```
+
+For example, see below time series produced by config with aggregation interval `1m` and the regular query:
+
+<img alt="total aggregation" src="stream-aggregation-check-max.webp">
+
+See also [min](#min) and [avg](#avg).
+
+### min
+
+`min` returns the minimum input [sample value](https://docs.victoriametrics.com/keyConcepts.html#raw-samples) over the given `interval`.
+
+The results of `min` is equal to the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
+
+```metricsql
+min(min_over_time(some_metric[interval]))
+```
+
+For example, see below time series produced by config with aggregation interval `1m` and the regular query:
+
+<img alt="min aggregation" src="stream-aggregation-check-min.webp">
+
+See also [max](#max) and [avg](#avg).
+
+### stddev
+
+`stddev` returns [standard deviation](https://en.wikipedia.org/wiki/Standard_deviation) for the input [sample values](https://docs.victoriametrics.com/keyConcepts.html#raw-samples)
+over the given `interval`.
+`stddev` makes sense only for aggregating [gauges](https://docs.victoriametrics.com/keyConcepts.html#gauge).
+
+The results of `stddev` is roughly equal to the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
+
+```metricsql
+histogram_stddev(sum(histogram_over_time(some_metric[interval])) by (vmrange))
+```
+
+See also [stdvar](#stdvar) and [avg](#avg).
+
+### stdvar
+
+`stdvar` returns [standard variance](https://en.wikipedia.org/wiki/Variance) for the input [sample values](https://docs.victoriametrics.com/keyConcepts.html#raw-samples)
+over the given `interval`.
+`stdvar` makes sense only for aggregating [gauges](https://docs.victoriametrics.com/keyConcepts.html#gauge).
+
+The results of `stdvar` is roughly equal to the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
+
+```metricsql
+histogram_stdvar(sum(histogram_over_time(some_metric[interval])) by (vmrange))
+```
+
+For example, see below time series produced by config with aggregation interval `1m` and the regular query:
+
+<img alt="stdvar aggregation" src="stream-aggregation-check-stdvar.webp">
+
+See also [stddev](#stddev) and [avg](#avg).
+
+### sum_samples
+
+`sum_samples` sums input [sample values](https://docs.victoriametrics.com/keyConcepts.html#raw-samples) over the given `interval`.
+`sum_samples` makes sense only for aggregating [gauges](https://docs.victoriametrics.com/keyConcepts.html#gauge).
+
+The results of `sum_samples` is equal to the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
+
+```metricsql
+sum(sum_over_time(some_metric[interval]))
+```
+
+For example, see below time series produced by config with aggregation interval `1m` and the regular query:
+
+<img alt="sum_samples aggregation" src="stream-aggregation-check-sum-samples.webp">
+
+See also [count_samples](#count_samples) and [count_series](#count_series).
 
 ### total
 
-`total` generates output [counter](https://docs.victoriametrics.com/keyConcepts.html#counter) by summing the input counters.
+`total` generates output [counter](https://docs.victoriametrics.com/keyConcepts.html#counter) by summing the input counters over the given `interval`.
-`total` only makes sense for aggregating [counter](https://docs.victoriametrics.com/keyConcepts.html#counter) metrics.
+`total` makes sense only for aggregating [counters](https://docs.victoriametrics.com/keyConcepts.html#counter).
 
-The results of `total` is equal to the `sum(some_counter)` query.
+The results of `total` is roughly equal to the the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
+
+```metricsql
+sum(running_sum(increase_pure(some_counter)))
+```
+
+`total` assumes that all the counters start from 0. For example, if the fist seen sample for new [time series](https://docs.victoriametrics.com/keyconcepts/#time-series)
+is `10`, then `total` assumes that the time series has been increased by `10`. If you need ignoring the first sample for new time series,
+then take a look at [total_prometheus](#total_prometheus).
 
 For example, see below time series produced by config with aggregation interval `1m` and `by: ["instance"]` and the regular query:
 
@@ -440,138 +674,64 @@ For example:
 
 <img alt="total aggregation counter reset" src="stream-aggregation-check-total-reset.webp">
 
-The same behavior will occur when creating or deleting new series in an aggregation group -
+The same behavior occurs when creating or deleting new series in an aggregation group -
-`total` will increase monotonically considering the values of the series set.  
+`total` output increases monotonically considering the values of the series set.
 An example of changing a set of series can be restarting a pod in the Kubernetes.
-This changes a label with pod's name in the series, but `total` account for such a scenario and do not reset the state of aggregated metric.
+This changes pod name label, but the `total` accounts for such a scenario and doesn't reset the state of aggregated metric.
 
 Aggregating irregular and sporadic metrics (received from [Lambdas](https://aws.amazon.com/lambda/)
 or [Cloud Functions](https://cloud.google.com/functions)) can be controlled via [staleness_inteval](#stream-aggregation-config).
 
-### increase
+See also [total_prometheus](#total_prometheus), [increase](#increase) and [increase_prometheus](#increase_prometheus).
 
-`increase` returns the increase of input [counters](https://docs.victoriametrics.com/keyConcepts.html#counter).
+### total_prometheus
-`increase` only makes sense for aggregating [counter](https://docs.victoriametrics.com/keyConcepts.html#counter) metrics.
 
-The results of `increase` with aggregation interval of `1m` is equal to the `increase(some_counter[1m])` query.
+`total_prometheus` generates output [counter](https://docs.victoriametrics.com/keyConcepts.html#counter) by summing the input counters over the given `interval`.
+`total_prometheus` makes sense only for aggregating [counters](https://docs.victoriametrics.com/keyConcepts.html#counter).
 
-For example, see below time series produced by config with aggregation interval `1m` and `by: ["instance"]` and  the regular query:
+The results of `total_prometheus` is roughly equal to the the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
 
-<img alt="increase aggregation" src="stream-aggregation-check-increase.webp">
+```metricsql
+sum(running_sum(increase_prometheus(some_counter)))
+```
 
-`increase` can be used as an alternative for [rate](https://docs.victoriametrics.com/MetricsQL.html#rate) function.
+`total_prometheus` skips the first seen sample value per each [time series](https://docs.victoriametrics.com/keyconcepts/#time-series).
-For example, if we have `increase` with `interval` of `5m` for a counter `some_counter`, then to get `rate` we should divide
+If you need taking into account the first sample per time series, then take a look at [total](#total).
-the resulting aggregation by the `interval` in seconds: `some_counter:5m_increase / 5m` is similar to `rate(some_counter[5m])`.
-Please note, opposite to [rate](https://docs.victoriametrics.com/MetricsQL.html#rate), `increase` aggregations can be 
-combined safely afterwards. This is helpful when the aggregation is calculated by more than one vmagent.
 
-Aggregating irregular and sporadic metrics (received from [Lambdas](https://aws.amazon.com/lambda/)
+`total_prometheus` is not affected by [counter resets](https://docs.victoriametrics.com/keyConcepts.html#counter) -
-or [Cloud Functions](https://cloud.google.com/functions)) can be controlled via [staleness_inteval](#stream-aggregation-config).
+it continues to increase monotonically with respect to the previous value.
+The counters are most often reset when the application is restarted.
 
-### count_series
+See also [total](#total), [increase](#increase) and [increase_prometheus](#increase_prometheus).
 
-`count_series` counts the number of unique [time series](https://docs.victoriametrics.com/keyConcepts.html#time-series).
+### unique_samples
 
-The results of `count_series` is equal to the `count(some_metric)` query.
+`unique_samples` counts the number of unique sample values over the given `interval`.
+`unique_samples` makes sense only for aggregating [gauges](https://docs.victoriametrics.com/keyConcepts.html#gauge).
 
-### count_samples
+The results of `unique_samples` is equal to the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
 
-`count_samples` counts the number of input [samples](https://docs.victoriametrics.com/keyConcepts.html#raw-samples).
+```metricsql
+count(count_values_over_time(some_metric[interval]))
+```
 
-The results of `count_samples` with aggregation interval of `1m` is equal to the `count_over_time(some_metric[1m])` query.
+See also [sum_samples](#sum_samples) and [count_series](#count_series).
-
-### sum_samples
-
-`sum_samples` sums input [sample values](https://docs.victoriametrics.com/keyConcepts.html#raw-samples).
-`sum_samples` makes sense only for aggregating [gauge](https://docs.victoriametrics.com/keyConcepts.html#gauge) metrics.
-
-The results of `sum_samples` with aggregation interval of `1m` is equal to the `sum_over_time(some_metric[1m])` query.
-
-For example, see below time series produced by config with aggregation interval `1m` and the regular query:
-
-<img alt="sum_samples aggregation" src="stream-aggregation-check-sum-samples.webp">
-
-### last
-
-`last` returns the last input [sample value](https://docs.victoriametrics.com/keyConcepts.html#raw-samples).
-
-The results of `last` with aggregation interval of `1m` is equal to the `last_over_time(some_metric[1m])` query.
-
-This aggregation output doesn't make much sense with `by` lists specified in the [config](#stream-aggregation-config). 
-The result of aggregation by labels in this case will be undetermined, because it depends on the order of processing the time series.
-
-### min
-
-`min` returns the minimum input [sample value](https://docs.victoriametrics.com/keyConcepts.html#raw-samples).
-
-The results of `min` with aggregation interval of `1m` is equal to the `min_over_time(some_metric[1m])` query.
-
-For example, see below time series produced by config with aggregation interval `1m` and the regular query:
-
-<img alt="min aggregation" src="stream-aggregation-check-min.webp">
-
-### max
-
-`max` returns the maximum input [sample value](https://docs.victoriametrics.com/keyConcepts.html#raw-samples).
-
-The results of `max` with aggregation interval of `1m` is equal to the `max_over_time(some_metric[1m])` query.
-
-For example, see below time series produced by config with aggregation interval `1m` and the regular query:
-
-<img alt="total aggregation" src="stream-aggregation-check-max.webp">
-
-### avg
-
-`avg` returns the average input [sample value](https://docs.victoriametrics.com/keyConcepts.html#raw-samples).
-
-The results of `avg` with aggregation interval of `1m` is equal to the `avg_over_time(some_metric[1m])` query.
-
-For example, see below time series produced by config with aggregation interval `1m` and `by: ["instance"]` and  the regular query:
-
-<img alt="avg aggregation" src="stream-aggregation-check-avg.webp">
-
-### stddev
-
-`stddev` returns [standard deviation](https://en.wikipedia.org/wiki/Standard_deviation) for the input [sample values](https://docs.victoriametrics.com/keyConcepts.html#raw-samples).
-`stddev` makes sense only for aggregating [gauge](https://docs.victoriametrics.com/keyConcepts.html#gauge) metrics.
-
-The results of `stddev` with aggregation interval of `1m` is equal to the `stddev_over_time(some_metric[1m])` query.
-
-### stdvar
-
-`stdvar` returns [standard variance](https://en.wikipedia.org/wiki/Variance) for the input [sample values](https://docs.victoriametrics.com/keyConcepts.html#raw-samples).
-`stdvar` makes sense only for aggregating [gauge](https://docs.victoriametrics.com/keyConcepts.html#gauge) metrics.
-
-The results of `stdvar` with aggregation interval of `1m` is equal to the `stdvar_over_time(some_metric[1m])` query.
-
-For example, see below time series produced by config with aggregation interval `1m` and the regular query:
-
-<img alt="stdvar aggregation" src="stream-aggregation-check-stdvar.webp">
-
-### histogram_bucket
-
-`histogram_bucket` returns [VictoriaMetrics histogram buckets](https://valyala.medium.com/improving-histogram-usability-for-prometheus-and-grafana-bc7e5df0e350)
-  for the input [sample values](https://docs.victoriametrics.com/keyConcepts.html#raw-samples).
-`histogram_bucket` makes sense only for aggregating [gauge](https://docs.victoriametrics.com/keyConcepts.html#gauge) metrics.
-See how to aggregate regular histograms [here](#aggregating-histograms).
-
-The results of `histogram_bucket` with aggregation interval of `1m` is equal to the `histogram_over_time(some_histogram_bucket[1m])` query.
-
-Aggregating irregular and sporadic metrics (received from [Lambdas](https://aws.amazon.com/lambda/)
-or [Cloud Functions](https://cloud.google.com/functions)) can be controlled via [staleness_inteval](#stream-aggregation-config).
 
 ### quantiles
 
 `quantiles(phi1, ..., phiN)` returns [percentiles](https://en.wikipedia.org/wiki/Percentile) for the given `phi*`
-over the input [sample values](https://docs.victoriametrics.com/keyConcepts.html#raw-samples). 
+over the input [sample values](https://docs.victoriametrics.com/keyConcepts.html#raw-samples) on the given `interval`.
-The `phi` must be in the range `[0..1]`, where `0` means `0th` percentile, while `1` means `100th` percentile.
+`phi` must be in the range `[0..1]`, where `0` means `0th` percentile, while `1` means `100th` percentile.
-`quantiles(...)` makes sense only for aggregating [gauge](https://docs.victoriametrics.com/keyConcepts.html#gauge) metrics.
+`quantiles(...)` makes sense only for aggregating [gauges](https://docs.victoriametrics.com/keyConcepts.html#gauge).
 
-The results of `quantiles(phi1, ..., phiN)` with aggregation interval of `1m` 
+The results of `quantiles(phi1, ..., phiN)` is equal to the following [MetricsQL](https://docs.victoriametrics.com/metricsql/) query:
-is equal to the `quantiles_over_time("quantile", phi1, ..., phiN, some_histogram_bucket[1m])` query.
+
+```metricsql
+histogram_quantiles("quantile", phi1, ..., phiN, sum(histogram_over_time(some_metric[interval])) by (vmrange))
+```
+
+See also [histogram_bucket](#histogram_bucket), [min](#min), [max](#max) and [avg](#avg).
 
-Please note, `quantiles` aggregation won't produce correct results when vmagent is in [cluster mode](#cluster-mode)
-since percentiles should be calculated only on the whole matched data set.
 
 ## Aggregating by labels
 
@@ -635,13 +795,21 @@ at [single-node VictoriaMetrics](https://docs.victoriametrics.com/Single-server-
   #
   interval: 1m
 
-  # staleness_interval defines an interval after which the series state will be reset if no samples have been sent during it.
+  # dedup_interval is an optional interval for de-duplication of input samples before the aggregation.
-  # It means that:
+  # Samples are de-duplicated on a per-series basis. See https://docs.victoriametrics.com/keyconcepts/#time-series
+  # and https://docs.victoriametrics.com/#deduplication
+  # The deduplication is performed after input_relabel_configs relabeling is applied.
+  # By default the deduplication is disabled.
+  #
+  # dedup_interval: 30s
+
+  # staleness_interval is an optional interval after which the series state will be reset if no samples have been sent during it.
+  # This means that:
   # - no data point will be written for a resulting time series if it didn't receive any updates during configured interval,
   # - if the series receives updates after the configured interval again, then the time series will be calculated from the initial state
   #   (it's like this series didn't exist until now).
   # Increase this parameter if it is expected for matched metrics to be delayed or collected with irregular intervals exceeding the `interval` value.
-  # By default, is equal to x2 of the `interval` field.
+  # By default, it equals to x2 of the `interval` field.
   # The parameter is only relevant for outputs: total, increase and histogram_bucket.
   #
   # staleness_interval: 2m
@@ -649,7 +817,8 @@ at [single-node VictoriaMetrics](https://docs.victoriametrics.com/Single-server-
   # flush_on_shutdown defines whether to flush the unfinished aggregation states on process restarts
   # or config reloads. It is not recommended changing this setting, unless unfinished aggregations states
   # are preferred to missing data points.
-  # Is `false` by default.
+  # Unfinished aggregation states aren't flushed on shutdown by default.
+  #
   # flush_on_shutdown: false
 
   # without is an optional list of labels, which must be removed from the output aggregation.
@@ -667,6 +836,13 @@ at [single-node VictoriaMetrics](https://docs.victoriametrics.com/Single-server-
   #
   outputs: [total]
 
+  # keep_metric_names instructs keeping the original metric names for the aggregated samples.
+  # This option can be set only if outputs list contains only a single output.
+  # By default a special suffix is added to original metric names in the aggregated samples.
+  # See https://docs.victoriametrics.com/stream-aggregation/#output-metric-names
+  #
+  # keep_metric_names: false
+
   # input_relabel_configs is an optional relabeling rules,
   # which are applied to the incoming samples after they pass the match filter
   # and before being aggregated.

lib/promutils/labelscompressor.go

@@ -0,0 +1,125 @@
+package promutils
+
+import (
+	"sync"
+	"sync/atomic"
+	"unsafe"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/encoding"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
+)
+
+// LabelsCompressor compresses []prompbmarshal.Label into short binary strings
+type LabelsCompressor struct {
+	labelToIdx sync.Map
+	idxToLabel sync.Map
+
+	nextIdx atomic.Uint64
+
+	totalSizeBytes atomic.Uint64
+}
+
+// SizeBytes returns the size of lc data in bytes
+func (lc *LabelsCompressor) SizeBytes() uint64 {
+	return uint64(unsafe.Sizeof(*lc)) + lc.totalSizeBytes.Load()
+}
+
+// ItemsCount returns the number of items in lc
+func (lc *LabelsCompressor) ItemsCount() uint64 {
+	return lc.nextIdx.Load()
+}
+
+// Compress compresses labels, appends the compressed labels to dst and returns the result.
+func (lc *LabelsCompressor) Compress(dst []byte, labels []prompbmarshal.Label) []byte {
+	if len(labels) == 0 {
+		// Fast path
+		return append(dst, 0)
+	}
+
+	a := encoding.GetUint64s(len(labels) + 1)
+	a.A[0] = uint64(len(labels))
+	lc.compress(a.A[1:], labels)
+	dst = encoding.MarshalVarUint64s(dst, a.A)
+	encoding.PutUint64s(a)
+	return dst
+}
+
+func (lc *LabelsCompressor) compress(dst []uint64, labels []prompbmarshal.Label) {
+	if len(labels) == 0 {
+		return
+	}
+	_ = dst[len(labels)-1]
+	for i := range labels {
+		label := &labels[i]
+		v, ok := lc.labelToIdx.Load(*label)
+		if !ok {
+			v = lc.nextIdx.Add(1)
+			labelCopy := cloneLabel(label)
+			lc.idxToLabel.Store(v, labelCopy)
+			lc.labelToIdx.Store(*labelCopy, v)
+
+			// Update lc.totalSizeBytes
+			labelSizeBytes := uint64(len(label.Name) + len(label.Value))
+			entrySizeBytes := labelSizeBytes + uint64(unsafe.Sizeof(label)+unsafe.Sizeof(*label)+2*unsafe.Sizeof(v))
+			lc.totalSizeBytes.Add(entrySizeBytes)
+		}
+		dst[i] = v.(uint64)
+	}
+}
+
+func cloneLabel(label *prompbmarshal.Label) *prompbmarshal.Label {
+	// pre-allocate memory for label name and value
+	n := len(label.Name) + len(label.Value)
+	buf := make([]byte, 0, n)
+
+	buf = append(buf, label.Name...)
+	labelName := bytesutil.ToUnsafeString(buf)
+
+	buf = append(buf, label.Value...)
+	labelValue := bytesutil.ToUnsafeString(buf[len(labelName):])
+	return &prompbmarshal.Label{
+		Name:  labelName,
+		Value: labelValue,
+	}
+}
+
+// Decompress decompresses src into []prompbmarshal.Label, appends it to dst and returns the result.
+func (lc *LabelsCompressor) Decompress(dst []prompbmarshal.Label, src []byte) []prompbmarshal.Label {
+	tail, labelsLen, err := encoding.UnmarshalVarUint64(src)
+	if err != nil {
+		logger.Panicf("BUG: cannot unmarshal labels length: %s", err)
+	}
+	if labelsLen == 0 {
+		// fast path - nothing to decode
+		if len(tail) > 0 {
+			logger.Panicf("BUG: unexpected non-empty tail left; len(tail)=%d; tail=%X", len(tail), tail)
+		}
+		return dst
+	}
+
+	a := encoding.GetUint64s(int(labelsLen))
+	tail, err = encoding.UnmarshalVarUint64s(a.A, tail)
+	if err != nil {
+		logger.Panicf("BUG: cannot unmarshal label indexes: %s", err)
+	}
+	if len(tail) > 0 {
+		logger.Panicf("BUG: unexpected non-empty tail left: len(tail)=%d; tail=%X", len(tail), tail)
+	}
+	dst = lc.decompress(dst, a.A)
+	encoding.PutUint64s(a)
+	return dst
+}
+
+func (lc *LabelsCompressor) decompress(dst []prompbmarshal.Label, src []uint64) []prompbmarshal.Label {
+	for _, idx := range src {
+		v, ok := lc.idxToLabel.Load(idx)
+		if !ok {
+			logger.Panicf("BUG: missing label for idx=%d", idx)
+		}
+		label := *(v.(*prompbmarshal.Label))
+		dst = append(dst, label)
+	}
+	return dst
+}

lib/promutils/labelscompressor_test.go

@@ -0,0 +1,119 @@
+package promutils
+
+import (
+	"fmt"
+	"sync"
+	"testing"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
+)
+
+func TestLabelsCompressorSerial(t *testing.T) {
+	var lc LabelsCompressor
+
+	f := func(labels []prompbmarshal.Label) {
+		t.Helper()
+
+		sExpected := labelsToString(labels)
+
+		data := lc.Compress(nil, labels)
+		labelsResult := lc.Decompress(nil, data)
+
+		sResult := labelsToString(labelsResult)
+		if sExpected != sResult {
+			t.Fatalf("unexpected result; got %s; want %s", sResult, sExpected)
+		}
+
+		if len(labels) > 0 {
+			if n := lc.SizeBytes(); n == 0 {
+				t.Fatalf("Unexpected zero SizeBytes()")
+			}
+			if n := lc.ItemsCount(); n == 0 {
+				t.Fatalf("Unexpected zero ItemsCount()")
+			}
+		}
+	}
+
+	// empty labels
+	f(nil)
+	f([]prompbmarshal.Label{})
+
+	// non-empty labels
+	f([]prompbmarshal.Label{
+		{
+			Name:  "instance",
+			Value: "12345.4342.342.3",
+		},
+		{
+			Name:  "job",
+			Value: "kube-pod-12323",
+		},
+	})
+	f([]prompbmarshal.Label{
+		{
+			Name:  "instance",
+			Value: "12345.4342.342.3",
+		},
+		{
+			Name:  "job",
+			Value: "kube-pod-12323",
+		},
+		{
+			Name:  "pod",
+			Value: "foo-bar-baz",
+		},
+	})
+}
+
+func TestLabelsCompressorConcurrent(t *testing.T) {
+	const concurrency = 5
+	var lc LabelsCompressor
+
+	var wg sync.WaitGroup
+	for i := 0; i < concurrency; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			series := newTestSeries(100, 20)
+			for i, labels := range series {
+				sExpected := labelsToString(labels)
+				data := lc.Compress(nil, labels)
+				labelsResult := lc.Decompress(nil, data)
+				sResult := labelsToString(labelsResult)
+				if sExpected != sResult {
+					panic(fmt.Errorf("unexpected result on iteration %d; got %s; want %s", i, sResult, sExpected))
+				}
+			}
+		}()
+	}
+	wg.Wait()
+
+	if n := lc.SizeBytes(); n == 0 {
+		t.Fatalf("Unexpected zero SizeBytes()")
+	}
+	if n := lc.ItemsCount(); n == 0 {
+		t.Fatalf("Unexpected zero ItemsCount()")
+	}
+}
+
+func labelsToString(labels []prompbmarshal.Label) string {
+	l := Labels{
+		Labels: labels,
+	}
+	return l.String()
+}
+
+func newTestSeries(seriesCount, labelsPerSeries int) [][]prompbmarshal.Label {
+	series := make([][]prompbmarshal.Label, seriesCount)
+	for i := 0; i < seriesCount; i++ {
+		labels := make([]prompbmarshal.Label, labelsPerSeries)
+		for j := 0; j < labelsPerSeries; j++ {
+			labels[j] = prompbmarshal.Label{
+				Name:  fmt.Sprintf("label_%d", j),
+				Value: fmt.Sprintf("value_%d_%d", i, j),
+			}
+		}
+		series[i] = labels
+	}
+	return series
+}

lib/promutils/labelscompressor_timing_test.go

@@ -0,0 +1,54 @@
+package promutils
+
+import (
+	"sync/atomic"
+	"testing"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
+)
+
+func BenchmarkLabelsCompressorCompress(b *testing.B) {
+	var lc LabelsCompressor
+	series := newTestSeries(100, 10)
+
+	b.ReportAllocs()
+	b.SetBytes(int64(len(series)))
+
+	b.RunParallel(func(pb *testing.PB) {
+		var dst []byte
+		for pb.Next() {
+			dst = dst[:0]
+			for _, labels := range series {
+				dst = lc.Compress(dst, labels)
+			}
+			Sink.Add(uint64(len(dst)))
+		}
+	})
+}
+
+func BenchmarkLabelsCompressorDecompress(b *testing.B) {
+	var lc LabelsCompressor
+	series := newTestSeries(100, 10)
+	datas := make([][]byte, len(series))
+	var dst []byte
+	for i, labels := range series {
+		dstLen := len(dst)
+		dst = lc.Compress(dst, labels)
+		datas[i] = dst[dstLen:]
+	}
+
+	b.ReportAllocs()
+	b.SetBytes(int64(len(series)))
+
+	b.RunParallel(func(pb *testing.PB) {
+		var labels []prompbmarshal.Label
+		for pb.Next() {
+			for _, data := range datas {
+				labels = lc.Decompress(labels[:0], data)
+			}
+			Sink.Add(uint64(len(labels)))
+		}
+	})
+}
+
+var Sink atomic.Uint64

lib/streamaggr/avg.go

@@ -1,6 +1,7 @@
 package streamaggr
 
 import (
+	"strings"
 	"sync"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
@@ -22,19 +23,24 @@ func newAvgAggrState() *avgAggrState {
 	return &avgAggrState{}
 }
 
-func (as *avgAggrState) pushSample(_, outputKey string, value float64) {
+func (as *avgAggrState) pushSamples(samples []pushSample) {
-again:
+	for i := range samples {
+		s := &samples[i]
+		outputKey := getOutputKey(s.key)
+
+	again:
 		v, ok := as.m.Load(outputKey)
 		if !ok {
 			// The entry is missing in the map. Try creating it.
 			v = &avgStateValue{
-			sum:   value,
+				sum:   s.value,
 				count: 1,
 			}
+			outputKey = strings.Clone(outputKey)
 			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if !loaded {
 				// The entry has been successfully stored
-			return
+				continue
 			}
 			// Update the entry created by a concurrent goroutine.
 			v = vNew
@@ -43,7 +49,7 @@ again:
 		sv.mu.Lock()
 		deleted := sv.deleted
 		if !deleted {
-		sv.sum += value
+			sv.sum += s.value
 			sv.count++
 		}
 		sv.mu.Unlock()
@@ -52,6 +58,7 @@ again:
 			// Try obtaining and updating the entry again.
 			goto again
 		}
+	}
 }
 
 func (as *avgAggrState) appendSeriesForFlush(ctx *flushCtx) {

lib/streamaggr/count_samples.go

@@ -1,12 +1,13 @@
 package streamaggr
 
 import (
+	"strings"
 	"sync"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
 )
 
-// countSamplesAggrState calculates output=countSamples, e.g. the count of input samples.
+// countSamplesAggrState calculates output=count_samples, e.g. the count of input samples.
 type countSamplesAggrState struct {
 	m sync.Map
 }
@@ -21,18 +22,23 @@ func newCountSamplesAggrState() *countSamplesAggrState {
 	return &countSamplesAggrState{}
 }
 
-func (as *countSamplesAggrState) pushSample(_, outputKey string, _ float64) {
+func (as *countSamplesAggrState) pushSamples(samples []pushSample) {
-again:
+	for i := range samples {
+		s := &samples[i]
+		outputKey := getOutputKey(s.key)
+
+	again:
 		v, ok := as.m.Load(outputKey)
 		if !ok {
 			// The entry is missing in the map. Try creating it.
 			v = &countSamplesStateValue{
 				n: 1,
 			}
+			outputKey = strings.Clone(outputKey)
 			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if !loaded {
 				// The new entry has been successfully created.
-			return
+				continue
 			}
 			// Use the entry created by a concurrent goroutine.
 			v = vNew
@@ -49,6 +55,7 @@ again:
 			// Try obtaining and updating the entry again.
 			goto again
 		}
+	}
 }
 
 func (as *countSamplesAggrState) appendSeriesForFlush(ctx *flushCtx) {

lib/streamaggr/count_series.go

@@ -1,9 +1,12 @@
 package streamaggr
 
 import (
+	"strings"
 	"sync"
 
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
+	"github.com/cespare/xxhash/v2"
 )
 
 // countSeriesAggrState calculates output=count_series, e.g. the number of unique series.
@@ -13,8 +16,7 @@ type countSeriesAggrState struct {
 
 type countSeriesStateValue struct {
 	mu      sync.Mutex
-	countedSeries map[string]struct{}
+	m       map[uint64]struct{}
-	n             uint64
 	deleted bool
 }
 
@@ -22,21 +24,29 @@ func newCountSeriesAggrState() *countSeriesAggrState {
 	return &countSeriesAggrState{}
 }
 
-func (as *countSeriesAggrState) pushSample(inputKey, outputKey string, _ float64) {
+func (as *countSeriesAggrState) pushSamples(samples []pushSample) {
-again:
+	for i := range samples {
+		s := &samples[i]
+		inputKey, outputKey := getInputOutputKey(s.key)
+
+		// Count unique hashes over the inputKeys instead of unique inputKey values.
+		// This reduces memory usage at the cost of possible hash collisions for distinct inputKey values.
+		h := xxhash.Sum64(bytesutil.ToUnsafeBytes(inputKey))
+
+	again:
 		v, ok := as.m.Load(outputKey)
 		if !ok {
 			// The entry is missing in the map. Try creating it.
 			v = &countSeriesStateValue{
-			countedSeries: map[string]struct{}{
+				m: map[uint64]struct{}{
-				inputKey: {},
+					h: {},
 				},
-			n: 1,
 			}
+			outputKey = strings.Clone(outputKey)
 			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if !loaded {
 				// The entry has been added to the map.
-			return
+				continue
 			}
 			// Update the entry created by a concurrent goroutine.
 			v = vNew
@@ -45,9 +55,8 @@ again:
 		sv.mu.Lock()
 		deleted := sv.deleted
 		if !deleted {
-		if _, ok := sv.countedSeries[inputKey]; !ok {
+			if _, ok := sv.m[h]; !ok {
-			sv.countedSeries[inputKey] = struct{}{}
+				sv.m[h] = struct{}{}
-			sv.n++
 			}
 		}
 		sv.mu.Unlock()
@@ -56,6 +65,7 @@ again:
 			// Try obtaining and updating the entry again.
 			goto again
 		}
+	}
 }
 
 func (as *countSeriesAggrState) appendSeriesForFlush(ctx *flushCtx) {
@@ -67,7 +77,7 @@ func (as *countSeriesAggrState) appendSeriesForFlush(ctx *flushCtx) {
 
 		sv := v.(*countSeriesStateValue)
 		sv.mu.Lock()
-		n := sv.n
+		n := len(sv.m)
 		// Mark the entry as deleted, so it won't be updated anymore by concurrent pushSample() calls.
 		sv.deleted = true
 		sv.mu.Unlock()

lib/streamaggr/dedup.go

@@ -0,0 +1,185 @@
+package streamaggr
+
+import (
+	"strings"
+	"sync"
+	"unsafe"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
+	"github.com/cespare/xxhash/v2"
+)
+
+const dedupAggrShardsCount = 128
+
+type dedupAggr struct {
+	shards []dedupAggrShard
+}
+
+type dedupAggrShard struct {
+	dedupAggrShardNopad
+
+	// The padding prevents false sharing on widespread platforms with
+	// 128 mod (cache line size) = 0 .
+	_ [128 - unsafe.Sizeof(dedupAggrShardNopad{})%128]byte
+}
+
+type dedupAggrShardNopad struct {
+	mu sync.Mutex
+	m  map[string]*dedupAggrSample
+}
+
+type dedupAggrSample struct {
+	value float64
+}
+
+func newDedupAggr() *dedupAggr {
+	shards := make([]dedupAggrShard, dedupAggrShardsCount)
+	return &dedupAggr{
+		shards: shards,
+	}
+}
+
+func (da *dedupAggr) sizeBytes() uint64 {
+	n := uint64(unsafe.Sizeof(*da))
+	for i := range da.shards {
+		n += da.shards[i].sizeBytes()
+	}
+	return n
+}
+
+func (da *dedupAggr) itemsCount() uint64 {
+	n := uint64(0)
+	for i := range da.shards {
+		n += da.shards[i].itemsCount()
+	}
+	return n
+}
+
+func (das *dedupAggrShard) sizeBytes() uint64 {
+	das.mu.Lock()
+	n := uint64(unsafe.Sizeof(*das))
+	for k, s := range das.m {
+		n += uint64(len(k)) + uint64(unsafe.Sizeof(k)+unsafe.Sizeof(s)+unsafe.Sizeof(*s))
+	}
+	das.mu.Unlock()
+	return n
+}
+
+func (das *dedupAggrShard) itemsCount() uint64 {
+	das.mu.Lock()
+	n := uint64(len(das.m))
+	das.mu.Unlock()
+	return n
+}
+
+func (da *dedupAggr) pushSamples(samples []pushSample) {
+	pss := getPerShardSamples()
+	shards := pss.shards
+	for _, sample := range samples {
+		h := xxhash.Sum64(bytesutil.ToUnsafeBytes(sample.key))
+		idx := h % uint64(len(shards))
+		shards[idx] = append(shards[idx], sample)
+	}
+	for i, shardSamples := range shards {
+		if len(shardSamples) == 0 {
+			continue
+		}
+		da.shards[i].pushSamples(shardSamples)
+	}
+	putPerShardSamples(pss)
+}
+
+type dedupFlushCtx struct {
+	samples []pushSample
+}
+
+func (ctx *dedupFlushCtx) reset() {
+	clear(ctx.samples)
+	ctx.samples = ctx.samples[:0]
+}
+
+func (da *dedupAggr) flush(f func(samples []pushSample)) {
+	ctx := &dedupFlushCtx{}
+	shards := da.shards
+	for i := range shards {
+		ctx.reset()
+		shards[i].flush(ctx, f)
+	}
+}
+
+type perShardSamples struct {
+	shards [][]pushSample
+}
+
+func (pss *perShardSamples) reset() {
+	shards := pss.shards
+	for i, shardSamples := range shards {
+		if len(shardSamples) > 0 {
+			clear(shardSamples)
+			shards[i] = shardSamples[:0]
+		}
+	}
+}
+
+func getPerShardSamples() *perShardSamples {
+	v := perShardSamplesPool.Get()
+	if v == nil {
+		return &perShardSamples{
+			shards: make([][]pushSample, dedupAggrShardsCount),
+		}
+	}
+	return v.(*perShardSamples)
+}
+
+func putPerShardSamples(pss *perShardSamples) {
+	pss.reset()
+	perShardSamplesPool.Put(pss)
+}
+
+var perShardSamplesPool sync.Pool
+
+func (das *dedupAggrShard) pushSamples(samples []pushSample) {
+	das.mu.Lock()
+	defer das.mu.Unlock()
+
+	m := das.m
+	if m == nil {
+		m = make(map[string]*dedupAggrSample, len(samples))
+		das.m = m
+	}
+	for _, sample := range samples {
+		s, ok := m[sample.key]
+		if ok {
+			s.value = sample.value
+		} else {
+			key := strings.Clone(sample.key)
+			m[key] = &dedupAggrSample{
+				value: sample.value,
+			}
+		}
+	}
+}
+
+func (das *dedupAggrShard) flush(ctx *dedupFlushCtx, f func(samples []pushSample)) {
+	das.mu.Lock()
+
+	m := das.m
+	das.m = nil
+
+	das.mu.Unlock()
+
+	if len(m) == 0 {
+		return
+	}
+
+	dstSamples := ctx.samples
+	for key, s := range m {
+		dstSamples = append(dstSamples, pushSample{
+			key:   key,
+			value: s.value,
+		})
+	}
+	ctx.samples = dstSamples
+
+	f(dstSamples)
+}

lib/streamaggr/dedup_test.go

@@ -0,0 +1,88 @@
+package streamaggr
+
+import (
+	"fmt"
+	"reflect"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"testing"
+)
+
+func TestDedupAggrSerial(t *testing.T) {
+	da := newDedupAggr()
+
+	const seriesCount = 100_000
+	expectedSamplesMap := make(map[string]pushSample)
+	for i := 0; i < 2; i++ {
+		samples := make([]pushSample, seriesCount)
+		for j := range samples {
+			sample := &samples[j]
+			sample.key = fmt.Sprintf("key_%d", j)
+			sample.value = float64(i + j)
+			expectedSamplesMap[sample.key] = *sample
+		}
+		da.pushSamples(samples)
+	}
+
+	if n := da.sizeBytes(); n > 4_200_000 {
+		t.Fatalf("too big dedupAggr state before flush: %d bytes; it shouldn't exceed 4_200_000 bytes", n)
+	}
+	if n := da.itemsCount(); n != seriesCount {
+		t.Fatalf("unexpected itemsCount; got %d; want %d", n, seriesCount)
+	}
+
+	flushedSamplesMap := make(map[string]pushSample)
+	flushSamples := func(samples []pushSample) {
+		for _, sample := range samples {
+			sample.key = strings.Clone(sample.key)
+			flushedSamplesMap[sample.key] = sample
+		}
+	}
+	da.flush(flushSamples)
+
+	if !reflect.DeepEqual(expectedSamplesMap, flushedSamplesMap) {
+		t.Fatalf("unexpected samples;\ngot\n%v\nwant\n%v", flushedSamplesMap, expectedSamplesMap)
+	}
+
+	if n := da.sizeBytes(); n > 17_000 {
+		t.Fatalf("too big dedupAggr state after flush; %d bytes; it shouldn't exceed 17_000 bytes", n)
+	}
+	if n := da.itemsCount(); n != 0 {
+		t.Fatalf("unexpected non-zero itemsCount after flush; got %d", n)
+	}
+}
+
+func TestDedupAggrConcurrent(t *testing.T) {
+	const concurrency = 5
+	const seriesCount = 10_000
+	da := newDedupAggr()
+
+	var samplesFlushed atomic.Int64
+	flushSamples := func(samples []pushSample) {
+		samplesFlushed.Add(int64(len(samples)))
+	}
+
+	var wg sync.WaitGroup
+	for i := 0; i < concurrency; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			for i := 0; i < 10; i++ {
+				samples := make([]pushSample, seriesCount)
+				for j := range samples {
+					sample := &samples[j]
+					sample.key = fmt.Sprintf("key_%d", j)
+					sample.value = float64(i + j)
+				}
+				da.pushSamples(samples)
+			}
+			da.flush(flushSamples)
+		}()
+	}
+	wg.Wait()
+
+	if n := samplesFlushed.Load(); n < seriesCount {
+		t.Fatalf("too small number of series flushed; got %d; want at least %d", n, seriesCount)
+	}
+}

lib/streamaggr/dedup_timing_test.go

@@ -0,0 +1,48 @@
+package streamaggr
+
+import (
+	"fmt"
+	"sync/atomic"
+	"testing"
+)
+
+func BenchmarkDedupAggr(b *testing.B) {
+	for _, samplesPerPush := range []int{1, 10, 100, 1_000, 10_000, 100_000} {
+		b.Run(fmt.Sprintf("samplesPerPush_%d", samplesPerPush), func(b *testing.B) {
+			benchmarkDedupAggr(b, samplesPerPush)
+		})
+	}
+}
+
+func benchmarkDedupAggr(b *testing.B, samplesPerPush int) {
+	flushSamples := func(samples []pushSample) {
+		Sink.Add(uint64(len(samples)))
+	}
+
+	const loops = 2
+	benchSamples := newBenchSamples(samplesPerPush)
+	da := newDedupAggr()
+
+	b.ReportAllocs()
+	b.SetBytes(int64(samplesPerPush * loops))
+	b.RunParallel(func(pb *testing.PB) {
+		for pb.Next() {
+			for i := 0; i < loops; i++ {
+				da.pushSamples(benchSamples)
+			}
+			da.flush(flushSamples)
+		}
+	})
+}
+
+func newBenchSamples(count int) []pushSample {
+	samples := make([]pushSample, count)
+	for i := range samples {
+		sample := &samples[i]
+		sample.key = fmt.Sprintf("key_%d", i)
+		sample.value = float64(i)
+	}
+	return samples
+}
+
+var Sink atomic.Uint64

lib/streamaggr/histogram_bucket.go

@@ -2,6 +2,7 @@ package streamaggr
 
 import (
 	"math"
+	"strings"
 	"sync"
 	"time"
 
@@ -9,7 +10,7 @@ import (
 	"github.com/VictoriaMetrics/metrics"
 )
 
-// histogramBucketAggrState calculates output=histogramBucket, e.g. VictoriaMetrics histogram over input samples.
+// histogramBucketAggrState calculates output=histogram_bucket, e.g. VictoriaMetrics histogram over input samples.
 type histogramBucketAggrState struct {
 	m sync.Map
 
@@ -30,15 +31,19 @@ func newHistogramBucketAggrState(stalenessInterval time.Duration) *histogramBuck
 	}
 }
 
-func (as *histogramBucketAggrState) pushSample(_, outputKey string, value float64) {
+func (as *histogramBucketAggrState) pushSamples(samples []pushSample) {
 	currentTime := fasttime.UnixTimestamp()
 	deleteDeadline := currentTime + as.stalenessSecs
+	for i := range samples {
+		s := &samples[i]
+		outputKey := getOutputKey(s.key)
 
-again:
+	again:
 		v, ok := as.m.Load(outputKey)
 		if !ok {
 			// The entry is missing in the map. Try creating it.
 			v = &histogramBucketStateValue{}
+			outputKey = strings.Clone(outputKey)
 			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if loaded {
 				// Use the entry created by a concurrent goroutine.
@@ -49,7 +54,7 @@ again:
 		sv.mu.Lock()
 		deleted := sv.deleted
 		if !deleted {
-		sv.h.Update(value)
+			sv.h.Update(s.value)
 			sv.deleteDeadline = deleteDeadline
 		}
 		sv.mu.Unlock()
@@ -58,6 +63,7 @@ again:
 			// Try obtaining and updating the entry again.
 			goto again
 		}
+	}
 }
 
 func (as *histogramBucketAggrState) removeOldEntries(currentTime uint64) {

lib/streamaggr/increase.go

@@ -1,129 +0,0 @@
-package streamaggr
-
-import (
-	"sync"
-	"time"
-
-	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
-)
-
-// increaseAggrState calculates output=increase, e.g. the increase over input counters.
-type increaseAggrState struct {
-	m sync.Map
-
-	ignoreInputDeadline uint64
-	stalenessSecs       uint64
-}
-
-type increaseStateValue struct {
-	mu             sync.Mutex
-	lastValues     map[string]*lastValueState
-	total          float64
-	deleteDeadline uint64
-	deleted        bool
-}
-
-func newIncreaseAggrState(interval time.Duration, stalenessInterval time.Duration) *increaseAggrState {
-	currentTime := fasttime.UnixTimestamp()
-	intervalSecs := roundDurationToSecs(interval)
-	stalenessSecs := roundDurationToSecs(stalenessInterval)
-	return &increaseAggrState{
-		ignoreInputDeadline: currentTime + intervalSecs,
-		stalenessSecs:       stalenessSecs,
-	}
-}
-
-func (as *increaseAggrState) pushSample(inputKey, outputKey string, value float64) {
-	currentTime := fasttime.UnixTimestamp()
-	deleteDeadline := currentTime + as.stalenessSecs
-
-again:
-	v, ok := as.m.Load(outputKey)
-	if !ok {
-		// The entry is missing in the map. Try creating it.
-		v = &increaseStateValue{
-			lastValues: make(map[string]*lastValueState),
-		}
-		vNew, loaded := as.m.LoadOrStore(outputKey, v)
-		if loaded {
-			// Use the entry created by a concurrent goroutine.
-			v = vNew
-		}
-	}
-	sv := v.(*increaseStateValue)
-	sv.mu.Lock()
-	deleted := sv.deleted
-	if !deleted {
-		lv, ok := sv.lastValues[inputKey]
-		if !ok {
-			lv = &lastValueState{}
-			sv.lastValues[inputKey] = lv
-		}
-		d := value
-		if ok && lv.value <= value {
-			d = value - lv.value
-		}
-		if ok || currentTime > as.ignoreInputDeadline {
-			sv.total += d
-		}
-		lv.value = value
-		lv.deleteDeadline = deleteDeadline
-		sv.deleteDeadline = deleteDeadline
-	}
-	sv.mu.Unlock()
-	if deleted {
-		// The entry has been deleted by the concurrent call to appendSeriesForFlush
-		// Try obtaining and updating the entry again.
-		goto again
-	}
-}
-
-func (as *increaseAggrState) removeOldEntries(currentTime uint64) {
-	m := &as.m
-	m.Range(func(k, v interface{}) bool {
-		sv := v.(*increaseStateValue)
-
-		sv.mu.Lock()
-		deleted := currentTime > sv.deleteDeadline
-		if deleted {
-			// Mark the current entry as deleted
-			sv.deleted = deleted
-		} else {
-			// Delete outdated entries in sv.lastValues
-			m := sv.lastValues
-			for k1, v1 := range m {
-				if currentTime > v1.deleteDeadline {
-					delete(m, k1)
-				}
-			}
-		}
-		sv.mu.Unlock()
-
-		if deleted {
-			m.Delete(k)
-		}
-		return true
-	})
-}
-
-func (as *increaseAggrState) appendSeriesForFlush(ctx *flushCtx) {
-	currentTime := fasttime.UnixTimestamp()
-	currentTimeMsec := int64(currentTime) * 1000
-
-	as.removeOldEntries(currentTime)
-
-	m := &as.m
-	m.Range(func(k, v interface{}) bool {
-		sv := v.(*increaseStateValue)
-		sv.mu.Lock()
-		increase := sv.total
-		sv.total = 0
-		deleted := sv.deleted
-		sv.mu.Unlock()
-		if !deleted {
-			key := k.(string)
-			ctx.appendSeries(key, "increase", currentTimeMsec, increase)
-		}
-		return true
-	})
-}

lib/streamaggr/last.go

@@ -1,6 +1,7 @@
 package streamaggr
 
 import (
+	"strings"
 	"sync"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
@@ -21,18 +22,23 @@ func newLastAggrState() *lastAggrState {
 	return &lastAggrState{}
 }
 
-func (as *lastAggrState) pushSample(_, outputKey string, value float64) {
+func (as *lastAggrState) pushSamples(samples []pushSample) {
-again:
+	for i := range samples {
+		s := &samples[i]
+		outputKey := getOutputKey(s.key)
+
+	again:
 		v, ok := as.m.Load(outputKey)
 		if !ok {
 			// The entry is missing in the map. Try creating it.
 			v = &lastStateValue{
-			last: value,
+				last: s.value,
 			}
+			outputKey = strings.Clone(outputKey)
 			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if !loaded {
 				// The new entry has been successfully created.
-			return
+				continue
 			}
 			// Use the entry created by a concurrent goroutine.
 			v = vNew
@@ -41,7 +47,7 @@ again:
 		sv.mu.Lock()
 		deleted := sv.deleted
 		if !deleted {
-		sv.last = value
+			sv.last = s.value
 		}
 		sv.mu.Unlock()
 		if deleted {
@@ -49,6 +55,7 @@ again:
 			// Try obtaining and updating the entry again.
 			goto again
 		}
+	}
 }
 
 func (as *lastAggrState) appendSeriesForFlush(ctx *flushCtx) {

lib/streamaggr/max.go

@@ -1,6 +1,7 @@
 package streamaggr
 
 import (
+	"strings"
 	"sync"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
@@ -21,18 +22,23 @@ func newMaxAggrState() *maxAggrState {
 	return &maxAggrState{}
 }
 
-func (as *maxAggrState) pushSample(_, outputKey string, value float64) {
+func (as *maxAggrState) pushSamples(samples []pushSample) {
-again:
+	for i := range samples {
+		s := &samples[i]
+		outputKey := getOutputKey(s.key)
+
+	again:
 		v, ok := as.m.Load(outputKey)
 		if !ok {
 			// The entry is missing in the map. Try creating it.
 			v = &maxStateValue{
-			max: value,
+				max: s.value,
 			}
+			outputKey = strings.Clone(outputKey)
 			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if !loaded {
 				// The new entry has been successfully created.
-			return
+				continue
 			}
 			// Use the entry created by a concurrent goroutine.
 			v = vNew
@@ -41,8 +47,8 @@ again:
 		sv.mu.Lock()
 		deleted := sv.deleted
 		if !deleted {
-		if value > sv.max {
+			if s.value > sv.max {
-			sv.max = value
+				sv.max = s.value
 			}
 		}
 		sv.mu.Unlock()
@@ -51,6 +57,7 @@ again:
 			// Try obtaining and updating the entry again.
 			goto again
 		}
+	}
 }
 
 func (as *maxAggrState) appendSeriesForFlush(ctx *flushCtx) {

lib/streamaggr/min.go

@@ -1,6 +1,7 @@
 package streamaggr
 
 import (
+	"strings"
 	"sync"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
@@ -21,18 +22,23 @@ func newMinAggrState() *minAggrState {
 	return &minAggrState{}
 }
 
-func (as *minAggrState) pushSample(_, outputKey string, value float64) {
+func (as *minAggrState) pushSamples(samples []pushSample) {
-again:
+	for i := range samples {
+		s := &samples[i]
+		outputKey := getOutputKey(s.key)
+
+	again:
 		v, ok := as.m.Load(outputKey)
 		if !ok {
 			// The entry is missing in the map. Try creating it.
 			v = &minStateValue{
-			min: value,
+				min: s.value,
 			}
+			outputKey = strings.Clone(outputKey)
 			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if !loaded {
 				// The new entry has been successfully created.
-			return
+				continue
 			}
 			// Use the entry created by a concurrent goroutine.
 			v = vNew
@@ -41,8 +47,8 @@ again:
 		sv.mu.Lock()
 		deleted := sv.deleted
 		if !deleted {
-		if value < sv.min {
+			if s.value < sv.min {
-			sv.min = value
+				sv.min = s.value
 			}
 		}
 		sv.mu.Unlock()
@@ -51,6 +57,7 @@ again:
 			// Try obtaining and updating the entry again.
 			goto again
 		}
+	}
 }
 
 func (as *minAggrState) appendSeriesForFlush(ctx *flushCtx) {

lib/streamaggr/quantiles.go

@@ -2,6 +2,7 @@ package streamaggr
 
 import (
 	"strconv"
+	"strings"
 	"sync"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
@@ -28,8 +29,12 @@ func newQuantilesAggrState(phis []float64) *quantilesAggrState {
 	}
 }
 
-func (as *quantilesAggrState) pushSample(_, outputKey string, value float64) {
+func (as *quantilesAggrState) pushSamples(samples []pushSample) {
-again:
+	for i := range samples {
+		s := &samples[i]
+		outputKey := getOutputKey(s.key)
+
+	again:
 		v, ok := as.m.Load(outputKey)
 		if !ok {
 			// The entry is missing in the map. Try creating it.
@@ -37,6 +42,7 @@ again:
 			v = &quantilesStateValue{
 				h: h,
 			}
+			outputKey = strings.Clone(outputKey)
 			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if loaded {
 				// Use the entry created by a concurrent goroutine.
@@ -48,7 +54,7 @@ again:
 		sv.mu.Lock()
 		deleted := sv.deleted
 		if !deleted {
-		sv.h.Update(value)
+			sv.h.Update(s.value)
 		}
 		sv.mu.Unlock()
 		if deleted {
@@ -56,6 +62,7 @@ again:
 			// Try obtaining and updating the entry again.
 			goto again
 		}
+	}
 }
 
 func (as *quantilesAggrState) appendSeriesForFlush(ctx *flushCtx) {

lib/streamaggr/stddev.go

@@ -2,6 +2,7 @@ package streamaggr
 
 import (
 	"math"
+	"strings"
 	"sync"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
@@ -24,12 +25,17 @@ func newStddevAggrState() *stddevAggrState {
 	return &stddevAggrState{}
 }
 
-func (as *stddevAggrState) pushSample(_, outputKey string, value float64) {
+func (as *stddevAggrState) pushSamples(samples []pushSample) {
-again:
+	for i := range samples {
+		s := &samples[i]
+		outputKey := getOutputKey(s.key)
+
+	again:
 		v, ok := as.m.Load(outputKey)
 		if !ok {
 			// The entry is missing in the map. Try creating it.
 			v = &stddevStateValue{}
+			outputKey = strings.Clone(outputKey)
 			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if loaded {
 				// Use the entry created by a concurrent goroutine.
@@ -42,8 +48,8 @@ again:
 		if !deleted {
 			// See `Rapid calculation methods` at https://en.wikipedia.org/wiki/Standard_deviation
 			sv.count++
-		avg := sv.avg + (value-sv.avg)/sv.count
+			avg := sv.avg + (s.value-sv.avg)/sv.count
-		sv.q += (value - sv.avg) * (value - avg)
+			sv.q += (s.value - sv.avg) * (s.value - avg)
 			sv.avg = avg
 		}
 		sv.mu.Unlock()
@@ -52,6 +58,7 @@ again:
 			// Try obtaining and updating the entry again.
 			goto again
 		}
+	}
 }
 
 func (as *stddevAggrState) appendSeriesForFlush(ctx *flushCtx) {

lib/streamaggr/stdvar.go

@@ -1,6 +1,7 @@
 package streamaggr
 
 import (
+	"strings"
 	"sync"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
@@ -23,12 +24,17 @@ func newStdvarAggrState() *stdvarAggrState {
 	return &stdvarAggrState{}
 }
 
-func (as *stdvarAggrState) pushSample(_, outputKey string, value float64) {
+func (as *stdvarAggrState) pushSamples(samples []pushSample) {
-again:
+	for i := range samples {
+		s := &samples[i]
+		outputKey := getOutputKey(s.key)
+
+	again:
 		v, ok := as.m.Load(outputKey)
 		if !ok {
 			// The entry is missing in the map. Try creating it.
 			v = &stdvarStateValue{}
+			outputKey = strings.Clone(outputKey)
 			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if loaded {
 				// Use the entry created by a concurrent goroutine.
@@ -41,8 +47,8 @@ again:
 		if !deleted {
 			// See `Rapid calculation methods` at https://en.wikipedia.org/wiki/Standard_deviation
 			sv.count++
-		avg := sv.avg + (value-sv.avg)/sv.count
+			avg := sv.avg + (s.value-sv.avg)/sv.count
-		sv.q += (value - sv.avg) * (value - avg)
+			sv.q += (s.value - sv.avg) * (s.value - avg)
 			sv.avg = avg
 		}
 		sv.mu.Unlock()
@@ -51,6 +57,7 @@ again:
 			// Try obtaining and updating the entry again.
 			goto again
 		}
+	}
 }
 
 func (as *stdvarAggrState) appendSeriesForFlush(ctx *flushCtx) {

lib/streamaggr/streamaggr.go

@@ -4,6 +4,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"math"
+	"slices"
 	"sort"
 	"strconv"
 	"strings"
@@ -19,14 +20,18 @@ import (
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/promrelabel"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/promutils"
+	"github.com/VictoriaMetrics/metrics"
 	"gopkg.in/yaml.v2"
 )
 
 var supportedOutputs = []string{
 	"total",
+	"total_prometheus",
 	"increase",
+	"increase_prometheus",
 	"count_series",
 	"count_samples",
+	"unique_samples",
 	"sum_samples",
 	"last",
 	"min",
@@ -80,8 +85,11 @@ type Config struct {
 	// Interval is the interval between aggregations.
 	Interval string `yaml:"interval"`
 
+	// DedupInterval is an optional interval for deduplication.
+	DedupInterval string `yaml:"dedup_interval,omitempty"`
+
 	// Staleness interval is interval after which the series state will be reset if no samples have been sent during it.
-	// The parameter is only relevant for outputs: total, increase and histogram_bucket.
+	// The parameter is only relevant for outputs: total, total_prometheus, increase, increase_prometheus and histogram_bucket.
 	StalenessInterval string `yaml:"staleness_interval,omitempty"`
 
 	// Outputs is a list of output aggregate functions to produce.
@@ -89,10 +97,13 @@ type Config struct {
 	// The following names are allowed:
 	//
 	// - total - aggregates input counters
-	// - increase - counts the increase over input counters
+	// - total_prometheus - aggregates input counters, ignoring the first sample in new time series
-	// - count_series - counts the input series
+	// - increase - calculates the increase over input series
+	// - increase_prometheus - calculates the increase over input series, ignoring the first sample in new time series
+	// - count_series - counts the number of unique input series
 	// - count_samples - counts the input samples
-	// - sum_samples - sums the input samples
+	// - unique_samples - counts the number of unique sample values
+	// - sum_samples - sums the input sample values
 	// - last - the last biggest sample value
 	// - min - the minimum sample value
 	// - max - the maximum sample value
@@ -102,12 +113,18 @@ type Config struct {
 	// - histogram_bucket - creates VictoriaMetrics histogram for input samples
 	// - quantiles(phi1, ..., phiN) - quantiles' estimation for phi in the range [0..1]
 	//
-	// The output time series will have the following names:
+	// The output time series will have the following names by default:
 	//
-	//   input_name:aggr_<interval>_<output>
+	//   input_name:<interval>[_by_<by_labels>][_without_<without_labels>]_<output>
+	//
+	// See also KeepMetricNames
 	//
 	Outputs []string `yaml:"outputs"`
 
+	// KeepMetricNames instructs to leave metric names as is for the output time series
+	// without adding any suffix.
+	KeepMetricNames bool `yaml:"keep_metric_names,omitempty"`
+
 	// By is an optional list of labels for grouping input series.
 	//
 	// See also Without.
@@ -144,6 +161,8 @@ type Aggregators struct {
 	// configData contains marshaled configs passed to NewAggregators().
 	// It is used in Equal() for comparing Aggregators.
 	configData []byte
+
+	ms *metrics.Set
 }
 
 // NewAggregators creates Aggregators from the given cfgs.
@@ -155,9 +174,10 @@ type Aggregators struct {
 //
 // MustStop must be called on the returned Aggregators when they are no longer needed.
 func NewAggregators(cfgs []*Config, pushFunc PushFunc, dedupInterval time.Duration) (*Aggregators, error) {
+	ms := metrics.NewSet()
 	as := make([]*aggregator, len(cfgs))
 	for i, cfg := range cfgs {
-		a, err := newAggregator(cfg, pushFunc, dedupInterval)
+		a, err := newAggregator(cfg, pushFunc, ms, dedupInterval)
 		if err != nil {
 			// Stop already initialized aggregators before returning the error.
 			for _, a := range as[:i] {
@@ -171,9 +191,46 @@ func NewAggregators(cfgs []*Config, pushFunc PushFunc, dedupInterval time.Durati
 	if err != nil {
 		logger.Panicf("BUG: cannot marshal the provided configs: %s", err)
 	}
+
+	_ = ms.NewGauge(`vm_streamaggr_dedup_state_size_bytes`, func() float64 {
+		n := uint64(0)
+		for _, aggr := range as {
+			if aggr.da != nil {
+				n += aggr.da.sizeBytes()
+			}
+		}
+		return float64(n)
+	})
+	_ = ms.NewGauge(`vm_streamaggr_dedup_state_items_count`, func() float64 {
+		n := uint64(0)
+		for _, aggr := range as {
+			if aggr.da != nil {
+				n += aggr.da.itemsCount()
+			}
+		}
+		return float64(n)
+	})
+
+	_ = ms.NewGauge(`vm_streamaggr_labels_compressor_size_bytes`, func() float64 {
+		n := uint64(0)
+		for _, aggr := range as {
+			n += aggr.lc.SizeBytes()
+		}
+		return float64(n)
+	})
+	_ = ms.NewGauge(`vm_streamaggr_labels_compressor_items_count`, func() float64 {
+		n := uint64(0)
+		for _, aggr := range as {
+			n += aggr.lc.ItemsCount()
+		}
+		return float64(n)
+	})
+
+	metrics.RegisterSet(ms)
 	return &Aggregators{
 		as:         as,
 		configData: configData,
+		ms:         ms,
 	}, nil
 }
 
@@ -182,9 +239,14 @@ func (a *Aggregators) MustStop() {
 	if a == nil {
 		return
 	}
+
+	metrics.UnregisterSet(a.ms)
+	a.ms = nil
+
 	for _, aggr := range a.as {
 		aggr.MustStop()
 	}
+	a.as = nil
 }
 
 // Equal returns true if a and b are initialized from identical configs.
@@ -208,12 +270,14 @@ func (a *Aggregators) Push(tss []prompbmarshal.TimeSeries, matchIdxs []byte) []b
 	for i := 0; i < len(matchIdxs); i++ {
 		matchIdxs[i] = 0
 	}
+	if a == nil {
+		return matchIdxs
+	}
 
-	if a != nil {
 	for _, aggr := range a.as {
 		aggr.Push(tss, matchIdxs)
 	}
-	}
+
 	return matchIdxs
 }
 
@@ -224,17 +288,22 @@ type aggregator struct {
 	inputRelabeling  *promrelabel.ParsedConfigs
 	outputRelabeling *promrelabel.ParsedConfigs
 
+	keepMetricNames bool
+
 	by                  []string
 	without             []string
 	aggregateOnlyByTime bool
 
-	// dedupAggr is set to non-nil if input samples must be de-duplicated according
+	// da is set to non-nil if input samples must be de-duplicated according
 	// to the dedupInterval passed to newAggregator().
-	dedupAggr *lastAggrState
+	da *dedupAggr
 
 	// aggrStates contains aggregate states for the given outputs
 	aggrStates []aggrState
 
+	// lc is used for compressing series keys before passing them to dedupAggr and aggrState.
+	lc promutils.LabelsCompressor
+
 	pushFunc PushFunc
 
 	// suffix contains a suffix, which should be added to aggregate metric names
@@ -250,10 +319,16 @@ type aggregator struct {
 
 	wg     sync.WaitGroup
 	stopCh chan struct{}
+
+	flushDuration      *metrics.Histogram
+	dedupFlushDuration *metrics.Histogram
+
+	flushTimeouts      *metrics.Counter
+	dedupFlushTimeouts *metrics.Counter
 }
 
 type aggrState interface {
-	pushSample(inputKey, outputKey string, value float64)
+	pushSamples(samples []pushSample)
 	appendSeriesForFlush(ctx *flushCtx)
 }
 
@@ -266,7 +341,7 @@ type PushFunc func(tss []prompbmarshal.TimeSeries)
 // e.g. only the last sample per each time series per each dedupInterval is aggregated.
 //
 // The returned aggregator must be stopped when no longer needed by calling MustStop().
-func newAggregator(cfg *Config, pushFunc PushFunc, dedupInterval time.Duration) (*aggregator, error) {
+func newAggregator(cfg *Config, pushFunc PushFunc, ms *metrics.Set, dedupInterval time.Duration) (*aggregator, error) {
 	// check cfg.Interval
 	interval, err := time.ParseDuration(cfg.Interval)
 	if err != nil {
@@ -276,6 +351,21 @@ func newAggregator(cfg *Config, pushFunc PushFunc, dedupInterval time.Duration)
 		return nil, fmt.Errorf("aggregation interval cannot be smaller than 1s; got %s", interval)
 	}
 
+	// check cfg.DedupInterval
+	if cfg.DedupInterval != "" {
+		di, err := time.ParseDuration(cfg.DedupInterval)
+		if err != nil {
+			return nil, fmt.Errorf("cannot parse `dedup_interval: %q`: %w", cfg.DedupInterval, err)
+		}
+		dedupInterval = di
+	}
+	if dedupInterval > interval {
+		return nil, fmt.Errorf("dedup_interval=%s cannot exceed interval=%s", dedupInterval, interval)
+	}
+	if dedupInterval > 0 && interval%dedupInterval != 0 {
+		return nil, fmt.Errorf("interval=%s must be a multiple of dedup_interval=%s", interval, dedupInterval)
+	}
+
 	// check cfg.StalenessInterval
 	stalenessInterval := interval * 2
 	if cfg.StalenessInterval != "" {
@@ -284,7 +374,7 @@ func newAggregator(cfg *Config, pushFunc PushFunc, dedupInterval time.Duration)
 			return nil, fmt.Errorf("cannot parse `staleness_interval: %q`: %w", cfg.StalenessInterval, err)
 		}
 		if stalenessInterval < interval {
-			return nil, fmt.Errorf("staleness_interval cannot be less than interval (%s); got %s", cfg.Interval, cfg.StalenessInterval)
+			return nil, fmt.Errorf("interval=%s cannot exceed staleness_interval=%s", cfg.Interval, cfg.StalenessInterval)
 		}
 	}
 
@@ -309,6 +399,16 @@ func newAggregator(cfg *Config, pushFunc PushFunc, dedupInterval time.Duration)
 		by = addMissingUnderscoreName(by)
 	}
 
+	// check cfg.KeepMetricNames
+	if cfg.KeepMetricNames {
+		if len(cfg.Outputs) != 1 {
+			return nil, fmt.Errorf("`ouputs` list must contain only a single entry if `keep_metric_names` is set; got %q", cfg.Outputs)
+		}
+		if cfg.Outputs[0] == "histogram_bucket" || strings.HasPrefix(cfg.Outputs[0], "quantiles(") && strings.Contains(cfg.Outputs[0], ",") {
+			return nil, fmt.Errorf("`keep_metric_names` cannot be applied to `outputs: %q`, since they can generate multiple time series", cfg.Outputs)
+		}
+	}
+
 	// initialize outputs list
 	if len(cfg.Outputs) == 0 {
 		return nil, fmt.Errorf("`outputs` list must contain at least a single entry from the list %s; "+
@@ -342,13 +442,19 @@ func newAggregator(cfg *Config, pushFunc PushFunc, dedupInterval time.Duration)
 		}
 		switch output {
 		case "total":
-			aggrStates[i] = newTotalAggrState(interval, stalenessInterval)
+			aggrStates[i] = newTotalAggrState(stalenessInterval, false, true)
+		case "total_prometheus":
+			aggrStates[i] = newTotalAggrState(stalenessInterval, false, false)
 		case "increase":
-			aggrStates[i] = newIncreaseAggrState(interval, stalenessInterval)
+			aggrStates[i] = newTotalAggrState(stalenessInterval, true, true)
+		case "increase_prometheus":
+			aggrStates[i] = newTotalAggrState(stalenessInterval, true, false)
 		case "count_series":
 			aggrStates[i] = newCountSeriesAggrState()
 		case "count_samples":
 			aggrStates[i] = newCountSamplesAggrState()
+		case "unique_samples":
+			aggrStates[i] = newUniqueSamplesAggrState()
 		case "sum_samples":
 			aggrStates[i] = newSumSamplesAggrState()
 		case "last":
@@ -381,11 +487,6 @@ func newAggregator(cfg *Config, pushFunc PushFunc, dedupInterval time.Duration)
 	}
 	suffix += "_"
 
-	var dedupAggr *lastAggrState
-	if dedupInterval > 0 {
-		dedupAggr = newLastAggrState()
-	}
-
 	// initialize the aggregator
 	a := &aggregator{
 		match: cfg.Match,
@@ -393,11 +494,12 @@ func newAggregator(cfg *Config, pushFunc PushFunc, dedupInterval time.Duration)
 		inputRelabeling:  inputRelabeling,
 		outputRelabeling: outputRelabeling,
 
+		keepMetricNames: cfg.KeepMetricNames,
+
 		by:                  by,
 		without:             without,
 		aggregateOnlyByTime: aggregateOnlyByTime,
 
-		dedupAggr:  dedupAggr,
 		aggrStates: aggrStates,
 		pushFunc:   pushFunc,
 
@@ -405,75 +507,84 @@ func newAggregator(cfg *Config, pushFunc PushFunc, dedupInterval time.Duration)
 		flushOnShutdown: cfg.FlushOnShutdown,
 
 		stopCh: make(chan struct{}),
+
+		flushDuration:      ms.GetOrCreateHistogram(`vm_streamaggr_flush_duration_seconds`),
+		dedupFlushDuration: ms.GetOrCreateHistogram(`vm_streamaggr_dedup_flush_duration_seconds`),
+
+		flushTimeouts:      ms.GetOrCreateCounter(`vm_streamaggr_flush_timeouts_total`),
+		dedupFlushTimeouts: ms.GetOrCreateCounter(`vm_streamaggr_dedup_flush_timeouts_total`),
+	}
+	if dedupInterval > 0 {
+		a.da = newDedupAggr()
 	}
 
-	if dedupAggr != nil {
 	a.wg.Add(1)
 	go func() {
-			a.runDedupFlusher(dedupInterval)
+		a.runFlusher(interval, dedupInterval)
-			a.wg.Done()
-		}()
-	}
-	a.wg.Add(1)
-	go func() {
-		a.runFlusher(interval)
 		a.wg.Done()
 	}()
 
 	return a, nil
 }
 
-func (a *aggregator) runDedupFlusher(interval time.Duration) {
+func (a *aggregator) runFlusher(interval, dedupInterval time.Duration) {
-	t := time.NewTicker(interval)
+	tickerFlush := time.NewTicker(interval)
+	defer tickerFlush.Stop()
+
+	var dedupTickerCh <-chan time.Time
+	if dedupInterval > 0 {
+		t := time.NewTicker(dedupInterval)
 		defer t.Stop()
+		dedupTickerCh = t.C
+	}
+
 	for {
 		select {
 		case <-a.stopCh:
 			return
-		case <-t.C:
+		case <-tickerFlush.C:
-		}
+			startTime := time.Now()
 
-		// Globally limit the concurrency for metrics' flush
-		// in order to limit memory usage when big number of aggregators
-		// are flushed at the same time.
-		flushConcurrencyCh <- struct{}{}
-		a.dedupFlush()
-		<-flushConcurrencyCh
-	}
-}
-
-func (a *aggregator) runFlusher(interval time.Duration) {
-	t := time.NewTicker(interval)
-	defer t.Stop()
-	for {
-		select {
-		case <-a.stopCh:
-			return
-		case <-t.C:
-		}
-
-		// Globally limit the concurrency for metrics' flush
-		// in order to limit memory usage when big number of aggregators
-		// are flushed at the same time.
 			flushConcurrencyCh <- struct{}{}
 			a.flush()
 			<-flushConcurrencyCh
+
+			d := time.Since(startTime)
+			a.flushDuration.Update(d.Seconds())
+			if d > interval {
+				a.flushTimeouts.Inc()
+				logger.Warnf("stream aggregation couldn't be finished in the configured interval=%s; it took %s; "+
+					"possible solutions: increase interval; use match filter matching smaller number of series; "+
+					"reduce samples' ingestion rate to stream aggregation", interval, d)
+			}
+		case <-dedupTickerCh:
+			startTime := time.Now()
+
+			flushConcurrencyCh <- struct{}{}
+			a.dedupFlush()
+			<-flushConcurrencyCh
+
+			d := time.Since(startTime)
+			a.dedupFlushDuration.Update(d.Seconds())
+			if d > dedupInterval {
+				a.dedupFlushTimeouts.Inc()
+				logger.Warnf("stream aggregation deduplication couldn't be finished in the configured dedup_interval=%s; it took %s; "+
+					"possible solutions: increase dedup_interval; use match filter matching smaller number of series; "+
+					"reduce samples' ingestion rate to stream aggregation", dedupInterval, d)
+			}
+		}
 	}
 }
 
 var flushConcurrencyCh = make(chan struct{}, cgroup.AvailableCPUs())
 
 func (a *aggregator) dedupFlush() {
-	ctx := &flushCtx{
+	a.da.flush(a.pushSamples)
-		skipAggrSuffix: true,
-	}
-	a.dedupAggr.appendSeriesForFlush(ctx)
-	a.push(ctx.tss, nil)
 }
 
 func (a *aggregator) flush() {
 	ctx := &flushCtx{
-		suffix: a.suffix,
+		a: a,
 	}
 	for _, as := range a.aggrStates {
 		ctx.reset()
@@ -521,7 +632,7 @@ func (a *aggregator) MustStop() {
 
 	// Flush the remaining data from the last interval if needed.
 	flushConcurrencyCh <- struct{}{}
-	if a.dedupAggr != nil {
+	if a.da != nil {
 		a.dedupFlush()
 	}
 	a.flush()
@@ -530,18 +641,15 @@ func (a *aggregator) MustStop() {
 
 // Push pushes tss to a.
 func (a *aggregator) Push(tss []prompbmarshal.TimeSeries, matchIdxs []byte) {
-	if a.dedupAggr == nil {
+	ctx := getPushCtx()
-		// Deduplication is disabled.
+	defer putPushCtx(ctx)
-		a.push(tss, matchIdxs)
+
-		return
+	samples := ctx.samples
-	}
+	labels := &ctx.labels
+	inputLabels := &ctx.inputLabels
+	outputLabels := &ctx.outputLabels
+	buf := ctx.buf
 
-	// Deduplication is enabled.
-	// push samples to dedupAggr, so later they will be pushed to the configured aggregators.
-	pushSample := a.dedupAggr.pushSample
-	inputKey := ""
-	bb := bbPool.Get()
-	labels := promutils.GetLabels()
 	for idx, ts := range tss {
 		if !a.match.Match(ts.Labels) {
 			continue
@@ -556,173 +664,140 @@ func (a *aggregator) Push(tss []prompbmarshal.TimeSeries, matchIdxs []byte) {
 		}
 		labels.Sort()
 
-		bb.B = marshalLabelsFast(bb.B[:0], labels.Labels)
+		inputLabels.Reset()
-		outputKey := bytesutil.InternBytes(bb.B)
+		outputLabels.Reset()
-		for _, sample := range ts.Samples {
-			pushSample(inputKey, outputKey, sample.Value)
-		}
-	}
-	promutils.PutLabels(labels)
-	bbPool.Put(bb)
-}
-
-func (a *aggregator) push(tss []prompbmarshal.TimeSeries, matchIdxs []byte) {
-	labels := promutils.GetLabels()
-	tmpLabels := promutils.GetLabels()
-	bb := bbPool.Get()
-	applyFilters := matchIdxs != nil
-	for idx, ts := range tss {
-		if applyFilters {
-			if !a.match.Match(ts.Labels) {
-				continue
-			}
-			matchIdxs[idx] = 1
-		}
-
-		labels.Labels = append(labels.Labels[:0], ts.Labels...)
-		if applyFilters {
-			labels.Labels = a.inputRelabeling.Apply(labels.Labels, 0)
-			if len(labels.Labels) == 0 {
-				// The metric has been deleted by the relabeling
-				continue
-			}
-			labels.Sort()
-		}
-
-		if a.aggregateOnlyByTime {
-			bb.B = marshalLabelsFast(bb.B[:0], labels.Labels)
-		} else {
-			tmpLabels.Labels = removeUnneededLabels(tmpLabels.Labels[:0], labels.Labels, a.by, a.without)
-			bb.B = marshalLabelsFast(bb.B[:0], tmpLabels.Labels)
-		}
-		outputKey := bytesutil.InternBytes(bb.B)
-		inputKey := ""
 		if !a.aggregateOnlyByTime {
-			tmpLabels.Labels = extractUnneededLabels(tmpLabels.Labels[:0], labels.Labels, a.by, a.without)
+			inputLabels.Labels, outputLabels.Labels = getInputOutputLabels(inputLabels.Labels, outputLabels.Labels, labels.Labels, a.by, a.without)
-			bb.B = marshalLabelsFast(bb.B[:0], tmpLabels.Labels)
+		} else {
-			inputKey = bytesutil.InternBytes(bb.B)
+			outputLabels.Labels = append(outputLabels.Labels, labels.Labels...)
 		}
 
+		bufLen := len(buf)
+		buf = a.compressLabels(buf, inputLabels.Labels, outputLabels.Labels)
 		for _, sample := range ts.Samples {
-			a.pushSample(inputKey, outputKey, sample.Value)
+			if math.IsNaN(sample.Value) {
+				// Skip NaN values
+				continue
 			}
-	}
+			samples = append(samples, pushSample{
-	bbPool.Put(bb)
+				key:   bytesutil.ToUnsafeString(buf[bufLen:]),
-	promutils.PutLabels(tmpLabels)
+				value: sample.Value,
-	promutils.PutLabels(labels)
-}
-
-var bbPool bytesutil.ByteBufferPool
-
-func (a *aggregator) pushSample(inputKey, outputKey string, value float64) {
-	if math.IsNaN(value) {
-		// Skip nan samples
-		return
-	}
-	for _, as := range a.aggrStates {
-		as.pushSample(inputKey, outputKey, value)
-	}
-}
-
-func extractUnneededLabels(dst, labels []prompbmarshal.Label, by, without []string) []prompbmarshal.Label {
-	if len(without) > 0 {
-		for _, label := range labels {
-			if hasInArray(label.Name, without) {
-				dst = append(dst, label)
-			}
-		}
-	} else {
-		for _, label := range labels {
-			if !hasInArray(label.Name, by) {
-				dst = append(dst, label)
-			}
-		}
-	}
-	return dst
-}
-
-func removeUnneededLabels(dst, labels []prompbmarshal.Label, by, without []string) []prompbmarshal.Label {
-	if len(without) > 0 {
-		for _, label := range labels {
-			if !hasInArray(label.Name, without) {
-				dst = append(dst, label)
-			}
-		}
-	} else {
-		for _, label := range labels {
-			if hasInArray(label.Name, by) {
-				dst = append(dst, label)
-			}
-		}
-	}
-	return dst
-}
-
-func hasInArray(name string, a []string) bool {
-	for _, s := range a {
-		if name == s {
-			return true
-		}
-	}
-	return false
-}
-
-func marshalLabelsFast(dst []byte, labels []prompbmarshal.Label) []byte {
-	dst = encoding.MarshalUint32(dst, uint32(len(labels)))
-	for _, label := range labels {
-		dst = encoding.MarshalUint32(dst, uint32(len(label.Name)))
-		dst = append(dst, label.Name...)
-		dst = encoding.MarshalUint32(dst, uint32(len(label.Value)))
-		dst = append(dst, label.Value...)
-	}
-	return dst
-}
-
-func unmarshalLabelsFast(dst []prompbmarshal.Label, src []byte) ([]prompbmarshal.Label, error) {
-	if len(src) < 4 {
-		return dst, fmt.Errorf("cannot unmarshal labels count from %d bytes; needs at least 4 bytes", len(src))
-	}
-	n := encoding.UnmarshalUint32(src)
-	src = src[4:]
-	for i := uint32(0); i < n; i++ {
-		// Unmarshal label name
-		if len(src) < 4 {
-			return dst, fmt.Errorf("cannot unmarshal label name length from %d bytes; needs at least 4 bytes", len(src))
-		}
-		labelNameLen := encoding.UnmarshalUint32(src)
-		src = src[4:]
-		if uint32(len(src)) < labelNameLen {
-			return dst, fmt.Errorf("cannot unmarshal label name from %d bytes; needs at least %d bytes", len(src), labelNameLen)
-		}
-		labelName := bytesutil.InternBytes(src[:labelNameLen])
-		src = src[labelNameLen:]
-
-		// Unmarshal label value
-		if len(src) < 4 {
-			return dst, fmt.Errorf("cannot unmarshal label value length from %d bytes; needs at least 4 bytes", len(src))
-		}
-		labelValueLen := encoding.UnmarshalUint32(src)
-		src = src[4:]
-		if uint32(len(src)) < labelValueLen {
-			return dst, fmt.Errorf("cannot unmarshal label value from %d bytes; needs at least %d bytes", len(src), labelValueLen)
-		}
-		labelValue := bytesutil.InternBytes(src[:labelValueLen])
-		src = src[labelValueLen:]
-
-		dst = append(dst, prompbmarshal.Label{
-			Name:  labelName,
-			Value: labelValue,
 			})
 		}
-	if len(src) > 0 {
-		return dst, fmt.Errorf("unexpected non-empty tail after unmarshaling labels; tail length is %d bytes", len(src))
 	}
-	return dst, nil
+	ctx.samples = samples
+	ctx.buf = buf
+
+	if a.da != nil {
+		a.da.pushSamples(samples)
+	} else {
+		a.pushSamples(samples)
+	}
+}
+
+func (a *aggregator) compressLabels(dst []byte, inputLabels, outputLabels []prompbmarshal.Label) []byte {
+	bb := bbPool.Get()
+	bb.B = a.lc.Compress(bb.B, inputLabels)
+	dst = encoding.MarshalVarUint64(dst, uint64(len(bb.B)))
+	dst = append(dst, bb.B...)
+	bbPool.Put(bb)
+	dst = a.lc.Compress(dst, outputLabels)
+	return dst
+}
+
+func (a *aggregator) decompressLabels(dst []prompbmarshal.Label, key string) []prompbmarshal.Label {
+	dst = a.lc.Decompress(dst, bytesutil.ToUnsafeBytes(key))
+	return dst
+}
+
+func getOutputKey(key string) string {
+	src := bytesutil.ToUnsafeBytes(key)
+	tail, inputKeyLen, err := encoding.UnmarshalVarUint64(src)
+	if err != nil {
+		logger.Panicf("BUG: cannot unmarshal inputKeyLen: %s", err)
+	}
+	outputKey := tail[inputKeyLen:]
+	return bytesutil.ToUnsafeString(outputKey)
+}
+
+func getInputOutputKey(key string) (string, string) {
+	src := bytesutil.ToUnsafeBytes(key)
+	tail, inputKeyLen, err := encoding.UnmarshalVarUint64(src)
+	if err != nil {
+		logger.Panicf("BUG: cannot unmarshal inputKeyLen: %s", err)
+	}
+	inputKey := tail[:inputKeyLen]
+	outputKey := tail[inputKeyLen:]
+	return bytesutil.ToUnsafeString(inputKey), bytesutil.ToUnsafeString(outputKey)
+}
+
+func (a *aggregator) pushSamples(samples []pushSample) {
+	for _, as := range a.aggrStates {
+		as.pushSamples(samples)
+	}
+}
+
+type pushCtx struct {
+	samples      []pushSample
+	labels       promutils.Labels
+	inputLabels  promutils.Labels
+	outputLabels promutils.Labels
+	buf          []byte
+}
+
+func (ctx *pushCtx) reset() {
+	clear(ctx.samples)
+	ctx.samples = ctx.samples[:0]
+
+	ctx.labels.Reset()
+	ctx.inputLabels.Reset()
+	ctx.outputLabels.Reset()
+	ctx.buf = ctx.buf[:0]
+}
+
+type pushSample struct {
+	key   string
+	value float64
+}
+
+func getPushCtx() *pushCtx {
+	v := pushCtxPool.Get()
+	if v == nil {
+		return &pushCtx{}
+	}
+	return v.(*pushCtx)
+}
+
+func putPushCtx(ctx *pushCtx) {
+	ctx.reset()
+	pushCtxPool.Put(ctx)
+}
+
+var pushCtxPool sync.Pool
+
+func getInputOutputLabels(dstInput, dstOutput, labels []prompbmarshal.Label, by, without []string) ([]prompbmarshal.Label, []prompbmarshal.Label) {
+	if len(without) > 0 {
+		for _, label := range labels {
+			if slices.Contains(without, label.Name) {
+				dstInput = append(dstInput, label)
+			} else {
+				dstOutput = append(dstOutput, label)
+			}
+		}
+	} else {
+		for _, label := range labels {
+			if !slices.Contains(by, label.Name) {
+				dstInput = append(dstInput, label)
+			} else {
+				dstOutput = append(dstOutput, label)
+			}
+		}
+	}
+	return dstInput, dstOutput
 }
 
 type flushCtx struct {
-	skipAggrSuffix bool
+	a *aggregator
-	suffix         string
 
 	tss     []prompbmarshal.TimeSeries
 	labels  []prompbmarshal.Label
@@ -736,16 +811,12 @@ func (ctx *flushCtx) reset() {
 	ctx.samples = ctx.samples[:0]
 }
 
-func (ctx *flushCtx) appendSeries(labelsMarshaled, suffix string, timestamp int64, value float64) {
+func (ctx *flushCtx) appendSeries(key, suffix string, timestamp int64, value float64) {
-	var err error
 	labelsLen := len(ctx.labels)
 	samplesLen := len(ctx.samples)
-	ctx.labels, err = unmarshalLabelsFast(ctx.labels, bytesutil.ToUnsafeBytes(labelsMarshaled))
+	ctx.labels = ctx.a.decompressLabels(ctx.labels, key)
-	if err != nil {
+	if !ctx.a.keepMetricNames {
-		logger.Panicf("BUG: cannot unmarshal labels from output key: %s", err)
+		ctx.labels = addMetricSuffix(ctx.labels, labelsLen, ctx.a.suffix, suffix)
-	}
-	if !ctx.skipAggrSuffix {
-		ctx.labels = addMetricSuffix(ctx.labels, labelsLen, ctx.suffix, suffix)
 	}
 	ctx.samples = append(ctx.samples, prompbmarshal.Sample{
 		Timestamp: timestamp,
@@ -757,15 +828,13 @@ func (ctx *flushCtx) appendSeries(labelsMarshaled, suffix string, timestamp int6
 	})
 }
 
-func (ctx *flushCtx) appendSeriesWithExtraLabel(labelsMarshaled, suffix string, timestamp int64, value float64, extraName, extraValue string) {
+func (ctx *flushCtx) appendSeriesWithExtraLabel(key, suffix string, timestamp int64, value float64, extraName, extraValue string) {
-	var err error
 	labelsLen := len(ctx.labels)
 	samplesLen := len(ctx.samples)
-	ctx.labels, err = unmarshalLabelsFast(ctx.labels, bytesutil.ToUnsafeBytes(labelsMarshaled))
+	ctx.labels = ctx.a.decompressLabels(ctx.labels, key)
-	if err != nil {
+	if !ctx.a.keepMetricNames {
-		logger.Panicf("BUG: cannot unmarshal labels from output key: %s", err)
+		ctx.labels = addMetricSuffix(ctx.labels, labelsLen, ctx.a.suffix, suffix)
 	}
-	ctx.labels = addMetricSuffix(ctx.labels, labelsLen, ctx.suffix, suffix)
 	ctx.labels = append(ctx.labels, prompbmarshal.Label{
 		Name:  extraName,
 		Value: extraValue,
@@ -843,3 +912,5 @@ func sortAndRemoveDuplicates(a []string) []string {
 	}
 	return dst
 }
+
+var bbPool bytesutil.ByteBufferPool

lib/streamaggr/streamaggr_test.go

@@ -56,9 +56,43 @@ func TestAggregatorsFailure(t *testing.T) {
 `)
 
 	// Negative interval
-	f(`- interval: -5m`)
+	f(`
+- outputs: [total]
+  interval: -5m
+`)
 	// Too small interval
-	f(`- interval: 10ms`)
+	f(`
+- outputs: [total]
+  interval: 10ms
+`)
+
+	// interval isn't multiple of dedup_interval
+	f(`
+- interval: 1m
+  dedup_interval: 35s
+  outputs: ["quantiles"]
+`)
+
+	// dedup_interval is bigger than dedup_interval
+	f(`
+- interval: 1m
+  dedup_interval: 1h
+  outputs: ["quantiles"]
+`)
+
+	// keep_metric_names is set for multiple inputs
+	f(`
+- interval: 1m
+  keep_metric_names: true
+  outputs: ["total", "increase"]
+`)
+
+	// keep_metric_names is set for unsupported input
+	f(`
+- interval: 1m
+  keep_metric_names: true
+  outputs: ["histogram_bucket"]
+`)
 
 	// Invalid input_relabel_configs
 	f(`
@@ -477,6 +511,17 @@ foo:1m_by_abc_sum_samples{abc="456"} 8
 `, `
 foo 123
 bar{baz="qwe"} 4.34
+`, `bar:1m_total{baz="qwe"} 4.34
+foo:1m_total 123
+`, "11")
+
+	// total_prometheus output for non-repeated series
+	f(`
+- interval: 1m
+  outputs: [total_prometheus]
+`, `
+foo 123
+bar{baz="qwe"} 4.34
 `, `bar:1m_total{baz="qwe"} 0
 foo:1m_total 0
 `, "11")
@@ -494,6 +539,25 @@ foo{baz="qwe"} -5
 bar{baz="qwer"} 343
 bar{baz="qwer"} 344
 foo{baz="qwe"} 10
+`, `bar:1m_total{baz="qwe"} 6.34
+bar:1m_total{baz="qwer"} 344
+foo:1m_total 123
+foo:1m_total{baz="qwe"} 10
+`, "11111111")
+
+	// total_prometheus output for repeated series
+	f(`
+- interval: 1m
+  outputs: [total_prometheus]
+`, `
+foo 123
+bar{baz="qwe"} 1.32
+bar{baz="qwe"} 4.34
+bar{baz="qwe"} 2
+foo{baz="qwe"} -5
+bar{baz="qwer"} 343
+bar{baz="qwer"} 344
+foo{baz="qwe"} 10
 `, `bar:1m_total{baz="qwe"} 5.02
 bar:1m_total{baz="qwer"} 1
 foo:1m_total 0
@@ -514,6 +578,24 @@ foo{baz="qwe"} -5
 bar{baz="qwer"} 343
 bar{baz="qwer"} 344
 foo{baz="qwe"} 10
+`, `bar:1m_total 350.34
+foo:1m_total 133
+`, "11111111")
+
+	// total_prometheus output for repeated series with group by __name__
+	f(`
+- interval: 1m
+  by: [__name__]
+  outputs: [total_prometheus]
+`, `
+foo 123
+bar{baz="qwe"} 1.32
+bar{baz="qwe"} 4.34
+bar{baz="qwe"} 2
+foo{baz="qwe"} -5
+bar{baz="qwer"} 343
+bar{baz="qwer"} 344
+foo{baz="qwe"} 10
 `, `bar:1m_total 6.02
 foo:1m_total 15
 `, "11111111")
@@ -525,6 +607,17 @@ foo:1m_total 15
 `, `
 foo 123
 bar{baz="qwe"} 4.34
+`, `bar:1m_increase{baz="qwe"} 4.34
+foo:1m_increase 123
+`, "11")
+
+	// increase_prometheus output for non-repeated series
+	f(`
+- interval: 1m
+  outputs: [increase_prometheus]
+`, `
+foo 123
+bar{baz="qwe"} 4.34
 `, `bar:1m_increase{baz="qwe"} 0
 foo:1m_increase 0
 `, "11")
@@ -542,12 +635,30 @@ foo{baz="qwe"} -5
 bar{baz="qwer"} 343
 bar{baz="qwer"} 344
 foo{baz="qwe"} 10
+`, `bar:1m_increase{baz="qwe"} 6.34
+bar:1m_increase{baz="qwer"} 344
+foo:1m_increase 123
+foo:1m_increase{baz="qwe"} 10
+`, "11111111")
+
+	// increase_prometheus output for repeated series
+	f(`
+- interval: 1m
+  outputs: [increase_prometheus]
+`, `
+foo 123
+bar{baz="qwe"} 1.32
+bar{baz="qwe"} 4.34
+bar{baz="qwe"} 2
+foo{baz="qwe"} -5
+bar{baz="qwer"} 343
+bar{baz="qwer"} 344
+foo{baz="qwe"} 10
 `, `bar:1m_increase{baz="qwe"} 5.02
 bar:1m_increase{baz="qwer"} 1
 foo:1m_increase 0
 foo:1m_increase{baz="qwe"} 15
 `, "11111111")
-
 	// multiple aggregate configs
 	f(`
 - interval: 1m
@@ -750,7 +861,7 @@ func TestAggregatorsWithDedupInterval(t *testing.T) {
 			}
 			tssOutputLock.Unlock()
 		}
-		const dedupInterval = time.Hour
+		const dedupInterval = 30 * time.Second
 		a, err := newAggregatorsFromData([]byte(config), pushFunc, dedupInterval)
 		if err != nil {
 			t.Fatalf("cannot initialize aggregators: %s", err)

lib/streamaggr/streamaggr_timing_test.go

@@ -11,9 +11,12 @@ import (
 func BenchmarkAggregatorsPushByJobAvg(b *testing.B) {
 	for _, output := range []string{
 		"total",
+		"total_prometheus",
 		"increase",
+		"increase_prometheus",
 		"count_series",
 		"count_samples",
+		"unique_samples",
 		"sum_samples",
 		"last",
 		"min",
@@ -44,7 +47,7 @@ func benchmarkAggregatorsPush(b *testing.B, output string) {
 	}
 	defer a.MustStop()
 
-	const loops = 5
+	const loops = 10
 
 	b.ReportAllocs()
 	b.SetBytes(int64(len(benchSeries) * loops))
@@ -52,7 +55,17 @@ func benchmarkAggregatorsPush(b *testing.B, output string) {
 		var matchIdxs []byte
 		for pb.Next() {
 			for i := 0; i < loops; i++ {
-				matchIdxs = a.Push(benchSeries, matchIdxs)
+				series := benchSeries
+				for len(series) > 0 {
+					chunk := series
+					if len(chunk) > 1_000 {
+						chunk = series[:1_000]
+						series = series[len(chunk):]
+					} else {
+						series = nil
+					}
+					matchIdxs = a.Push(chunk, matchIdxs)
+				}
 			}
 		}
 	})

lib/streamaggr/sum_samples.go

@@ -1,6 +1,7 @@
 package streamaggr
 
 import (
+	"strings"
 	"sync"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
@@ -21,18 +22,23 @@ func newSumSamplesAggrState() *sumSamplesAggrState {
 	return &sumSamplesAggrState{}
 }
 
-func (as *sumSamplesAggrState) pushSample(_, outputKey string, value float64) {
+func (as *sumSamplesAggrState) pushSamples(samples []pushSample) {
-again:
+	for i := range samples {
+		s := &samples[i]
+		outputKey := getOutputKey(s.key)
+
+	again:
 		v, ok := as.m.Load(outputKey)
 		if !ok {
 			// The entry is missing in the map. Try creating it.
 			v = &sumSamplesStateValue{
-			sum: value,
+				sum: s.value,
 			}
+			outputKey = strings.Clone(outputKey)
 			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if !loaded {
 				// The new entry has been successfully created.
-			return
+				continue
 			}
 			// Use the entry created by a concurrent goroutine.
 			v = vNew
@@ -41,7 +47,7 @@ again:
 		sv.mu.Lock()
 		deleted := sv.deleted
 		if !deleted {
-		sv.sum += value
+			sv.sum += s.value
 		}
 		sv.mu.Unlock()
 		if deleted {
@@ -49,6 +55,7 @@ again:
 			// Try obtaining and updating the entry again.
 			goto again
 		}
+	}
 }
 
 func (as *sumSamplesAggrState) appendSeriesForFlush(ctx *flushCtx) {

lib/streamaggr/total.go

@@ -2,6 +2,7 @@ package streamaggr
 
 import (
 	"math"
+	"strings"
 	"sync"
 	"time"
 
@@ -12,7 +13,9 @@ import (
 type totalAggrState struct {
 	m sync.Map
 
-	ignoreInputDeadline uint64
+	suffix            string
+	resetTotalOnFlush bool
+	keepFirstSample   bool
 	stalenessSecs     uint64
 }
 
@@ -29,27 +32,34 @@ type lastValueState struct {
 	deleteDeadline uint64
 }
 
-func newTotalAggrState(interval time.Duration, stalenessInterval time.Duration) *totalAggrState {
+func newTotalAggrState(stalenessInterval time.Duration, resetTotalOnFlush, keepFirstSample bool) *totalAggrState {
-	currentTime := fasttime.UnixTimestamp()
-	intervalSecs := roundDurationToSecs(interval)
 	stalenessSecs := roundDurationToSecs(stalenessInterval)
+	suffix := "total"
+	if resetTotalOnFlush {
+		suffix = "increase"
+	}
 	return &totalAggrState{
-		ignoreInputDeadline: currentTime + intervalSecs,
+		suffix:            suffix,
+		resetTotalOnFlush: resetTotalOnFlush,
+		keepFirstSample:   keepFirstSample,
 		stalenessSecs:     stalenessSecs,
 	}
 }
 
-func (as *totalAggrState) pushSample(inputKey, outputKey string, value float64) {
+func (as *totalAggrState) pushSamples(samples []pushSample) {
-	currentTime := fasttime.UnixTimestamp()
+	deleteDeadline := fasttime.UnixTimestamp() + as.stalenessSecs
-	deleteDeadline := currentTime + as.stalenessSecs
+	for i := range samples {
+		s := &samples[i]
+		inputKey, outputKey := getInputOutputKey(s.key)
 
-again:
+	again:
 		v, ok := as.m.Load(outputKey)
 		if !ok {
 			// The entry is missing in the map. Try creating it.
 			v = &totalStateValue{
 				lastValues: make(map[string]*lastValueState),
 			}
+			outputKey = strings.Clone(outputKey)
 			vNew, loaded := as.m.LoadOrStore(outputKey, v)
 			if loaded {
 				// Use the entry created by a concurrent goroutine.
@@ -63,16 +73,18 @@ again:
 			lv, ok := sv.lastValues[inputKey]
 			if !ok {
 				lv = &lastValueState{}
+				inputKey = strings.Clone(inputKey)
 				sv.lastValues[inputKey] = lv
 			}
-		d := value
+			if ok || as.keepFirstSample {
-		if ok && lv.value <= value {
+				if s.value >= lv.value {
-			d = value - lv.value
+					sv.total += s.value - lv.value
+				} else {
+					// counter reset
+					sv.total += s.value
 				}
-		if ok || currentTime > as.ignoreInputDeadline {
-			sv.total += d
 			}
-		lv.value = value
+			lv.value = s.value
 			lv.deleteDeadline = deleteDeadline
 			sv.deleteDeadline = deleteDeadline
 		}
@@ -82,6 +94,7 @@ again:
 			// Try obtaining and updating the entry again.
 			goto again
 		}
+	}
 }
 
 func (as *totalAggrState) removeOldEntries(currentTime uint64) {
@@ -123,7 +136,9 @@ func (as *totalAggrState) appendSeriesForFlush(ctx *flushCtx) {
 		sv := v.(*totalStateValue)
 		sv.mu.Lock()
 		total := sv.total
-		if math.Abs(sv.total) >= (1 << 53) {
+		if as.resetTotalOnFlush {
+			sv.total = 0
+		} else if math.Abs(sv.total) >= (1 << 53) {
 			// It is time to reset the entry, since it starts losing float64 precision
 			sv.total = 0
 		}
@@ -131,7 +146,7 @@ func (as *totalAggrState) appendSeriesForFlush(ctx *flushCtx) {
 		sv.mu.Unlock()
 		if !deleted {
 			key := k.(string)
-			ctx.appendSeries(key, "total", currentTimeMsec, total)
+			ctx.appendSeries(key, as.suffix, currentTimeMsec, total)
 		}
 		return true
 	})

lib/streamaggr/unique_samples.go

@@ -0,0 +1,82 @@
+package streamaggr
+
+import (
+	"strings"
+	"sync"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/fasttime"
+)
+
+// uniqueSamplesAggrState calculates output=unique_samples, e.g. the number of unique sample values.
+type uniqueSamplesAggrState struct {
+	m sync.Map
+}
+
+type uniqueSamplesStateValue struct {
+	mu      sync.Mutex
+	m       map[float64]struct{}
+	deleted bool
+}
+
+func newUniqueSamplesAggrState() *uniqueSamplesAggrState {
+	return &uniqueSamplesAggrState{}
+}
+
+func (as *uniqueSamplesAggrState) pushSamples(samples []pushSample) {
+	for i := range samples {
+		s := &samples[i]
+		outputKey := getOutputKey(s.key)
+
+	again:
+		v, ok := as.m.Load(outputKey)
+		if !ok {
+			// The entry is missing in the map. Try creating it.
+			v = &uniqueSamplesStateValue{
+				m: map[float64]struct{}{
+					s.value: {},
+				},
+			}
+			outputKey = strings.Clone(outputKey)
+			vNew, loaded := as.m.LoadOrStore(outputKey, v)
+			if !loaded {
+				// The new entry has been successfully created.
+				continue
+			}
+			// Use the entry created by a concurrent goroutine.
+			v = vNew
+		}
+		sv := v.(*uniqueSamplesStateValue)
+		sv.mu.Lock()
+		deleted := sv.deleted
+		if !deleted {
+			if _, ok := sv.m[s.value]; !ok {
+				sv.m[s.value] = struct{}{}
+			}
+		}
+		sv.mu.Unlock()
+		if deleted {
+			// The entry has been deleted by the concurrent call to appendSeriesForFlush
+			// Try obtaining and updating the entry again.
+			goto again
+		}
+	}
+}
+
+func (as *uniqueSamplesAggrState) appendSeriesForFlush(ctx *flushCtx) {
+	currentTimeMsec := int64(fasttime.UnixTimestamp()) * 1000
+	m := &as.m
+	m.Range(func(k, v interface{}) bool {
+		// Atomically delete the entry from the map, so new entry is created for the next flush.
+		m.Delete(k)
+
+		sv := v.(*uniqueSamplesStateValue)
+		sv.mu.Lock()
+		n := len(sv.m)
+		// Mark the entry as deleted, so it won't be updated anymore by concurrent pushSample() calls.
+		sv.deleted = true
+		sv.mu.Unlock()
+		key := k.(string)
+		ctx.appendSeries(key, "unique_samples", currentTimeMsec, float64(n))
+		return true
+	})
+}