Commit graph

11 commits

Author SHA1 Message Date
AndrewChubatiuk
c0ca03aee5
review comments 2024-11-25 17:07:38 +02:00
AndrewChubatiuk
4c5ca49297
lib/streamaggr: added aggregation windows 2024-11-25 17:07:32 +02:00
Aliaksandr Valialkin
04981c7a7f
lib/streamaggr: remove resetState arg from aggrState.flushState()
The resetState arg was used only for the BenchmarkAggregatorsFlushInternalSerial benchmark.
This benchmark was testing aggregate state flush performance by keeping the same state across flushes.
The benhmark didn't reflect the performance and scalability of stream aggregation in production,
while it led to non-trivial code changes related to resetState arg handling.

So let's drop the benchmark together with all the code related to resetState handling,
in order to simplify the code at lib/streamaggr a bit.

Thanks to @AndrewChubatiuk for the original idea at https://github.com/VictoriaMetrics/VictoriaMetrics/pull/6314
2024-08-07 11:39:14 +02:00
Aliaksandr Valialkin
86c7afd126
lib/streamaggr: consistently use the same timestamp across all the output aggregated samples in a single aggregation interval
Prevsiously every aggregation output was using its own timestamp for the output aggregated samples
in a single aggregation interval. This could result in unexpected inconsitent timesetamps for the output
aggregated samples.

This commit consistently uses the same timestamp across all the output aggregated samples.
This commit makes sure that the duration between subsequent timestamps strictly equals
the configured aggregation interval.

Thanks to @AndrewChubatiuk for the original idea at https://github.com/VictoriaMetrics/VictoriaMetrics/pull/6314
This commit should help https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4580
2024-08-07 11:39:13 +02:00
Aliaksandr Valialkin
3c02937a34
all: consistently use 'any' instead of 'interface{}'
'any' type is supported starting from Go1.18. Let's consistently use it
instead of 'interface{}' type across the code base, since `any` is easier to read than 'interface{}'.
2024-07-10 00:20:37 +02:00
Aliaksandr Valialkin
36be090cd5
lib/streamaggr: follow-up for 7cb894a777
- Use bytesutil.InternString() instead of strings.Clone() for inputKey and outputKey in aggregatorpushSamples().
  This should reduce string allocation rate, since strings can be re-used between aggrState flushes.
- Reduce memory allocations at dedupAggrShard by storing dedupAggrSample by value in the active series map.
- Remove duplicate call to bytesutil.InternBytes() at Deduplicator, since it is already called inside dedupAggr.pushSamples().
- Add missing string interning at rateAggrState.pushSamples().

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/6402
2024-06-07 16:27:26 +02:00
Roman Khavronenko
7cb894a777
lib/streamaggr: reduce number of inuse objects (#6402)
The main change is getting rid of interning of sample key. It was
discovered that for cases with many unique time series aggregated by
vmagent interned keys could grow up to hundreds of millions of objects.
This has negative impact on the following aspects:
1. It slows down garbage collection cycles, as GC has to scan all inuse
objects periodically. The higher is the number of inuse objects, the
longer it takes/the more CPU it takes.
2. It slows down the hot path of samples aggregation where each key
needs to be looked up in the map first.

The change makes code more fragile, but suppose to provide performance
optimization for heavy-loaded vmagents with stream aggregation enabled.

---------

Signed-off-by: hagen1778 <roman@victoriametrics.com>
Co-authored-by: Aliaksandr Valialkin <valyala@victoriametrics.com>
2024-06-07 15:45:52 +02:00
Aliaksandr Valialkin
6319d029a8
lib/streamaggr: benchmark only flush routines in BenchmarkDedupAggrFlushSerial and BenchmarkAggregatorsFlushSerial 2024-03-04 19:12:28 +02:00
Aliaksandr Valialkin
946814afee
lib/streamaggr: reduce memory allocations when registering new series in deduplication and aggregation structs 2024-03-04 17:00:19 +02:00
Aliaksandr Valialkin
0422ae01ba
lib/streamaggr: flush dedup state and aggregation state in parallel on all the available CPU cores
This should reduce the time needed for aggregation state flush on systems with many CPU cores
2024-03-04 01:21:50 +02:00
Aliaksandr Valialkin
28a9e92b5e
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
2024-03-02 02:42:50 +02:00