Commit Graph

73 Commits

Author SHA1 Message Date
Aliaksandr Valialkin
9ed9e766e8
lib/streamaggr: improve performance for dedupAggr.sizeBytes() and dedupAggr.itemsCount()
These functions are called every time `/metrics` page is scraped, so it would be great
if they could be sped up for the cases when dedupAggr tracks tens of millions of active time series.
2024-06-10 16:00:05 +02:00
Aliaksandr Valialkin
387c22da49
lib/streamaggr: remove flushState arg at dedupAggr.flush(), since it is always set to true in production 2024-06-10 16:00:05 +02:00
Aliaksandr Valialkin
32aa0751a1
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 16:35:53 +02:00
Roman Khavronenko
78121642df
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 16:35:52 +02:00
Andrii Chubatiuk
93cd08f15f
lib/streamaggr: metrics to track dropped, nan samples and samples lag (#6358)
### Describe Your Changes

Added streamaggr metrics to:
 - `vm_streamaggr_samples_lag_seconds` - samples lag
- `vm_streamaggr_ignored_samples_total{reason="nan"}` - ignored NaN
samples
- `vm_streamaggr_ignored_samples_total{reason="too_old"}` - ignored old
samples

(cherry picked from commit 185fac03b3)
2024-06-06 19:22:45 +02:00
Roman Khavronenko
3e8b5e74d5
lib/streamaggr: skip empty aggregators (#6307)
Prevent excessive resource usage when stream aggregation config file
contains no matchers by prevent pushing data into Aggregators object.
Before this change a lot of extra work was invoked without reason.

Signed-off-by: hagen1778 <roman@victoriametrics.com>

(cherry picked from commit 7ce052b32d)
Signed-off-by: hagen1778 <roman@victoriametrics.com>
2024-05-20 14:46:36 +02:00
Andrii Chubatiuk
fe332c3419
app/vmagent: add global aggregator (#6268)
Add global stream aggregation for VMAgent

https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5467
(cherry picked from commit f153f54d11)
2024-05-17 14:01:31 +02:00
Aliaksandr Valialkin
5dbc4ad5ef
lib/streamaggr: properly return output key from getOutputKey
The bug has been introduced in cc2647d212

(cherry picked from commit b617dc9c0b)
2024-05-16 09:35:53 +02:00
Aliaksandr Valialkin
32193b6059
lib/encoding: optimize UnmarshalVarUint64, UnmarshalVarInt64 and UnmarshalBytes a bit
Change the return values for these functions - now they return the unmarshaled result plus
the size of the unmarshaled result in bytes, so the caller could re-slice the src for further unmarshaling.

This improves performance of these functions in hot loops of VictoriaLogs a bit.
2024-05-14 01:30:25 +02:00
Andrii Chubatiuk
b9eb527d98
lib/streamaggr: added rate_sum and rate_avg to benchmarks, lint fix (#6264)
fixed lint for rate outputs

(cherry picked from commit ce25d68b45)
2024-05-13 16:49:59 +02:00
Andrii Chubatiuk
d9cddf1ad8
lib/streamaggr: added rate and rate_avg output (#6243)
Added `rate` and `rate_avg` output
Resource usage is the same as for increase output, tested on a benchmark

---------

Signed-off-by: hagen1778 <roman@victoriametrics.com>
Co-authored-by: hagen1778 <roman@victoriametrics.com>
(cherry picked from commit 9c3d44c8c9)
2024-05-13 16:49:39 +02:00
Roman Khavronenko
7be6fcd8fd
lib/streamaggr: set correct suffix <output>_prometheus (#6228)
Set correct suffix `<output>_prometheus` for aggregation outputs
`increase_prometheus` and `total_prometheus`
Before, outputs `total` and `total_prometheus` or `increase` and
`increase_prometheus` had the same suffix.

Signed-off-by: hagen1778 <roman@victoriametrics.com>
(cherry picked from commit 8a03e987cb)
2024-05-10 14:29:01 +02:00
Andrii Chubatiuk
f3d65ba902
streamaggr: made labels compressor shared (#6173)
Though labels compressor is quite resource intensive, each aggregator
and deduplicator instance has it's own compressor. Made it shared across
all aggregators to consume less resources while using multiple
aggregators.

Co-authored-by: Roman Khavronenko <hagen1778@gmail.com>
(cherry picked from commit a9283e06a3)
2024-05-10 14:28:59 +02:00
hagen1778
342290275e
app/streamaggr: follow-up after c0e4ccb7b5
* rm vmagent mentions from vminsert flags
* improve documentation wording, add links to related sections
* mention `ignore_first_intervals` in the stream aggr options
* update flags description
* add basic test for config parsing validation

Signed-off-by: hagen1778 <roman@victoriametrics.com>

(cherry picked from commit bae3874e6a)
Signed-off-by: hagen1778 <roman@victoriametrics.com>
2024-04-22 14:39:23 +02:00
Andrii Chubatiuk
131367fb59
lib/streamaggr: add option to ignore first N aggregation intervals (#6137)
Stream aggregation may yield inaccurate results if it processes incomplete data.
This issue can arise when data is sourced from clients that maintain a queue of unsent data, such as Prometheus or vmagent.
 If the queue isn't fully cleared within the aggregation interval, only a portion of the time series may be included in that period, leading to distorted calculations.
To mitigate this we add an option to ignore first N aggregation intervals. It is expected, that client queues
will be cleared during the time while aggregation ignores first N intervals and all subsequent aggregations
will be correct.

(cherry picked from commit c0e4ccb7b5)
Signed-off-by: hagen1778 <roman@victoriametrics.com>
2024-04-22 14:34:36 +02:00
Aliaksandr Valialkin
728bcf0585
all: replace old https://docs.victoriametrics.com/stream-aggregation.html url with the new one - https://docs.victoriametrics.com/stream-aggregation/ 2024-04-18 02:20:00 +02:00
Aliaksandr Valialkin
222e8b5a7b
lib/streamaggr: update the minimum allowed timestamp for incoming samples before flushing the samples to the storage
This should prevent from dropping samples with old timestamps during long flushes.

This is a follow-up for 1cedaf61cb
2024-04-04 02:26:08 +03:00
Aliaksandr Valialkin
00f59d6ddf
all: fix golangci-lint(revive) warnings after 0c0ed61ce7
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/6001
2024-04-03 03:00:45 +03:00
Aliaksandr Valialkin
cd222d6502
lib/streamaggr: ignore out of order samples for last output
This is a follow-up for 6a465f6e29

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5931
2024-03-18 01:03:58 +02:00
Aliaksandr Valialkin
111d0aa2bf
app/{vmagent,vminsert}: add an ability to ignore input samples outside the current aggregation interval for stream aggregation
See https://docs.victoriametrics.com/stream-aggregation.html#ignoring-old-samples
2024-03-17 23:30:46 +02:00
Aliaksandr Valialkin
e70b644f1f
lib/streamaggr: ignore out of order samples when calculating increase, increase_prometheus, total and total_prometheus outputs
Out of order samples may result in unexpected spikes for these outputs.
So it is better to ignore such samples.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5931
2024-03-17 23:24:14 +02:00
Aliaksandr Valialkin
1f753a049a
lib/streamaggr: follow-up for 15e33d56f1
- Properly set pushSample.timestamp when flushing de-duplicated samples to stream aggregation
  This is needed for https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5931

- Re-classify this change as feature instead of bugfix at docs/CHANGELOG.md

- Verify de-duplication logic for samples with different timestamps

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5643
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5939
2024-03-17 23:23:57 +02:00
Andrii Chubatiuk
a58a81d80b
lib/streamaggr: pick sample with bigger timestamp or value on deduplicator (#5939)
Apply the same deduplication logic as in https://docs.victoriametrics.com/#deduplication
This would require more memory for deduplication, since we need to track timestamp
for each record. However, deduplication should become more consistent.

https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5643

---------

Co-authored-by: Roman Khavronenko <roman@victoriametrics.com>
2024-03-17 23:06:37 +02:00
Aliaksandr Valialkin
5817bf1c59
lib/streamaggr: add tests for keep_metric_names and drop_input_labels options 2024-03-06 20:06:23 +02:00
Aliaksandr Valialkin
27b9e8ed3e
app/{vmagent,vminsert}: add -streamAggr.dropInputSamples command-line flag for dropping the specified labels from input samples before deduplication and streaming aggregation 2024-03-05 02:27:27 +02:00
Aliaksandr Valialkin
c38c45d71f
app/{vminsert,vmagent}: allow using -streamAggr.dedupInterval without -streamAggr.config
This allows performing online de-duplication of incoming samples
2024-03-05 00:47:23 +02:00
Aliaksandr Valialkin
4352544d61
lib/streamaggr: do not reset aggregation state after the aggregation took longer than the configured interval
It is better from user PoV preserving this state until the next flush
2024-03-04 20:03:45 +02:00
Aliaksandr Valialkin
9f81450b38
lib/streamaggr: add missing "s" suffix in the warning message when the de-duplication or aggregation couldnt be finished in a timely manner 2024-03-04 19:38:39 +02:00
Aliaksandr Valialkin
d10932bd99
lib/streamaggr: benchmark only flush routines in BenchmarkDedupAggrFlushSerial and BenchmarkAggregatorsFlushSerial 2024-03-04 19:13:50 +02:00
Aliaksandr Valialkin
36ee08cad4
Revert "lib/streamaggr: do not flush dedup shards in parallel"
This reverts commit eb40395a1c.

Reason for revert: it has been appeared that the performance gain on multiple CPU cores
wasn't visible because the benchmark was generating incorrect pushSample.key.

See a207e0bf687d65f5198207477248d70c69284296
2024-03-04 19:13:50 +02:00
Aliaksandr Valialkin
9728aaf5d9
lib/streamaggr: properly generate pushSample.key in benchmarks 2024-03-04 19:13:49 +02:00
Aliaksandr Valialkin
93a057e4e6
lib/streamaggr: reduce the number of pointers at "total" aggregation state
This should reduce load on GC when scanning heap objects.
2024-03-04 19:13:49 +02:00
Aliaksandr Valialkin
9e00d8ad60
lib/streamaggr: use multiple job label values in BenchmarkAggregatorsPush instead of single value
This should make the benchmark closer to production cases
2024-03-04 19:13:48 +02:00
Aliaksandr Valialkin
9773ad200e
lib/streamaggr: use multiple job labels in BenchmarkAggregatorsPush 2024-03-04 19:13:48 +02:00
Aliaksandr Valialkin
482560a1f3
lib/streamaggr: do not flush dedup shards in parallel
This significantly increases CPU usage on systems with many CPU cores, while doesn't reduce flush latency too much
2024-03-04 17:01:42 +02:00
Aliaksandr Valialkin
d7252fce79
lib/streamaggr: reduce memory allocations when registering new series in deduplication and aggregation structs 2024-03-04 17:01:41 +02:00
Aliaksandr Valialkin
402dc14ec0
lib/streamaggr: make aggregate.runFlusher() more roubst and clear 2024-03-04 17:01:41 +02:00
Aliaksandr Valialkin
2ffef39bb3
lib/streamaggr: properly drop samples on the first incomplete interval
Previously samples were dropped on the first incomplete interval and the next complete interval.
Also make sure that the de-duplication is performed just before flushing the aggregate state.
This should help the case then dedup_interval = interval.
2024-03-04 17:01:40 +02:00
Aliaksandr Valialkin
c2dae136b3
lib/streamaggr: explicitly call resetSeries after flushSeries
This makes the code less fragile
2024-03-04 06:23:36 +02:00
Aliaksandr Valialkin
48a425898a
lib/streamaggr: enable time alignment for aggregate flushed to multiples of interval
For example, if `interval: 1m`, then data flush occurs at the end of every minute,
while `interval: 1h` leads to data flush at the end of every hour.

Add `no_align_flush_to_interval` option, which can be used for disabling the alignment.
2024-03-04 06:23:35 +02:00
Aliaksandr Valialkin
d80deaeaf4
lib/streamaggr: ignore the first sample in new time series during staleness_interval seconds after the stream aggregation start for total and increase outputs 2024-03-04 03:04:58 +02:00
Aliaksandr Valialkin
5e9cbfd4db
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 01:22:41 +02:00
Aliaksandr Valialkin
1e741ed6db
lib/streamaggr: add a benchmark for flushing dedup state 2024-03-04 01:22:40 +02:00
Aliaksandr Valialkin
5205972b83
lib/streamaggr: add a benchmark for measuring the performance of aggregator.flush 2024-03-04 01:22:40 +02:00
Aliaksandr Valialkin
8daf7a3f43
lib/streamaggr: add a benchmark for de-duplicating of 1M samples 2024-03-04 01:22:39 +02:00
Aliaksandr Valialkin
0d5d46f9db
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 03:15:43 +02:00
Aliaksandr Valialkin
31f0dc4b97
lib/streamaggr: allow one second aggregation interval 2024-03-01 21:35:43 +02:00
Aliaksandr Valialkin
8187244153
lib/streamaggr: make the BenchmarkAggregatorsPushByJobAvg closer to production case with long list of labels per sample 2024-02-29 02:41:48 +02:00
hagen1778
2ff94b2bfa
lib/streamaggr: fix incorrect err message for min interval value
Signed-off-by: hagen1778 <roman@victoriametrics.com>
2024-01-29 17:27:23 +01:00
Roman Khavronenko
9e9f170fe7
lib/streamaggr: skip unfinished aggregation state on shutdown by default (#5689)
Sending unfinished aggregate states tend to produce unexpected anomalies with lower values than expected.
The old behavior can be restored by specifying `flush_on_shutdown: true` setting in streaming aggregation config

Signed-off-by: hagen1778 <roman@victoriametrics.com>
2024-01-26 22:45:45 +01:00