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 00:42:26 +00:00
|
|
|
package streamaggr
|
|
|
|
|
|
|
|
import (
|
|
|
|
"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
|
2024-06-07 14:24:09 +00:00
|
|
|
m map[string]dedupAggrSample
|
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 00:42:26 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
type dedupAggrSample struct {
|
2024-03-12 21:47:29 +00:00
|
|
|
value float64
|
|
|
|
timestamp int64
|
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 00:42:26 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
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 {
|
2024-03-04 14:44:31 +00:00
|
|
|
n += uint64(len(k)) + uint64(unsafe.Sizeof(k)+unsafe.Sizeof(s))
|
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 00:42:26 +00:00
|
|
|
}
|
|
|
|
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)
|
|
|
|
}
|
|
|
|
|
2024-03-03 23:21:39 +00:00
|
|
|
func getDedupFlushCtx() *dedupFlushCtx {
|
|
|
|
v := dedupFlushCtxPool.Get()
|
|
|
|
if v == nil {
|
|
|
|
return &dedupFlushCtx{}
|
|
|
|
}
|
|
|
|
return v.(*dedupFlushCtx)
|
|
|
|
}
|
|
|
|
|
|
|
|
func putDedupFlushCtx(ctx *dedupFlushCtx) {
|
|
|
|
ctx.reset()
|
|
|
|
dedupFlushCtxPool.Put(ctx)
|
|
|
|
}
|
|
|
|
|
|
|
|
var dedupFlushCtxPool sync.Pool
|
|
|
|
|
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 00:42:26 +00:00
|
|
|
type dedupFlushCtx struct {
|
|
|
|
samples []pushSample
|
|
|
|
}
|
|
|
|
|
|
|
|
func (ctx *dedupFlushCtx) reset() {
|
|
|
|
clear(ctx.samples)
|
|
|
|
ctx.samples = ctx.samples[:0]
|
|
|
|
}
|
|
|
|
|
2024-03-04 17:12:06 +00:00
|
|
|
func (da *dedupAggr) flush(f func(samples []pushSample), resetState bool) {
|
2024-03-04 15:57:50 +00:00
|
|
|
var wg sync.WaitGroup
|
2024-03-03 23:21:39 +00:00
|
|
|
for i := range da.shards {
|
2024-03-04 15:57:50 +00:00
|
|
|
flushConcurrencyCh <- struct{}{}
|
|
|
|
wg.Add(1)
|
|
|
|
go func(shard *dedupAggrShard) {
|
|
|
|
defer func() {
|
|
|
|
<-flushConcurrencyCh
|
|
|
|
wg.Done()
|
|
|
|
}()
|
|
|
|
|
|
|
|
ctx := getDedupFlushCtx()
|
2024-03-04 17:12:06 +00:00
|
|
|
shard.flush(ctx, f, resetState)
|
2024-03-04 15:57:50 +00:00
|
|
|
putDedupFlushCtx(ctx)
|
|
|
|
}(&da.shards[i])
|
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 00:42:26 +00:00
|
|
|
}
|
2024-03-04 15:57:50 +00:00
|
|
|
wg.Wait()
|
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 00:42:26 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
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 {
|
2024-06-07 14:24:09 +00:00
|
|
|
m = make(map[string]dedupAggrSample, len(samples))
|
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 00:42:26 +00:00
|
|
|
das.m = m
|
|
|
|
}
|
|
|
|
for _, sample := range samples {
|
2024-03-12 21:47:29 +00:00
|
|
|
s, ok := m[sample.key]
|
|
|
|
if !ok {
|
2024-06-07 14:24:09 +00:00
|
|
|
key := bytesutil.InternString(sample.key)
|
|
|
|
m[key] = dedupAggrSample{
|
2024-03-12 21:47:29 +00:00
|
|
|
value: sample.value,
|
|
|
|
timestamp: sample.timestamp,
|
|
|
|
}
|
|
|
|
continue
|
|
|
|
}
|
2024-03-17 19:37:14 +00:00
|
|
|
// Update the existing value according to logic described at https://docs.victoriametrics.com/#deduplication
|
2024-03-12 21:47:29 +00:00
|
|
|
if sample.timestamp > s.timestamp || (sample.timestamp == s.timestamp && sample.value > s.value) {
|
2024-06-07 14:24:09 +00:00
|
|
|
key := bytesutil.InternString(sample.key)
|
|
|
|
m[key] = dedupAggrSample{
|
|
|
|
value: sample.value,
|
|
|
|
timestamp: sample.timestamp,
|
|
|
|
}
|
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 00:42:26 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2024-03-04 17:12:06 +00:00
|
|
|
func (das *dedupAggrShard) flush(ctx *dedupFlushCtx, f func(samples []pushSample), resetState bool) {
|
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 00:42:26 +00:00
|
|
|
das.mu.Lock()
|
|
|
|
|
|
|
|
m := das.m
|
2024-03-04 17:12:06 +00:00
|
|
|
if resetState && len(m) > 0 {
|
2024-06-07 14:24:09 +00:00
|
|
|
das.m = make(map[string]dedupAggrSample, len(m))
|
2024-03-04 14:44:31 +00:00
|
|
|
}
|
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 00:42:26 +00:00
|
|
|
|
|
|
|
das.mu.Unlock()
|
|
|
|
|
|
|
|
if len(m) == 0 {
|
|
|
|
return
|
|
|
|
}
|
|
|
|
|
|
|
|
dstSamples := ctx.samples
|
|
|
|
for key, s := range m {
|
|
|
|
dstSamples = append(dstSamples, pushSample{
|
2024-03-17 19:37:14 +00:00
|
|
|
key: key,
|
|
|
|
value: s.value,
|
|
|
|
timestamp: s.timestamp,
|
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 00:42:26 +00:00
|
|
|
})
|
|
|
|
|
2024-03-03 23:21:39 +00:00
|
|
|
// Limit the number of samples per each flush in order to limit memory usage.
|
|
|
|
if len(dstSamples) >= 100_000 {
|
|
|
|
f(dstSamples)
|
|
|
|
clear(dstSamples)
|
|
|
|
dstSamples = dstSamples[:0]
|
|
|
|
}
|
|
|
|
}
|
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 00:42:26 +00:00
|
|
|
f(dstSamples)
|
2024-03-03 23:21:39 +00:00
|
|
|
ctx.samples = dstSamples
|
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 00:42:26 +00:00
|
|
|
}
|