Commit Graph

31 Commits

Author SHA1 Message Date
Aliaksandr Valialkin
eb1505ba14
lib/uint64set: go fmt after c0a9b87f46 2024-02-15 14:50:44 +02:00
Aliaksandr Valialkin
c0a9b87f46
lib/mergeset: optimize Set.AddMulti() a bit for len(items) < 10000
This should improve the search speed for time series matching the given label filters
2024-02-15 14:30:15 +02:00
Aliaksandr Valialkin
a7694092b8 Revert "lib/uint64set: allow reusing bucket16 structs inside uint64set.Set via uint64set.Release method"
This reverts commit 7c6d3981bf.

Reason for revert: high contention at bucket16Pool on systems with big number of CPU cores.
This slows down query processing significantly.
2021-07-06 18:21:35 +03:00
Aliaksandr Valialkin
7c6d3981bf lib/uint64set: allow reusing bucket16 structs inside uint64set.Set via uint64set.Release method
This reduces the load on memory allocator in Go runtime in production workload.
2021-07-06 15:35:03 +03:00
Aliaksandr Valialkin
2233d6ed8a lib/uint64set: store pointers to bucket16 instead of bucket16 objects in bucket32
This speeds up bucket32.addBucketAtPos() when bucket32.buckets contains big number of items,
since the copying of bucket16 pointers is much faster than the copying of bucket16 objects.

This is a cpu profile for copying bucket16 objects:

      10ms     13.43s (flat, cum) 32.01% of Total
      10ms      120ms    650:	b.b16his = append(b.b16his[:pos+1], b.b16his[pos:]...)
         .          .    651:	b.b16his[pos] = hi
         .     13.31s    652:	b.buckets = append(b.buckets[:pos+1], b.buckets[pos:]...)
         .          .    653:	b16 := &b.buckets[pos]
         .          .    654:	*b16 = bucket16{}
         .          .    655:	return b16
         .          .    656:}

This is a cpu profile for copying pointers to bucket16:

      10ms      1.14s (flat, cum)  2.19% of Total
         .      100ms    647:	b.b16his = append(b.b16his[:pos+1], b.b16his[pos:]...)
         .          .    648:	b.b16his[pos] = hi
      10ms      700ms    649:	b.buckets = append(b.buckets[:pos+1], b.buckets[pos:]...)
         .      330ms    650:	b16 := &bucket16{}
         .          .    651:	b.buckets[pos] = b16
         .          .    652:	return b16
         .          .    653:}
2021-05-25 14:20:52 +03:00
Aliaksandr Valialkin
9b1999a5ff lib/uint64set: remove memory allocation in getOrCreateSmallPool()
This partially reverts fb82c4b9fa

It has been appeared that the additional memory allocation may result in higher GC pauses.
It is better to spend CPU time on copying bigger bucket16 structs instead of increasing query latencies due to higher GC pauses
2021-04-14 12:30:19 +03:00
Aliaksandr Valialkin
b75c2ce659 lib/uint64set: improve Set.Has() performance scalability on multi-CPU system
Do not update bucket32.hint on Set.Has() call, since it leads to memory ping-pong between CPU cores multi-CPU system
2021-03-29 12:33:47 +03:00
Aliaksandr Valialkin
35bb44d317 lib/uint64set: optimize bucket16.addMulti a bit 2021-03-16 21:09:07 +02:00
Aliaksandr Valialkin
fb82c4b9fa lib/uint64set: reduce the size of bucket16 by storing smallPool by pointer.
This reduces CPU time spent on bucket16 copying inside bucket13.addBucketAtPos.
2021-03-15 17:23:31 +02:00
Aliaksandr Valialkin
eb103e1527 lib/uint64set: optimize Set.AddMulti for large sorted sets 2021-03-15 17:10:40 +02:00
Aliaksandr Valialkin
43504ebd14 lib/uint64set: optimize bucket16.add and bucket16.addMulti a bit 2021-03-15 16:58:12 +02:00
Aliaksandr Valialkin
1bf6cd814d lib/uint64set: remove memory allocation in bucket16.appendTo when sorting smallPool 2021-02-16 15:31:49 +02:00
Aliaksandr Valialkin
188325f0fc lib/promscrape: eliminate data race in stream parse mode
Previously `-promscrape.streamParse` mode could result in garbage labels for the scraped metrics because of data race.
See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/825#issuecomment-723198247
2020-11-07 12:45:17 +02:00
Aliaksandr Valialkin
d3442b40b2 lib/uint64set: optimize adding items to the set via Set.AddMulti 2020-07-21 20:56:59 +03:00
Aliaksandr Valialkin
047849e855 lib/uint64set: remove zero buckets after Set.Intersect 2020-03-27 01:15:58 +02:00
Aliaksandr Valialkin
f3ec424e7d lib/uint64set: small code cleanup and perf tuning
* Remember the last accessed bucket on Has() call.
* Inline fast paths inside Add() and Has() calls.
* Remove fragile code with maxUnsortedBuckets inside bucket32.
2020-03-25 15:30:25 +02:00
Aliaksandr Valialkin
257e61195a lib/uint64set: add missing bucket32.b16his values 2020-01-18 14:26:04 +02:00
Aliaksandr Valialkin
4cc0c44b9e lib/uint64set: optimize Set.Union
This should improve performance for queries over big number of time series
2020-01-18 13:47:03 +02:00
Aliaksandr Valialkin
bc0f897fcb lib/uint64set: reduce memory allocations in Set.AppendTo 2020-01-17 22:33:09 +02:00
Aliaksandr Valialkin
f9289b804a lib/storage: reduce memory allocations when merging metricID sets 2020-01-17 22:10:44 +02:00
Aliaksandr Valialkin
0c8ad08578 lib/uint64set: typo fix in Set.Intersect 2020-01-17 18:10:58 +02:00
Aliaksandr Valialkin
9f027ec176 lib/uint64set: optimize Intersect, Subtract and Union functions
This should improve performance for queries over big number of time series.
2020-01-17 16:11:49 +02:00
Aliaksandr Valialkin
605d588ba6 lib/uint64set: reduce memory usage in Union, Intersect and Subtract methods
Iterate items with newly added Set.ForEach method instead of allocating `[]uint64`
slice for all the items before the iteration.
2020-01-15 12:12:49 +02:00
Aliaksandr Valialkin
695682232f lib/uint64set: reduce memory usage when storing big number of sparse metric_id values 2020-01-03 18:16:44 +02:00
Aliaksandr Valialkin
ca259864e2 lib/storage: return back inmemory inverted index for recent hour
Issues fixed:
- Slow startup times. Now the index is loaded from cache during start.
- High memory usage related to superflouos index copies every 10 seconds.
2019-11-13 13:11:04 +02:00
Aliaksandr Valialkin
5c25070556 lib/uint64set: remove superflouos check for item existence before deleting it in Set.Subtract 2019-11-09 14:19:47 +02:00
Aliaksandr Valialkin
d888b21657 lib/storage: add inmemory inverted index for the last hour
It should improve performance for `last N hours` dashboards with update intervals smaller than 1 hour.
2019-11-08 21:21:07 +02:00
Aliaksandr Valialkin
6be4456d88 lib/{storage,uint64set}: add Set.Union() function and use it 2019-11-04 00:44:37 +02:00
Aliaksandr Valialkin
370cfbb365 lib/uint64set: return an emptry set instead of nil set from Set.Clone, since the caller may add data to the cloned set
This fixes the following panic in v1.28.1:

panic: runtime error: invalid memory address or nil pointer dereference
[signal SIGSEGV: segmentation violation code=0x1 addr=0x10 pc=0x783a7e]

goroutine 1155 [running]:
github.com/VictoriaMetrics/VictoriaMetrics/lib/uint64set.(*Set).Add(0x0, 0x15b3bfb41e8b71ec)
  github.com/VictoriaMetrics/VictoriaMetrics@/lib/uint64set/uint64set.go:57 +0x2e
github.com/VictoriaMetrics/VictoriaMetrics/lib/storage.(*indexSearch).getMetricIDsForRecentHours(0xc5bdc0dd40, 0x16e273f6b50, 0x16e2745d3f0, 0x5b8d95, 0x10, 0x4a2f51, 0xaa01000000000000)
  github.com/VictoriaMetrics/VictoriaMetrics@/lib/storage/index_db.go:1951 +0x260
github.com/VictoriaMetrics/VictoriaMetrics/lib/storage.(*indexSearch).getMetricIDsForTimeRange(0xc5bdc0dd40, 0x16e273f6b50, 0x16e2745d3f0, 0x5b8d95, 0x10, 0xb296c0, 0xc00009cd80, 0x9bc640)
2019-11-01 16:12:44 +02:00
Aliaksandr Valialkin
2cb811b42f lib/uint64set: optimize Set.AppendTo 2019-09-25 00:34:17 +03:00
Aliaksandr Valialkin
b986516fbe lib/storage: create and use lib/uint64set instead of map[uint64]struct{}
This should improve inverted index search performance for filters matching big number of time series,
since `lib/uint64set.Set` is faster than `map[uint64]struct{}` for both `Add` and `Has` calls.
See the corresponding benchmarks in `lib/uint64set`.
2019-09-24 21:17:55 +03:00