在VictoriaMetrics集群版本中,-dedup.minScrapeInterval用于数据去重,它可以配置在vmselect和vmstorage的启动参数上:
-
配置在vmselect上:
- 由于vm存储时间戳的时间精度是millisecond,同一个vminsert的数据发往不同vmstorage存储时,存储的是相同的millisecond;
- 故通常在vmselect上配置-dedup.minScrapeInterval=1ms,这样可以去重不同节点的重复数据;
-
配置在vmstorage上:
- 若两个vmagent推送相同的数据时,通常配置vmstorage的-dedup.minScrapeInterval=scrape_interval,这样可以防止单个节点上存储相同的数据;
VictoriaMetrics stores timestamps with millisecond precision, so -dedup.minScrapeInterval=1ms command-line flag must be passed to vmselect nodes when the replication is enabled, so they could de-duplicate replicated samples obtained from distinct vmstorage nodes during querying.
If duplicate data is pushed to VictoriaMetrics from identically configured vmagent instances or Prometheus instances, then the -dedup.minScrapeInterval must be set to scrape_interval from scrape configs according to deduplication docs.
一. vmselect
vm存储timestamps的精度为ms,通常配置vmselect的
- -dedup.minScrapeInterval=1ms
- 这样在多副本的场景下,可以对查询结果进行去重。
查询过程中,相关的去重代码:
// app/vmselect/netstorage/netstorage.go
func (pts *packedTimeseries) Unpack(dst *Result, tbf *tmpBlocksFile, tr storage.TimeRange) error {
...
dedupInterval := storage.GetDedupInterval() //读配置参数-dedup.minScrapeInterval
mergeSortBlocks(dst, sbs, dedupInterval) //执行合并(含去重逻辑)
return nil
}
func mergeSortBlocks(dst *Result, sbh sortBlocksHeap, dedupInterval int64) {
...
//对timestamps和values的值,用dedupInterval去重
timestamps, values := storage.DeduplicateSamples(dst.Timestamps, dst.Values, dedupInterval)
dedups := len(dst.Timestamps) - len(timestamps)
dedupsDuringSelect.Add(dedups)
dst.Timestamps = timestamps
dst.Values = values
}
去重的具体逻辑:
- 以dedupInterval为段(tsNext),每段仅取一个值(该段中ts最大的值);
// lib/storage/dedup.go
func DeduplicateSamples(srcTimestamps []int64, srcValues []float64, dedupInterval int64) ([]int64, []float64) {
....
tsNext := srcTimestamps[0] + dedupInterval - 1
tsNext -= tsNext % dedupInterval
dstTimestamps := srcTimestamps[:0]
dstValues := srcValues[:0]
for i, ts := range srcTimestamps[1:] {
if ts
二. vmstorage
vmstorage配置-dedup.minScrapeInterval的目的,通常是为了在存储上去重:
- 多个vmagent采集相同的target,均push到vm;
- 多个prometheus实例,相同的external_labels,采集相同的target,均push到vm;
此时,可以配置-dedup.minScrapeInterval=scrape_interval,保证一个间隔仅存储一个sample。
vmstorage在启动时,读取dedup.minScrapeInterval的值:
// app/vmstorage/main.go
var (
...
minScrapeInterval = flag.Duration("dedup.minScrapeInterval", 0, "Leave only the last sample in every time series per each discrete interval "+
"equal to -dedup.minScrapeInterval > 0. See https://docs.victoriametrics.com/#deduplication for details")
...
)
func main() {
...
storage.SetDedupInterval(*minScrapeInterval)
...
}
在将内存数据merge到硬盘的parts过程中,会执行去重:
// lib/storage/block_stream_writer.go
func (bsw *blockStreamWriter) WriteExternalBlock(b *Block, ph *partHeader, rowsMerged *uint64) {
...
b.deduplicateSamplesDuringMerge() //去重
headerData, timestampsData, valuesData := b.MarshalData(bsw.timestampsBlockOffset, bsw.valuesBlockOffset)
...
fs.MustWriteData(bsw.timestampsWriter, timestampsData)
fs.MustWriteData(bsw.valuesWriter, valuesData)
...
}
对数据反序列化,然后分别对timestamps和values进行去重:
- dedup.minScrapeInterval
- 首先,对block的数据进行反序列化;
- 然后,调用去重的逻辑,对timestamps和values去重;
// lib/storage/block.go
func (b *Block) deduplicateSamplesDuringMerge() {
if !isDedupEnabled() { //dedup.minScrapeInterval
具体的去重逻辑:
-
这里的去重逻辑,跟vmselect相同;
- 保证一个dedupInterval时间段上,只采用一个数据点(该段中ts最大的值);
-
区别在于:
- vmselect是对查询结果去重,去重的对象为[]float64;
- vmstorage是对merge到磁盘的数据去重,去重的对象为[]int64;
// lib/storage/dedup.go
func deduplicateSamplesDuringMerge(srcTimestamps, srcValues []int64, dedupInterval int64) ([]int64, []int64) {
...
tsNext := srcTimestamps[0] + dedupInterval - 1
tsNext -= tsNext % dedupInterval
dstTimestamps := srcTimestamps[:0]
dstValues := srcValues[:0]
for i, ts := range srcTimestamps[1:] {
if ts
参考:
1.https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html…
服务器托管,北京服务器托管,服务器租用 http://www.fwqtg.net