关于时序数据库:VictoriaMetrics使用dedupminScrapeInterval进行数据去重

在VictoriaMetrics集群版本中，-dedup.minScrapeInterval用于数据去重，它能够配置在vmselect和vmstorage的启动参数上：

• 配置在vmselect上：

  • 因为vm存储工夫戳的工夫精度是millisecond，同一个vminsert的数据发往不同vmstorage存储时，存储的是雷同的millisecond；
  • 故通常在vmselect上配置-dedup.minScrapeInterval=1ms，这样能够去重不同节点的反复数据；

• 配置在vmstorage上：

  • 若两个vmagent推送雷同的数据时，通常配置vmstorage的-dedup.minScrapeInterval=scrape_interval，这样能够避免单个节点上存储雷同的数据；

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
• 这样在多正本的场景下，能够对查问后果进行去重。

查问过程中，相干的去重代码：

// 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 <= tsNext {
            continue
        }
        dstTimestamps = append(dstTimestamps, srcTimestamps[i])
        dstValues = append(dstValues, srcValues[i])
        tsNext += dedupInterval
        if tsNext < ts {
            tsNext = ts + dedupInterval - 1
            tsNext -= tsNext % dedupInterval
        }
    }
    dstTimestamps = append(dstTimestamps, srcTimestamps[len(srcTimestamps)-1])
    dstValues = append(dstValues, srcValues[len(srcValues)-1])
    return dstTimestamps, dstValues
}

二. vmstorage

vmstorage配置-dedup.minScrapeInterval的目标，通常是为了在存储下来重：

• 多个vmagent采集雷同的target，均push到vm；
• 多个prometheus实例，雷同的external_labels，采集雷同的target，均push到vm；

此时，能够配置-dedup.minScrapeInterval=scrape_interval，保障一个距离仅存储一个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，而后别离对timestamps和values进行去重：

• dedup.minScrapeInterval <= 0时，不须要去重；
• 首先，对block的数据进行反序列化；
• 而后，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 <= 0时，不须要去重
        // Deduplication is disabled
        return
    }
    // Unmarshal block if it isn't unmarshaled yet in order to apply the de-duplication to unmarshaled samples.
    // 1.反序列化
    if err := b.UnmarshalData(); err != nil {
        logger.Panicf("FATAL: cannot unmarshal block: %s", err)
    }
    srcTimestamps := b.timestamps[b.nextIdx:]
    ...
    dedupInterval := GetDedupInterval()
    ...
    srcValues := b.values[b.nextIdx:]
    // 2.调用去重的逻辑
    timestamps, values := deduplicateSamplesDuringMerge(srcTimestamps, srcValues, dedupInterval)
    dedups := len(srcTimestamps) - len(timestamps)
    b.timestamps = b.timestamps[:b.nextIdx+len(timestamps)]
    b.values = b.values[:b.nextIdx+len(values)]
}

具体的去重逻辑：

• 这里的去重逻辑，跟vmselect雷同；

  • 保障一个dedupInterval时间段上，只采纳一个数据点(该段中ts最大的值)；

• 区别在于：

  • vmselect是对查问后果去重，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 <= tsNext {
            continue
        }
        dstTimestamps = append(dstTimestamps, srcTimestamps[i])
        dstValues = append(dstValues, srcValues[i])
        tsNext += dedupInterval
        if tsNext < ts {
            tsNext = ts + dedupInterval - 1
            tsNext -= tsNext % dedupInterval
        }
    }
    dstTimestamps = append(dstTimestamps, srcTimestamps[len(srcTimestamps)-1])
    dstValues = append(dstValues, srcValues[len(srcValues)-1])
    return dstTimestamps, dstValues
}

参考：

1.https://docs.victoriametrics.com/Cluster-VictoriaMetrics.html…