Component schema

TimeSeriesAggregationSample

object1 direct reference

Definition

The frequency at which metric data points are aggregated via these rules. This is a required parameter. If you don't which to downsample your metrics, then set this to the scrape/export interval of your metrics. Otherwise, use a larger number to reduce the granularity of your metrics.

Duration

The amount of time to account for delays in scraping, exporting, and data arrival. We suggest to always add a little bit of a delay to account for this. But you can increase it in case you have larger delays in your scrape and export processes.

Duration

Duration after which a source time series with no new data is dropped from memory. This is a retention bound, and not a guarantee about exported values.

When a source time series is dropped and new data arrives later, that source series starts over: its first new reading becomes a subtraction base rather than an increment, so no increment is derived across the gap. The aggregated output series itself continues, and its cumulative total is not reset, because resetting it would be read downstream as a counter reset.

A source time series may be held slightly beyond this duration while an aggregation window that has not been exported yet still needs its last reading as a subtraction base. That is bounded by one export cycle, so it shifts when a series is dropped rather than whether.

Must be at least delay plus interval, and is raised to that when a smaller value is configured: a window is not exported until delay after it closes, so a series declared dead sooner could never contribute to one.

That minimum is necessary but not sufficient. This must also be longer than the reporting period of the sources the rule matches, or each reading is dropped before the next one arrives and the aggregated series exports nothing at all. Only the rule's own widths can be validated, so this one is on you: for a source scraped every 5 minutes, a staleAfter of a few seconds over delay plus interval is legal and useless.

Default is 5 times the interval, subject to that same minimum.

Used by 1

These schemas and endpoints reference this type directly.

SchemaTimeSeriesAggregationSpec

json

OpenAPI source1234567891011121314151617181920
{
  "properties": {
    "interval": {
      "$ref": "#/components/schemas/Duration",
      "description": "The frequency at which metric data points are aggregated via these rules. This is a required parameter.\nIf you don't which to downsample your metrics, then set this to the scrape/export interval of your\nmetrics. Otherwise, use a larger number to reduce the granularity of your metrics.\n"
    },
    "delay": {
      "$ref": "#/components/schemas/Duration",
      "description": "The amount of time to account for delays in scraping, exporting, and data arrival. We suggest to always\nadd a little bit of a delay to account for this. But you can increase it in case you have larger delays in\nyour scrape and export processes.\n",
      "default": "20s"
    },
    "staleAfter": {
      "$ref": "#/components/schemas/Duration",
      "description": "Duration after which a source time series with no new data is dropped from memory. This is a\nretention bound, and not a guarantee about exported values.\n\nWhen a source time series is dropped and new data arrives later, that source series starts\nover: its first new reading becomes a subtraction base rather than an increment, so no\nincrement is derived across the gap. The aggregated output series itself continues, and its\ncumulative total is not reset, because resetting it would be read downstream as a counter\nreset.\n\nA source time series may be held slightly beyond this duration while an aggregation window\nthat has not been exported yet still needs its last reading as a subtraction base. That is\nbounded by one export cycle, so it shifts when a series is dropped rather than whether.\n\nMust be at least delay plus interval, and is raised to that when a smaller value is\nconfigured: a window is not exported until delay after it closes, so a series declared dead\nsooner could never contribute to one.\n\nThat minimum is necessary but not sufficient. This must also be longer than the reporting\nperiod of the sources the rule matches, or each reading is dropped before the next one\narrives and the aggregated series exports nothing at all. Only the rule's own widths can be\nvalidated, so this one is on you: for a source scraped every 5 minutes, a staleAfter of a\nfew seconds over delay plus interval is legal and useless.\n\nDefault is 5 times the interval, subject to that same minimum.\n"
    }
  },
  "required": [
    "interval"
  ]
}