github-mirrors/VictoriaMetrics
1
0
Fork 0
mirror of https://github.com/VictoriaMetrics/VictoriaMetrics.git synced 2024-11-21 14:44:00 +00:00
Code Issues Projects Releases Packages Wiki Activity

app/vmselect: add -search.estimatedSeriesCountAfterAggregation command-line flag for tuning the probability of OOMs vs false-positive not enough memory errors

Browse source
This commit is contained in:
Aliaksandr Valialkin 2020-04-28 12:51:36 +03:00
parent 83aca79137
commit 17d96e4503
1 changed files with 7 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
app/vmselect/promql/eval.go
View file

@ -18,6 +18,9 @@ import (
var (
maxPointsPerTimeseries = flag.Int("search.maxPointsPerTimeseries", 30e3, "The maximum points per a single timeseries returned from the search")
estimatedSeriesCountAfterAggregation = flag.Int("search.estimatedSeriesCountAfterAggregation", 1000, "Estimated number of series returned by aggregation with grouping "+
"such as `sum(...) by (...)`. Increase this value in order to reduce the probability of OOMs. Reduce this value in order to reduce 'not enough memory' errors "+
"for queries containing aggregation with grouping")
)
// The minimum number of points per timeseries for enabling time rounding.
@ -668,8 +671,7 @@ func evalRollupFuncWithMetricExpr(ec *EvalConfig, name string, rf rollupFunc,
if iafc.ae.Modifier.Op != "" {
// Increase the number of timeseries for non-empty group list: `aggr() by (something)`,
// since each group can have own set of time series in memory.
// Estimate the number of such groups is lower than 1000 :)
timeseriesLen *= 1000
timeseriesLen *= *estimatedSeriesCountAfterAggregation
}
}
rollupPoints := mulNoOverflow(pointsPerTimeseries, int64(timeseriesLen*len(rcs)))
@ -679,8 +681,8 @@ func evalRollupFuncWithMetricExpr(ec *EvalConfig, name string, rf rollupFunc,
rss.Cancel()
return nil, fmt.Errorf("not enough memory for processing %d data points across %d time series with %d points in each time series; "+
"possible solutions are: reducing the number of matching time series; switching to node with more RAM; "+
"increasing -memory.allowedPercent; increasing `step` query arg (%gs)",
rollupPoints, rssLen*len(rcs), pointsPerTimeseries, float64(ec.Step)/1e3)
"increasing -memory.allowedPercent; increasing `step` query arg (%gs); reducing -search.estimatedSeriesCountAfterAggregation",
rollupPoints, timeseriesLen*len(rcs), pointsPerTimeseries, float64(ec.Step)/1e3)
}
defer rml.Put(uint64(rollupMemorySize))