Aliaksandr Valialkin
9240bc36a3
app/vmselect/promql/aggr_incremental.go: eliminate unnecessary memory allocation in incrementalAggrFuncContext.updateTimeseries
2024-01-23 02:28:30 +02:00
Aliaksandr Valialkin
41a0fdaf39
app/vmselect/promql: optimize repeated SLI-like instant queries with lookbehind windows >= 1d
...
Repeated instant queries with long lookbehind windows, which contain one of the following rollup functions,
are optimized via partial result caching:
- sum_over_time()
- count_over_time()
- avg_over_time()
- increase()
- rate()
The basic idea of optimization is to calculate
rf(m[d] @ t)
as
rf(m[offset] @ t) + rf(m[d] @ (t-offset)) - rf(m[offset] @ (t-d))
where rf(m[d] @ (t-offset)) is cached query result, which was calculated previously
The offset may be in the range of up to 1 hour.
2023-10-31 19:25:23 +01:00
Nikolay
1f91f22b5f
app/vmselect: reduce lock contention for heavy aggregation requests ( #5119 )
...
reduce lock contention for heavy aggregation requests
previously lock contetion may happen on machine with big number of CPU due to enabled string interning. sync.Map was a choke point for all aggregation requests.
Now instead of interning, new string is created. It may increase CPU and memory usage for some cases.
https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5087
2023-10-10 13:45:20 +02:00
Aliaksandr Valialkin
4856a4cf5a
app/vmselect: optimize incremental aggregates a bit
...
Substitute sync.Map with an ordinary slice indexed by workerID.
This should reduce the overhead when updating the incremental aggregate state
2023-03-20 15:37:06 -07:00
Aliaksandr Valialkin
21ee9a1fab
app/vmselect/promql: intern output series names during incremental aggregation
...
This should reduce the number of memory allocations for repeated queries
2023-01-09 22:11:36 -08:00
Aliaksandr Valialkin
fb1cc3cc94
app/vmselect/promql: increase scalability of incremental aggregate calculations on systems with many CPU cores
...
Use sync.Map instead of a global mutex there. This should lift scalability limits
on systems with many CPU cores.
2022-10-01 20:00:03 +03:00
Aliaksandr Valialkin
0da202023b
app/vmselect/promql: return empty values from group() if all the time series have no values at the given timestamp
...
This aligns `group()` behaviour to Prometheus
2020-07-28 13:40:11 +03:00
Aliaksandr Valialkin
6b5ad535ae
app/vmselect/promql: optimize group(rollup(m)) calculations
2020-07-17 16:47:16 +03:00
Aliaksandr Valialkin
7882a0dbbf
app/vmselect/promql: suppress "SA4006: this value of dstValues is never used" error in golangci-lint
2020-05-13 11:47:08 +03:00
Aliaksandr Valialkin
faf92a0965
app/vmselect/promql: fix any(..) calculations - return all the data points instead of the first one
2020-05-12 20:36:42 +03:00
Aliaksandr Valialkin
cc311e20fe
app/vmselect/promql: add any(x) by (y) aggregate function, which returns any time series from q for each group y
2020-05-12 19:45:56 +03:00
Aliaksandr Valialkin
574289c3fb
app/vmselect/promql: support for sum(x) by (y) limit N syntax in order to limit the number of output time series after aggregation
2020-05-12 19:45:54 +03:00
Aliaksandr Valialkin
4e4f57b121
lib/metricsql: move it to a separate repository - github.com/VictoriaMetrics/metrics
2020-04-28 15:28:22 +03:00
Aliaksandr Valialkin
1925ee038d
Rename lib/promql to lib/metricsql and apply small fixes
2019-12-25 22:03:59 +02:00
Mike Poindexter
bec62e4e43
Split Extended PromQL parsing to a separate library
2019-12-25 22:03:51 +02:00
Aliaksandr Valialkin
17096b5750
app/vmselect/promql: return NaN from count() over zero time series
...
This aligns `count` behavior with Prometheus.
2019-07-25 22:02:30 +03:00
Aliaksandr Valialkin
ab88890523
app/vmselect/promql: parallelize incremental aggregation to multiple CPU cores
...
This may reduce response times for aggregation over big number of time series
with small step between output data points.
2019-07-12 15:52:22 +03:00
Aliaksandr Valialkin
a336bb4e22
app/vmselect/promql: reduce RAM usage for aggregates over big number of time series
...
Calculate incremental aggregates for `aggr(metric_selector)` function instead of
keeping all the time series matching the given `metric_selector` in memory.
2019-07-10 13:04:39 +03:00
