Commit graph

127 commits

Author SHA1 Message Date
Aliaksandr Valialkin
f7d28bddbf Revert "app/vmselect/promql: remove metric name from results of certain rollup functions in order to be consistent with Prometheus"
This reverts commit e5202a4eae.

Reason for revert: the previous behavior for VictoriaMetrics is easier to understand and use by users -
functions, which don't change the meaning of the time series shouldn't drop metric name.

Now the following functions do not drop metric name:

* max_over_time
* min_over_time
* avg_over_time
* quantile_over_time
* geomean_over_time
* mode_over_time
* holt_winters
* predict_linear

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/674
2020-10-12 11:35:18 +03:00
Aliaksandr Valialkin
c584aece38 app/vmselect/promql: properly limit implicitly set rollup window to -search.maxStalenessInterval
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/784
2020-09-23 23:23:59 +03:00
Aliaksandr Valialkin
62919eaf7e app/vmselect/promql: go fmt 2020-09-08 15:19:59 +03:00
Aliaksandr Valialkin
e6da63dffe app/vmselect/promql: adjust integrate() calculations to be more similar to calculations from InfluxDB: attempt #2
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/701
2020-09-08 14:35:50 +03:00
Aliaksandr Valialkin
8e85b56737 app/vmselect/promql: adjust integrate() calculations to be more similar to calculations from InfluxDB
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/701
2020-09-08 14:23:39 +03:00
Aliaksandr Valialkin
c0343a661b app/vmselect/promql: increase floating point calculations accuracy by dividing by 1e3 instead of multiplying by 1e-3 2020-09-08 14:00:47 +03:00
Aliaksandr Valialkin
f4e7e5fb90 app/vmselect/promql: add count_le_over_time(m[d], le) and count_gt_over_time(m[d], gt) functions
These functions returns the number of raw samples that don't exceed `le` or are bigger than `gt`.
These functions are complement to already existing `share_le_over_time(m[d], le)` and `share_gt_over_time(m[d], gt)`.
2020-09-03 15:29:10 +03:00
Aliaksandr Valialkin
e5202a4eae app/vmselect/promql: remove metric name from results of certain rollup functions in order to be consistent with Prometheus
Rollup functions:

  - avg_over_time
  - min_over_time
  - max_over_time
  - quantile_over_time

This improves VictoriaMetrics results at https://promlabs.com/promql-compliance-test-results-victoriametrics/
2020-08-06 23:29:13 +03:00
Aliaksandr Valialkin
e6eee2bebf app/vmselect/promql: add zscore-related functions: zscore_over_time(m[d]) and zscore(q) by (...) 2020-08-03 21:52:18 +03:00
Aliaksandr Valialkin
328b52e5ff app/vmselect/promql: return non-empty value from rate_over_sum(m[d]) even if a single data point is located in the given [d] window
Just divide the data point value by the window duration in this case.
2020-07-29 12:37:58 +03:00
Aliaksandr Valialkin
700737c181 app/vmselect/promql: remove rollupFuncArg.realPrevValue handling, since the corner case in increase() is handled in another way now
See e00cfc854d for the approach used now.
2020-07-29 12:37:58 +03:00
Aliaksandr Valialkin
2f735f112d app/vmselect/promql: fill gaps with 0 in rate_over_sum response when the last value before the selected time window isnt empty 2020-07-29 12:37:58 +03:00
Aliaksandr Valialkin
3107c633e3 app/vmselect/promql: typo fix after 3e557c9861 2020-07-24 02:15:58 +03:00
Aliaksandr Valialkin
3e557c9861 app/vmselect/promql: add rate_over_sum(m[d]) function to MetricsQL, which returns rate over sum of m values over d duration
Something like `sum_over_time(m[d]) / d`, but more accurate.
2020-07-24 01:17:42 +03:00
Aliaksandr Valialkin
b1f6843bd0 app/vmselect/promql: allow setting [d] window smaller than the interval between raw points for avg_over_time
This makes `avg_over_time` behavior consistent with `sum_over_time` and `count_over_time` behaviors.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/636
2020-07-23 22:25:43 +03:00
Aliaksandr Valialkin
e00cfc854d app/vmselect/promql: skip the first value in time series passed to increase() if it exceeds by more than 10x the delta between the next value and the first value
This should prvent from inflated `increase()` results for time series that start from big initial values.
Such cases may occur when a label value changes in a metric without counter reset.
2020-07-21 17:24:10 +03:00
Aliaksandr Valialkin
ecb1b2564a app/vmselect/promql: add mode() aggregate function 2020-07-20 15:31:20 +03:00
Aliaksandr Valialkin
1c641037e8 app/vmselect/promql: add mode_over_time(m[d]) function
See https://en.wikipedia.org/wiki/Mode_(statistics) and https://stackoverflow.com/questions/61134078/promql-query-to-return-the-value-from-a-range-vector-which-occurs-maximum-no-of
2020-07-17 18:28:45 +03:00
Aliaksandr Valialkin
d5dddb0953 all: use %w instead of %s for wrapping errors in fmt.Errorf
This will simplify examining the returned errors such as httpserver.ErrorWithStatusCode .
See https://blog.golang.org/go1.13-errors for details.
2020-06-30 23:05:11 +03:00
Aliaksandr Valialkin
dbd8beccfa app/vmselect/promql: add ascent_over_time(m[d]) and descent_over_time(m[d]) functions
These functions could be useful in GPS tracking apps for calculating the summary for height gain/loss
over the given duration `d`.
2020-05-21 12:07:48 +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
9ebc937685 app/vmselect: add -search.minStalenessInterval command-line flag for removing gaps on graphs built from time series with irregular duration between samples
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/426
2020-04-20 19:42:15 +03:00
Aliaksandr Valialkin
fe57d46687 app/vmselect: merge -search.maxLookback and -search.maxStalenessInterval flags, since it has been appeared they have identical purpose :(
Leave both flags for backwards compatibility reasons.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/209
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/426
2020-04-20 19:26:31 +03:00
Aliaksandr Valialkin
6f7f64f757 app/vmselect: handle timestamp(metric offset X) the same way as Prometheus does
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/415
2020-04-15 12:01:00 +03:00
Aliaksandr Valialkin
f9b24d4899 app/vmselect/promql: keep metric name after applying first_over_time and last_over_time functions 2020-04-04 14:54:13 +03:00
Aliaksandr Valialkin
8023ad7dbd app/vmselect: add -search.maxStalenessInterval for tuning Prometheus data model closer to Influx-style data model 2020-03-11 16:43:34 +02:00
Aliaksandr Valialkin
ae51300973 app/vmselect/promql: properly take into account the first datapoint when calculating rollup_candlestick
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/309
2020-02-24 13:24:30 +02:00
Aliaksandr Valialkin
e65ec88779 app/vmselect/promql: do not take into account values outside the current window in rollup_candlestick
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/309
2020-02-23 18:03:57 +02:00
Yaroslav
a6d0645539
fix rollupOpen(), rollupHigh(), rollupLow() functions (#328) 2020-02-23 18:01:53 +02:00
Aliaksandr Valialkin
78b62dee87 app/vmselect/promql: properly adjust time range for data to select
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/309
2020-02-05 21:22:18 +02:00
Aliaksandr Valialkin
ada6a3da8d app/vmselect/promql: adjust rollup_candlestick calculations to the exepcted results
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/309
2020-02-04 22:42:13 +02:00
Aliaksandr Valialkin
adbbc4fa1a app/vmselect/promql: return expected results from increase() over the beginning of time series, which start from big value
Examples for such counters: OS-level counters for network or cpu stats.
2020-01-28 16:30:11 +02:00
Aliaksandr Valialkin
582dd01f42 app/vmselect/promql: add range_over_time(m[d]) function for calculating value range for m over d 2020-01-21 19:05:17 +02:00
Aliaksandr Valialkin
8b14572f70 app/vmselect/promql: add hoeffding_bound_upper(phi, m[d]) and hoeffding_bound_lower(phi, m[d]) functions
These functions can be used for calculating Hoeffding bounds
for `m` over `d` time range and for the given `phi` in the range `[0..1]`.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/283
2020-01-11 14:46:23 +02:00
Aliaksandr Valialkin
8eaced8cae app/vmselect/promql: return continuous values for min_over_time and max_over_time when step is smaller than scrape_interval 2020-01-11 12:47:50 +02:00
Aliaksandr Valialkin
46bfdbe6cf app/vmselect/promql: do not take into account the previous point before time window in square brackets for min_over_time, max_over_time, rollup_first and rollup_last functions
This makes the behaviour for these functions similar to Prometheus when processing broken time series with irregular data points
like `gitlab_runner_jobs`. See https://gitlab.com/gitlab-org/gitlab-exporter/issues/50 for details.
2020-01-11 00:26:26 +02:00
Aliaksandr Valialkin
b829fe5e39 app/vmselect/promql: properly handle aggr(aggr_over_time(...)) 2020-01-10 21:57:18 +02:00
Aliaksandr Valialkin
164278151f app/vmselect/promql: add aggr_over_time(("aggr_func1", "aggr_func2", ...), m[d]) function
This function can be used for simultaneous calculating of multiple `aggr_func*` functions
that accept range vector. For example, `aggr_over_time(("min_over_time", "max_over_time"), m[d])`
would calculate `min_over_time` and `max_over_time` for `m[d]`.
2020-01-10 21:18:06 +02:00
Aliaksandr Valialkin
c4632faa9d app/vmselect/promql: add tmin_over_time(m[d]) and tmax_over_time(m[d]) functions
These functions return timestamp in seconds for the minimum and maximum value for `m` over time range `d`
2020-01-10 19:39:28 +02:00
Aliaksandr Valialkin
d5aeda0e1a app/vmselect/promql: skip rate calculation for the first point on time series 2020-01-08 14:42:53 +02:00
Aliaksandr Valialkin
b1ded7cf9a app/vmselect/promql: add absent_over_time(m[d]) func similar to the function in Prometheus 2.16
See https://github.com/prometheus/prometheus/issues/2882
2020-01-04 12:45:07 +02:00
Aliaksandr Valialkin
a8360d04c0 app/vmselect/promql: add histogram_over_time(m[d]) rollup function 2020-01-04 12:44:56 +02:00
Aliaksandr Valialkin
a774120460 app/vmselect/promql: use scrapeInterval instead of window in denominator when calculating rate for the first point on the time series
This should provide better estimation for `rate` in the beginning of time series.
2020-01-03 19:01:50 +02:00
Aliaksandr Valialkin
cb3a342882 app/vmselect/promql: add share_le_over_time and share_gt_over_time functions for SLI and SLO calculations 2020-01-03 00:41:16 +02:00
Mike Poindexter
bec62e4e43 Split Extended PromQL parsing to a separate library 2019-12-25 22:03:51 +02:00
Aliaksandr Valialkin
9e19949c6b app/vmselect/promql: adjust calculations for rate and increase for the first value
These calculations should trigger alerts on `/api/v1/query` for counters starting from values greater than 0.
2019-12-24 19:39:25 +02:00
Aliaksandr Valialkin
0455c03cb9 app/vmselect/promql: properly calculate rate on the first data point
It is calculated as `value / scrape_interval`, since the value was missing on the previous scrape,
i.e. we can assume its value was 0 at this time.
2019-12-24 15:55:52 +02:00
Aliaksandr Valialkin
9476b73527 app/vmselect/promql: estimate per-series scrape interval as 0.6 quantile for the first 100 intervals
This should improve scrape interval estimation for tiem series with gaps.
2019-12-02 13:42:33 +02:00
Aliaksandr Valialkin
75ff524a4e app/vmselect/promql: fix corner case for increase over time series with gaps
In this case `increase` could return invalid high value for the first point after the gap.
2019-11-30 01:34:56 +02:00
Aliaksandr Valialkin
da98703748 app/vmselect/promql: optimize binary search over big number of samples during rollup calculations 2019-11-25 14:01:46 +02:00
Aliaksandr Valialkin
3f1637fae8 app/vmselect/promql: properly calculate integrate(q[d]) 2019-11-13 21:10:41 +02:00
Aliaksandr Valialkin
2f58f37f07 app/vmselect/promql: add lag(q[d]) function, which returns the lag between the current timestamp and the timstamp for the last data point in q 2019-11-01 12:21:33 +02:00
Aliaksandr Valialkin
c66a691593 app/vmselect/prometheus: add -search.maxLookback command-line flag for overriding dynamic calculations for max lookback interval
This flag is similar to `-search.lookback-delta` if set. The max lookback interval is determined dynamically
from interval between datapoints for each input time series if the flag isn't set.

The interval can be overriden on per-query basis by passing `max_lookback=<duration>` query arg to `/api/v1/query` and `/api/v1/query_range`.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/209
2019-10-15 21:31:48 +03:00
Aliaksandr Valialkin
956fdd89d3 app/vmselect/promql: take into account the previous point when calculating max_over_time and min_over_time
This lines up with `first_over_time` function used in `rollup_candlestick`, so `rollup=low` always returns
the minimum value.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/204
2019-10-08 12:30:05 +03:00
Aliaksandr Valialkin
26dc21cf64 app/vmselect/promql: add increases_over_time and decreases_over_time functions
`increases_over_time(q[d])` returns the number of `q` increases during the given duration `d`.
`decreases_over_time(q[d])` returns the number of `q` decreases during the given duration `d`.
2019-09-25 20:38:44 +03:00
Aliaksandr Valialkin
305e7bc981 app/vmselect/promql: do not return artificial points beyond the last point in time series 2019-09-04 16:35:34 +03:00
Aliaksandr Valialkin
dd86e6130c app/vmselect/promql: independently track offset hints for tStart and tEnd
This should improve performance if timeseries starts or ends on the selected time range
2019-08-19 13:40:14 +03:00
Aliaksandr Valialkin
6a27657d73 app/vmselect/promql: optimize search for timestamp boundaries in rollupConfig.Do
This should improve the performance of queries over big number of time series
with big number of output points.
2019-08-19 13:03:29 +03:00
Aliaksandr Valialkin
e74fb23189 app/vmselect/promql: add scrape_interval(q[d]) function, which would return scrape interval for q over d 2019-08-18 21:08:26 +03:00
Aliaksandr Valialkin
1c108fc494 app/vmselect/promql: add lifetime(q[d]) function, which returns the lifetime of q over d in seconds.
This function is useful for determining time series lifetime.
`d` must exceed the expected lifetime of the time series, otherwise
the function would return values close to `d`.
2019-08-16 11:59:32 +03:00
Aliaksandr Valialkin
d6b5ed6d39 app/vmselect/promql: fix corner-case calculation for ideriv 2019-08-16 11:59:28 +03:00
Aliaksandr Valialkin
639b14e8ab app/vmselect/promql: properly handle corner cases for rollup functions 2019-08-15 23:29:59 +03:00
Aliaksandr Valialkin
1c69f4eadc app/vmselect/promql: tune automatic window adjustement
Increase the windows adjustement for small scrape intervals,
since they usually have higher jitter.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/139
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/134
2019-08-04 19:34:05 +03:00
Aliaksandr Valialkin
8d93b15b86 app/vmselect/promql: further increase the allowed jitter for scrape interval
Real-world production data shows higher jitter than 1/8 of scrape interval.
This may results in gaps on the graph. So increase the allowed jitter to 1/4
of scrape interval in order to reduce the probability of gaps on the graphs
over time series with high jitter for scrape_interval.
2019-08-02 20:10:23 +03:00
Aliaksandr Valialkin
f090b2e917 app/vmselect/promql: tolerate higher jitter in scrape interval
Allow jitter for up to 1/8 instead of 1/16 for the scrape interval.
This should imrpove graphs when `step` is smaller than the `scrape_interval`.
2019-08-01 23:26:00 +03:00
Aliaksandr Valialkin
4aaa5c2efc app/vmselect/promql: allow adjusting window for default rollup function
Default rollup function is `last_over_time`. It must support adjusting
the provided window in order to prevent from gaps on the graph
for window values smaller than scrape interval.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/134
2019-07-31 00:45:54 +03:00
Aliaksandr Valialkin
e06866005d app/vmselect/promql: gracefully handle duplicate timestamps in irate and rollup_rate funcs
Previously such timestamps result in `+Inf` results. Now the previous timestamp is used
for the calculations.
2019-07-03 12:39:55 +03:00
Aliaksandr Valialkin
74c0fb04f3 app/vmselect/promql: consistency renaming: candlestick -> rollup_candlestick 2019-06-29 03:13:02 +03:00
Aliaksandr Valialkin
796b010139 app/vmselect: add candlestick(m[d]) func for returning open, close, low and high rollups on the given time range d
This function is frequently used in financial apps. See https://en.wikipedia.org/wiki/Candlestick_chart
2019-06-27 18:46:13 +03:00
Aliaksandr Valialkin
4bb738ddd9 app/vmselect/promql: adjust value returned by linearRegression to the end of time range like Prometheus does
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/71
2019-06-24 22:45:58 +03:00
Aliaksandr Valialkin
90e72c2a42 app/vmselect/promql: add sum2 and sum2_over_time, geomean and geomean_over_time funcs.
These functions may be useful for statistic calculations.
2019-06-24 16:44:44 +03:00
Aliaksandr Valialkin
61926bae01 app/vmselect/promql: adjust the provided window only for range functions with dt in denominator
This should fix range function calculations such as `changes(m[d])` where `d` is smaller
than the scrape interval.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/72
2019-06-23 19:27:31 +03:00
Aliaksandr Valialkin
146c6e1f72 app/vmselect/promql: typo fixes in comments 2019-06-21 23:22:59 +03:00
Aliaksandr Valialkin
d261fa2885 app/vmselect/promql: add deriv_fast function for calculating fast derivative
`deriv_fast` calculates derivative based on the first and the last point on the interval
instead of calculating linear regression based on all the data points on the interval.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/73
2019-06-21 23:05:39 +03:00
Aliaksandr Valialkin
5b47c00910 app/vmselect/promql: use linear regression in deriv func like Prometheus does
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/73
2019-06-21 22:59:46 +03:00
Aliaksandr Valialkin
9e1119dab8 app/vmselect/promql: ajdust data model to the model used in Prometheus
Do not take into account data points on the range `[timestamp .. timestamp+step)`
when calculating value on the given `timestamp`.
Use only data points from the past when performing these calculations like Prometheus does.

This should reduce discrepancies between results returned by VictoriaMetrics
and results returned by Prometheus.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/72
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/71
2019-06-21 21:54:48 +03:00
Aliaksandr Valialkin
1836c415e6 all: open-sourcing single-node version 2019-05-23 00:18:06 +03:00