Commit Graph

489 Commits

Author SHA1 Message Date
Aliaksandr Valialkin
c4638553a3
lib/fs: rename WriteFileAtomically to MustWriteAtomic
Callers of this function log the returned error and exit.
So let's just log the error with the given filepath and the call stack
inside the function itself and then exit. This simplifies the code
at callers' place while leaves the same level of debuggability in case of errors.
2023-04-13 22:43:30 -07:00
Aliaksandr Valialkin
54b9537a76
app/vmselect/promql: follow-up for 79e1c6a6fc
- Document the fix at docs/CHANGELOG.md
- Add tests with multiple adjancent zero buckets
- Simplify the fix a bit

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/296
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/4021
2023-03-27 18:04:30 -07:00
Ze'ev Klapow
680a661ec0
fix le buckets when adjacent vmrange is empty ()
There is a bug here where if you have a single bucket like:

foo{vmrange="4.084e+02...4.642e+02"} 2 123

The expected output is three le encoded buckets like:

foo{le="4.084e+02"} 0 123
foo{le="4.642e+02"} 2 123
foo{le="+Inf"} 2 123

This correctly encodes the start and end of the vmrange.
If however, the input contains the previous bucket, and that bucket is
empty then you only get the end le and +Inf out currently, i.e:

foo{vmrange="7.743e+05...8.799e+05"} 5 123
foo{vmrange="6.813e+05...7.743e+05"} 0 123

results in:

foo{le="8.799e+05"} 5 123
foo{le="+Inf"} 5 123

This causes issues when you go to compute a quantile because this means
that the assumed lower bound of the buckets is 0 and this we interpolate
between 0->end rather than the vmrange start->end as expected.
2023-03-27 18:04:29 -07:00
Aliaksandr Valialkin
9387793f47
app/vmselect: follow-up for 10ab086366
- Expose stats.seriesFetched at `/api/v1/query_range` responses too
  for the sake of consistency.

- Initialize QueryStats when it is needed and pass it to EvalConfig then.
  This guarantees that the QueryStats is properly collected when the query
  contains some subqueries.
2023-03-27 15:11:42 -07:00
Roman Khavronenko
10ab086366
app/vmselect: export seriesFetched stat for /query responses ()
The change adds a new field `seriesFetched` to EvalConfig object.
Since EvalConfig object can be copied inside `Exec`,
`seriesFetched` is a pointer which can be updated by all copied
objects.

The reason for having stats is that other components, like vmalert,
could benefit from this information.

Signed-off-by: hagen1778 <roman@victoriametrics.com>
Co-authored-by: Aliaksandr Valialkin <valyala@victoriametrics.com>
2023-03-27 08:51:33 -07:00
Aliaksandr Valialkin
740fa57fdc
app/vmselect/promql: typo fix after e7f46a0aab 2023-03-24 23:47:11 -07:00
Aliaksandr Valialkin
7aff6f872f
app/vmselect/promql: follow-up for 7205c79c5a
- Allocate and initialize seriesByWorkerID slice in a single go instead
  of initializing every item in the list separately.
  This should reduce CPU usage a bit.
- Properly set anti-false sharing padding at timeseriesWithPadding structure
- Document the change at docs/CHANGELOG.md

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3966
2023-03-24 23:39:43 -07:00
Zakhar Bessarab
fec87e3ada
app/vmselect/promql: use lock-less approach to gather results of parallel processing for evalRollup* funcs ()
* vmselect/promql: refactor `evalRollupNoIncrementalAggregate` to use lock-less approach for parallel workers computation

Locking there is causing issues when running on highly multi-core system as it introduces lock contention during results merge.

New implementation uses lock less approach to store results per workerID and merges final result in the end, this is expected to significantly reduce lock contention and CPU usage for systems with high number of cores.

Related: 
Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com>

* vmselect/promql: add pooling for `timeseriesWithPadding` to reduce allocations

Related: 
Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com>

* vmselect/promql: refactor `evalRollupFuncWithSubquery` to avoid using locks

Uses same approach as `evalRollupNoIncrementalAggregate` to remove locking between workers and reduce lock contention.

Related: 
Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com>

---------

Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com>
2023-03-24 23:39:41 -07:00
Aliaksandr Valialkin
79d8f0e7c6
app/vmselect/promql: pass workerID to the callback inside doParallel()
This opens the possibility to remove tssLock from evalRollupFuncWithSubquery()
in the follow-up commit from @zekker6 in order to speed up the code
for systems with many CPU cores.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3966
2023-03-20 20:57:34 -07:00
Aliaksandr Valialkin
e749a015a9
app/vmselect/promql: fix TestIncrementalAggr test on systems less than 3 CPU cores
This is a follow-up for 4856a4cf5a
2023-03-20 20:37:44 -07:00
Aliaksandr Valialkin
18af01c387
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 15:42:13 -07:00
oliverpool
8c708ca1e9
app/vmselect/promql: add test to ensure 8-byte alignment ()
See 0af9e2b693
2023-03-16 22:07:13 -07:00
Aliaksandr Valialkin
3b4a3583bc
app/vmselect/promql: prevent from cannot unmarshal timeseries from rollupResultCache panic after the upgrade to v1.89.0 2023-03-12 19:09:11 -07:00
Aliaksandr Valialkin
a6a4beb89a
app/vmselect: remove data race on updating EvalConfig.IsPartialResponse from concurrently running goroutines
This properly returns `is_partial: true` for partial responses.
2023-03-12 16:53:03 -07:00
Aliaksandr Valialkin
5cd60c54d3
app/vmselect/promql: prevent from SIGBUS crash on architecures, which deny unaligned access to 8-byte words (e.g. ARM)
Thanks to @oliverpool for nailing down the root cause of the issue and for the initial attempt to fix it
at https://github.com/VictoriaMetrics/VictoriaMetrics/pull/3927
2023-03-12 16:29:18 -07:00
Aliaksandr Valialkin
d3605ad072
app/vmselect/promql: fix panic when calculating aggr_func(rollup*())
The panic has been introduced in dac21d874b
2023-02-27 11:48:38 -08:00
Aliaksandr Valialkin
18dd0d1dbf
.golangci.yml: properly enable revive linter and fix all the warnings it detects 2023-02-26 12:19:58 -08:00
Roman Khavronenko
66d0b45651
vmselect/promql: check for deadline in count_values fn ()
* vmselect/promql: check for deadline in `count_values` fn

`count_values` could be very slow during the data processing.
Checking for deadline between iterations supposed to reduce
probability of exceeding `search.maxQueryDuration`.

The change also adds a new trace record, which captures the time
spent in aggregation function. Before that, the trace for aggr funcs
could be confusing since it doesn't account for all the places where
time was spent.

Signed-off-by: hagen1778 <roman@victoriametrics.com>

* wip

---------

Signed-off-by: hagen1778 <roman@victoriametrics.com>
Co-authored-by: Aliaksandr Valialkin <valyala@victoriametrics.com>
2023-02-24 17:10:38 -08:00
Roman Khavronenko
79eb33556e
metricsql: support optional 2nd argument for rollup functions ()
* metricsql: support optional 2nd argument for rollup functions

Support optional 2nd argument `min`, `max` or `avg` for rollup functions:
 * rollup
 * rollup_delta
 * rollup_deriv
 * rollup_increase
 * rollup_rate
 * rollup_scrape_interval

 If second argument is passed, then rollup function will return only the selected aggregation type.
 This change can be useful for situations where only one type of rollup calculation is needed.
 For example, `rollup_rate(requests_total[5m], "max")`.

Signed-off-by: hagen1778 <roman@victoriametrics.com>

* wip

---------

Signed-off-by: hagen1778 <roman@victoriametrics.com>
Co-authored-by: Aliaksandr Valialkin <valyala@victoriametrics.com>
2023-02-24 13:48:30 -08:00
Aliaksandr Valialkin
8efa9159cf
app/vmselect/promql: measure the time required for calculating the aggregate function from the prepared source time series 2023-02-23 20:06:02 -08:00
Aliaksandr Valialkin
6369c88a68
app/vmselect: add -search.logQueryMemoryUsage command-line flag for logging queries, which take big amounts of memory
Thanks to @michal-kralik for initial attempts for this feature:

- https://github.com/VictoriaMetrics/VictoriaMetrics/pull/3651
- https://github.com/VictoriaMetrics/VictoriaMetrics/pull/3715

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3553
2023-02-23 18:52:44 -08:00
Alexander Marshalov
9bb9bd266b
fix interpolate function for filling only intermediate gaps () ()
* fix interpolate function for filling only intermediate gaps ()

* Update docs/CHANGELOG.md

---------

Co-authored-by: Aliaksandr Valialkin <valyala@victoriametrics.com>
2023-02-22 18:41:29 -08:00
Aliaksandr Valialkin
a15da5ff73
app/vmselect/promql: add share(q) aggregate function for normalizing results across multiple time series in [0..1] value range per each timestamp and aggregation group 2023-02-18 22:43:54 -08:00
Aliaksandr Valialkin
84b5532bc1
app/vmselect/promql: add range_zscore(q) and range_trim_zscore(z, q) functions
These functions may be useful for dropping outliers at https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3759
2023-02-18 22:43:53 -08:00
Aliaksandr Valialkin
450b6f6d39
app/vmselect/promql: add range_mad(q) and range_trim_outliers(k, q) functions
These functions may help trimming outliers during query time
for the use case described at https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3759
2023-02-18 15:18:47 -08:00
Oleksandr Redko
0e1c395609
app,lib: fix typos in comments () 2023-02-13 09:32:35 -08:00
Aliaksandr Valialkin
db7f237da9
app/vmselect/promql: add mad_over_time(m[d]) function
See https://github.com/prometheus/prometheus/issues/5514
2023-02-11 01:06:39 -08:00
Aliaksandr Valialkin
6ff15ca135
app/vmselect: use consistent randomizer in tests
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3683
2023-01-23 19:27:40 -08:00
Aliaksandr Valialkin
e01f52d517
app/vmselect/promql: updates tests for https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3664 2023-01-17 23:26:06 -08:00
Aliaksandr Valialkin
ce47faf102
app/vmselect/promql: reduce memory allocations when searching for time series pairs with identical labelsets in q1 op q2 queries
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3641
2023-01-15 13:00:28 -08:00
Aliaksandr Valialkin
fe8802bbc8
app/vmselect/promql: reduce the number of memory allocations inside getCommonLabelFilters()
This should improve performance a bit for `q1 op q2` queries
2023-01-15 12:56:21 -08:00
Aliaksandr Valialkin
d33a65e401
app/vmselect/promql: reduce memory allocations at getCommonLabelFilters() function
Intern tag keys and values there
2023-01-12 01:27:34 -08:00
Aliaksandr Valialkin
8a35377cf3
app/vmselect/promql: move the eval function args in parallel query trace outside the loop 2023-01-10 22:23:43 -08:00
Aliaksandr Valialkin
c8bd3534cb
app/vmselect/promql: eliminate memory allocation when sorting values inside float64s 2023-01-09 23:06:57 -08:00
Aliaksandr Valialkin
7956b0d974
app/vmselect/promql: pre-allocate memory for values to be merged in mergeTimeseries()
This should reduce the number of memory re-allocations
2023-01-09 22:52:19 -08:00
Aliaksandr Valialkin
895d5d9d22
app/vmselect/promql: consistently intern series names obtained from marshalMetricNameSorted
This reduces memory allocations when the returned series names are used as map keys later
2023-01-09 22:46:30 -08:00
Aliaksandr Valialkin
12e2bcdf81
app/vmselect/promql: avoid memory allocations and copying from source timeseries to the returned result at timeseriesToResult() 2023-01-09 22:39:15 -08:00
Aliaksandr Valialkin
dd92e2050f
app/vmselect/promql: remove memory allocations from sortMetricTags() 2023-01-09 22:22:58 -08:00
Aliaksandr Valialkin
8050f5a18c
app/vmselect/promql: intern output series names inside timeseriesToResult()
This reduces the number of memory allocations for repeated queries,
which return (almost) the same set of time series.
2023-01-09 22:20:34 -08:00
Aliaksandr Valialkin
efb3c630fe
app/vmselect/promql: intern output series names during normal aggregation 2023-01-09 22:15:31 -08:00
Aliaksandr Valialkin
c0de651558
app/vmselect/promql: intern output series names during incremental aggregation
This should reduce the number of memory allocations for repeated queries
2023-01-09 22:12:05 -08:00
Aliaksandr Valialkin
78114e85d6
vendor: update github.com/VictoriaMetrics/metricsql from v0.50.0 to v0.51.0
Updates https://github.com/VictoriaMetrics/metricsql/pull/7
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3589
2023-01-05 01:50:26 -08:00
Aliaksandr Valialkin
507e719c3f
app/vmselect/promql: add tests for d3de110070 2022-12-21 20:25:37 -08:00
Aliaksandr Valialkin
a77bb09e69
app/vmselect/promql: make sure that label_replace() doesn't create an empty dst_label if the src_label doesn't match regex 2022-12-21 20:25:37 -08:00
Aliaksandr Valialkin
7fb314bb59
app/vmselect/promql: do not extend too short lookbehind window for rate() function if it is set explicitly
Previously too short lookbehind window d for rate(m[d]) could be automatically extended
if it didn't cover at least two raw samples. This was needed in order to guarantee
non-empty results from rate(m[d]) on short time ranges.

Now the lookbehind window isn't extended if it is set explicitly,
since it is expected that the user knows what he is doing.

The lookbehind window continues to be extended when needed if it isn't set explicitly.
For example, in the case of rate(m).

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3483
2022-12-20 00:18:32 -08:00
Aliaksandr Valialkin
808ce815e4
app/vmselect/promql: allow passing inf arg into functions, which accept numeric limit on the number of output time series
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3461
2022-12-10 22:48:01 -08:00
Aliaksandr Valialkin
c902500efc
app/vmselect/promql: add range_trim_spikes(phi, q) function for trimming phi percent of largest spikes per each time series returned by q 2022-12-05 21:55:24 -08:00
Aliaksandr Valialkin
ad548abd6a
app/vmselect/promql: add range_normalize(q1, ..., qN) function for normalizing query results into [0..1] value range
This may be useful for analyzing correlation between time series with different value ranges
2022-11-21 23:25:35 +02:00
Aliaksandr Valialkin
2d361a834d
app/vmselect/promql: properly return an empty result from limit_offset() if offset exceeds the number of inner time series
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3312
2022-11-21 16:48:10 +02:00
Aliaksandr Valialkin
b0fefe562a
app/vmselect/promql: optimize e1 op e2 when e1 returns an empty result
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3349
2022-11-21 16:09:49 +02:00