Commit Graph

160 Commits

Author SHA1 Message Date
Aliaksandr Valialkin
e3adc095bd all: add -dedup.minScrapeInterval command-line flag for data de-duplication
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/86
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/278
2020-01-31 01:18:54 +02:00
Aliaksandr Valialkin
a45f25699c lib/storage: re-use indexSearch inside Storage.prefetchMetricNames 2020-01-31 01:18:53 +02:00
Aliaksandr Valialkin
da19fffa08 all: rename ReadAt* to MustReadAt* in order to dont clash with io.ReaderAt 2020-01-30 15:16:16 +02:00
Aliaksandr Valialkin
1332ddc15e lib/storage: pass missing AccountID and ProjectID to searchMetricName 2020-01-30 15:16:16 +02:00
Aliaksandr Valialkin
4ed5e9a7ce lib/storage: pre-fetch metricNames for the found metricIDs in Search.Init
This should speed up Search.NextMetricBlock loop for big number of found time series.
2020-01-30 15:16:16 +02:00
Aliaksandr Valialkin
ea53a21b02 all: consistently log durations in seconds with millisecond precision
This should improve logs readability
2020-01-22 18:35:24 +02:00
Aliaksandr Valialkin
62b041e90a lib/{mergeset,storage}: properly update lastAccessTime in index and data block cache entries 2020-01-20 15:00:10 +02:00
Aliaksandr Valialkin
a851c75703 lib/storage: skip recovering timestamps order for lossless compression (PrecisionBits=64) 2020-01-17 23:59:19 +02:00
Aliaksandr Valialkin
476c7fb109 lib/storage: reduce memory allocations when merging metricID sets 2020-01-17 22:10:56 +02:00
Aliaksandr Valialkin
7d429e2806 lib/uint64set: reduce memory usage in Union, Intersect and Subtract methods
Iterate items with newly added Set.ForEach method instead of allocating `[]uint64`
slice for all the items before the iteration.
2020-01-15 12:15:48 +02:00
Aliaksandr Valialkin
caffb0cd01 lib/{mergeset,storage}: fix uint64 counters alignment for 32-bit architectures (GOARCH=386, GOARCH=arm) 2020-01-14 22:47:42 +02:00
Aliaksandr Valialkin
b03ccbf6f7 lib/{storage,mergeset}: gradually remove stale entries from block cache and index caches
This should reduce memory usage in the long run when old blocks and indexes
aren't accessed anymore.
2020-01-14 21:38:29 +02:00
Aliaksandr Valialkin
53e176ed67 lib/storage: limit maxRaRowsPerPartition by 500K for any number of rawRowsShardsPerPartition
This should reduce write amplification for high ingestion rate on multi-CPU systems
2020-01-04 23:58:23 +02:00
Aliaksandr Valialkin
a37a006f11 lib/storage: scale ingestion performance by sharding rawRows on systems with more than 8 CPU cores 2019-12-19 18:17:05 +02:00
Aliaksandr Valialkin
8d79412b26 lib/storage: optimize bulk import performance when multiple data points are inserted for the same time series
This should speed up `/api/v1/import` and make it more scalable on multi-core systems.
2019-12-19 15:13:36 +02:00
Aliaksandr Valialkin
3694efd005 lib/{mergeset,storage}: log info message when both source and destination part paths from txn are missing during startup
This is expected condition after unclean shutdown (OOM, hard reset, `kill -9`) on NFS disk.
2019-12-09 15:45:23 +02:00
Aliaksandr Valialkin
639967db59 lib/{mergeset,storage}: make sure pending transaction deletions are finished before and after runTransactions call.
`runTransactions` call issues async deletions for transaction files. The previously issued transaction deletions
can race with the next call to `runTransactions`. Prevent this by waiting until all the pending transaction
deletions are funished in the beginning of `runTransactions`. Also make sure that all the pending transaction
deletions are finished before returning from `runTransactions`.
2019-12-04 21:40:52 +02:00
Aliaksandr Valialkin
534da0a8c3 lib/storage: fall back to global inverted index if a filter match too many time series in per-day index
Previously this resulted to error message. The query may succeed via search in global index.
2019-12-03 14:48:08 +02:00
Aliaksandr Valialkin
6eb698d1cc lib/storage: fix printing tag filters in TagFilters.String 2019-12-03 14:25:20 +02:00
Aliaksandr Valialkin
c04f60db35 lib/storage: print __name__ instead of empty string in user-visible tag filters 2019-12-03 14:18:18 +02:00
Aliaksandr Valialkin
625f6ca761 lib/storage: optimize regexp filter search 2019-12-03 00:33:53 +02:00
Aliaksandr Valialkin
b9616c017f lib/{mergeset,storage}: remove transaction files only after the mentioned dirs are really removed
This should fix the issue on NFS when incompletely removed dirs may be left
after unclean shutdown (OOM, kill -9, hard reset, etc.), while the corresponding transaction
files are already removed.

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/162
2019-12-02 21:34:37 +02:00
Aliaksandr Valialkin
4e22b521c2 lib/storage: remove metricID with missing metricID->metricName entry
The metricID->metricName entry can be missing in the indexdb after unclean shutdown
when only a part of entries for new time series is written into indexdb.

Recover from such a situation by removing the broken metricID. New metricID
will be automatically created for time series with the given metricName
when new data point will arive to it.
2019-12-02 20:52:13 +02:00
Aliaksandr Valialkin
5a62415bec lib/storage: protect from time drift during indexdb rotation
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/248
2019-12-02 14:43:11 +02:00
Aliaksandr Valialkin
f055dbefda lib/storage: generate more human-friendly result in TagFilters.String 2019-12-02 13:56:40 +02:00
Aliaksandr Valialkin
0f184affa7 app/vmselect/promql: optimize binary search over big number of samples during rollup calculations 2019-11-25 14:01:54 +02:00
Aliaksandr Valialkin
b9e53490b9 lib/storage: move non-matching tag filters to the top at matchTagFilters
This should reduce the amount of useless work needed for matching the next metricNames.
2019-11-21 21:40:36 +02:00
Aliaksandr Valialkin
33d9d63393 lib/storage: speed up time series search for queries with multiple filters
Use optimized specialized binary search for uint64 metricIDs instead of generic sort.Search.
2019-11-21 18:43:40 +02:00
Aliaksandr Valialkin
a02a57fbe9 lib/storage: verify the number of returned metricIDs in BenchmarkHeadPostingForMatchers 2019-11-20 15:40:03 +02:00
Aliaksandr Valialkin
6ca4b94511 lib/storage: increase the number of created time series in BenchmarkHeadPostingForMatchers in order to be on par with Promethues
The previous commit was accidentally creating 10x smaller number of time series than Prometheus
and this led to invalid benchmark results.

The updated benchmark results:

benchmark                                                          old ns/op      new ns/op     delta
BenchmarkHeadPostingForMatchers/n="1"                              272756688      6194893       -97.73%
BenchmarkHeadPostingForMatchers/n="1",j="foo"                      138132923      10781372      -92.19%
BenchmarkHeadPostingForMatchers/j="foo",n="1"                      134723762      10632834      -92.11%
BenchmarkHeadPostingForMatchers/n="1",j!="foo"                     195823953      10679975      -94.55%
BenchmarkHeadPostingForMatchers/i=~".*"                            7962582919     100118510     -98.74%
BenchmarkHeadPostingForMatchers/i=~".+"                            7589543864     154955671     -97.96%
BenchmarkHeadPostingForMatchers/i=~""                              1142371741     258003769     -77.42%
BenchmarkHeadPostingForMatchers/i!=""                              9964150263     159783895     -98.40%
BenchmarkHeadPostingForMatchers/n="1",i=~".*",j="foo"              216995884      10937895      -94.96%
BenchmarkHeadPostingForMatchers/n="1",i=~".*",i!="2",j="foo"       202541348      10990027      -94.57%
BenchmarkHeadPostingForMatchers/n="1",i!=""                        486285711      87004349      -82.11%
BenchmarkHeadPostingForMatchers/n="1",i!="",j="foo"                350776931      53342793      -84.79%
BenchmarkHeadPostingForMatchers/n="1",i=~".+",j="foo"              380888565      54256156      -85.76%
BenchmarkHeadPostingForMatchers/n="1",i=~"1.+",j="foo"             89500296       21823279      -75.62%
BenchmarkHeadPostingForMatchers/n="1",i=~".+",i!="2",j="foo"       379529654      46671359      -87.70%
BenchmarkHeadPostingForMatchers/n="1",i=~".+",i!~"2.*",j="foo"     424563825      53915842      -87.30%

VictoriaMetrics uses 1GB of RAM during the benchmark (vs 3.5GB of RAM for Prometheus)
2019-11-18 19:48:27 +02:00
Aliaksandr Valialkin
6f61fd367a lib/storage: add BenchmarkHeadPostingForMatchers similar to the benchmark from Prometheus
See the corresponding benchmark in Prometheus - 23c0299d85/tsdb/head_bench_test.go (L52)

The benchmark allows performing apples-to-apples comparison of time series search
in Prometheus and VictoriaMetrics. The following article - https://www.robustperception.io/evaluating-performance-and-correctness -
contains incorrect numbers for VictoriaMetrics, since there wasn't this benchmark yet. Fix it.

Benchmarks can be repeated with the following commands from Prometheus and VictoriaMetrics source code roots:

- Prometheus: GOMAXPROCS=1 go test ./tsdb/ -run=111 -bench=BenchmarkHeadPostingForMatchers
- VictoriaMetrics: GOMAXPROCS=1 go test ./lib/storage/ -run=111 -bench=BenchmarkHeadPostingForMatchers

Benchmark results:
benchmark                                                          old ns/op      new ns/op     delta
BenchmarkHeadPostingForMatchers/n="1"                              272756688      364977        -99.87%
BenchmarkHeadPostingForMatchers/n="1",j="foo"                      138132923      1181636       -99.14%
BenchmarkHeadPostingForMatchers/j="foo",n="1"                      134723762      1141578       -99.15%
BenchmarkHeadPostingForMatchers/n="1",j!="foo"                     195823953      1148056       -99.41%
BenchmarkHeadPostingForMatchers/i=~".*"                            7962582919     8716755       -99.89%
BenchmarkHeadPostingForMatchers/i=~".+"                            7589543864     12096587      -99.84%
BenchmarkHeadPostingForMatchers/i=~""                              1142371741     16164560      -98.59%
BenchmarkHeadPostingForMatchers/i!=""                              9964150263     12230021      -99.88%
BenchmarkHeadPostingForMatchers/n="1",i=~".*",j="foo"              216995884      1173476       -99.46%
BenchmarkHeadPostingForMatchers/n="1",i=~".*",i!="2",j="foo"       202541348      1299743       -99.36%
BenchmarkHeadPostingForMatchers/n="1",i!=""                        486285711      11555193      -97.62%
BenchmarkHeadPostingForMatchers/n="1",i!="",j="foo"                350776931      5607506       -98.40%
BenchmarkHeadPostingForMatchers/n="1",i=~".+",j="foo"              380888565      6380335       -98.32%
BenchmarkHeadPostingForMatchers/n="1",i=~"1.+",j="foo"             89500296       2078970       -97.68%
BenchmarkHeadPostingForMatchers/n="1",i=~".+",i!="2",j="foo"       379529654      6561368       -98.27%
BenchmarkHeadPostingForMatchers/n="1",i=~".+",i!~"2.*",j="foo"     424563825      6757132       -98.41%

The first column (old) is for Prometheus, the second column (new) is for VictoriaMetrics.

Prometheus was using 3.5GB of RAM during the benchmark, while VictoriaMetrics was using 400MB of RAM.
2019-11-18 18:47:02 +02:00
Aliaksandr Valialkin
d297b65089 lib/storage: add vm_cache_size_bytes{type="storage/hour_metric_ids"} metric 2019-11-13 20:26:05 +02:00
Aliaksandr Valialkin
494ad0fdb3 lib/storage: remove inmemory index for recent hour, since it uses too much memory
Production workload shows that the index requires ~4Kb of RAM per active time series.
This is too much for high number of active time series, so let's delete this index.

Now the queries should fall back to the index for the current day instead of the index
for the recent hour. The query performance for the current day index should be good enough
given the 100M rows/sec scan speed per CPU core.
2019-11-13 18:08:58 +02:00
Aliaksandr Valialkin
633dd81bb5 lib/storage: add -disableRecentHourIndex flag for disabling inmemory index for recent hour
This may be useful for saving RAM on high number of time series aka high cardinality
2019-11-13 15:10:12 +02:00
Aliaksandr Valialkin
f1620ba7c0 lib/storage: fix inmemory inverted index issues found in v1.29
Issues fixed:
- Slow startup times. Now the index is loaded from cache during start.
- High memory usage related to superflouos index copies every 10 seconds.
2019-11-13 13:35:38 +02:00
Mike Poindexter
955a592106 Add test for invalid caching of tsids (#232)
* Add test for invalid caching of tsids

* Clean up error handling
2019-11-12 15:52:46 +02:00
Oleg Kovalov
74ba42d111 fix misspelled words (#229) 2019-11-12 00:18:24 +02:00
Aliaksandr Valialkin
c48e39eea9 lib/storage: add tests for dateMetricIDCache 2019-11-11 13:21:05 +02:00
Aliaksandr Valialkin
6bdde0d6d4 lib/storage: eliminate data race when updating lastSyncTime in dateMetricIDCache.Has 2019-11-10 22:04:23 +02:00
Aliaksandr Valialkin
9ea2bd822e lib/storage: implement per-day inverted index 2019-11-10 00:20:32 +02:00
Aliaksandr Valialkin
dea2f3efed lib/storage: use specialized cache for (date, metricID) entries
This improves ingestion performance.
2019-11-09 23:09:18 +02:00
Aliaksandr Valialkin
9a43902bd8 lib/storage: remove unused code from getMetricIDsForTimeRange: it is expected that time range is always non-zero 2019-11-09 19:03:51 +02:00
Aliaksandr Valialkin
c16e17dede lib/storage: properly set time range when deleting time series 2019-11-09 18:50:02 +02:00
Aliaksandr Valialkin
8126007c15 lib/storage: obtain all the time series ids from (tag->metricIDs) rows instead of (metricID->TSID) rows, since this much faster 2019-11-09 18:04:26 +02:00
Aliaksandr Valialkin
50773348d3 lib/storage: small code prettifying 2019-11-09 14:01:24 +02:00
Aliaksandr Valialkin
0bc54c23ce lib/storage: inmemoryInvertedIndex prettifying 2019-11-09 14:01:24 +02:00
Aliaksandr Valialkin
46e67bb78c lib/storage: export vm_new_timeseries_created_total metric for determining time series churn rate 2019-11-08 19:58:21 +02:00
Aliaksandr Valialkin
0063c857f5 lib/storage: add inmemory inverted index for the last hour
It should improve performance for `last N hours` dashboards with update intervals smaller than 1 hour.
2019-11-08 19:37:46 +02:00
Aliaksandr Valialkin
89c03a5464 lib/storage: populate partition names from both small and big directories
Certain partition directories may be missing after restoring from backups
if they had no data. Re-create such directories on start.
2019-11-06 19:50:21 +02:00
Aliaksandr Valialkin
1c777e0245 lib/storage: substitute error message about unsorted items in the index block after metricIDs merge with counter
The origin of the error has been detected and documented in the code,
so it is enough to export a counter for such errors at `vm_index_blocks_with_metric_ids_incorrect_order_total`,
so it could be monitored and alerted on high error rates.

Export also the counter for processed index blocks with metricIDs - `vm_index_blocks_with_metric_ids_processed_total`,
so its' rate could be compared to `rate(vm_index_blocks_with_metric_ids_incorrect_order_total)`.
2019-11-06 14:32:41 +02:00