This reverts commit eb40395a1c.
Reason for revert: it has been appeared that the performance gain on multiple CPU cores
wasn't visible because the benchmark was generating incorrect pushSample.key.
See a207e0bf687d65f5198207477248d70c69284296
Previously samples were dropped on the first incomplete interval and the next complete interval.
Also make sure that the de-duplication is performed just before flushing the aggregate state.
This should help the case then dedup_interval = interval.
For example, if `interval: 1m`, then data flush occurs at the end of every minute,
while `interval: 1h` leads to data flush at the end of every hour.
Add `no_align_flush_to_interval` option, which can be used for disabling the alignment.
The labelsMap struct employs the fact that label indexes are condensed around 0,
so it stores the referred labels in a slice instead of map and uses slice index as label key.
This allows increasing the LabelsCompressor.Decompress performance by up to 3x.
This also reduces the latency of data flush in stream aggregation.
- Reduce memory usage by up to 5x when de-duplicating samples across big number of time series.
- Reduce memory usage by up to 5x when aggregating across big number of output time series.
- Add lib/promutils.LabelsCompressor, which is going to be used by other VictoriaMetrics components
for reducing memory usage for marshaled []prompbmarshal.Label.
- Add `dedup_interval` option at aggregation config, which allows setting individual
deduplication intervals per each aggregation.
- Add `keep_metric_names` option at aggregation config, which allows keeping the original
metric names in the output samples.
- Add `unique_samples` output, which counts the number of unique sample values.
- Add `increase_prometheus` and `total_prometheus` outputs, which ignore the first sample
per each newly encountered time series.
- Use 64-bit hashes instead of marshaled labels as map keys when calculating `count_series` output.
This makes obsolete https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5579
- Expose various metrics, which may help debugging stream aggregation:
- vm_streamaggr_dedup_state_size_bytes - the size of data structures responsible for deduplication
- vm_streamaggr_dedup_state_items_count - the number of items in the deduplication data structures
- vm_streamaggr_labels_compressor_size_bytes - the size of labels compressor data structures
- vm_streamaggr_labels_compressor_items_count - the number of entries in the labels compressor
- vm_streamaggr_flush_duration_seconds - a histogram, which shows the duration of stream aggregation flushes
- vm_streamaggr_dedup_flush_duration_seconds - a histogram, which shows the duration of deduplication flushes
- vm_streamaggr_flush_timeouts_total - counter for timed out stream aggregation flushes,
which took longer than the configured interval
- vm_streamaggr_dedup_flush_timeouts_total - counter for timed out deduplication flushes,
which took longer than the configured dedup_interval
- Actualize docs/stream-aggregation.md
The memory usage reduction increases CPU usage during stream aggregation by up to 30%.
This commit is based on https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5850
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5898
- Document the ability to read OpenTelemetry data from Amazon Firehose at docs/CHANGELOG.md
- Simplify parsing Firehose data. There is no need in trying to optimize the parsing with fastjson
and byte slice tricks, since OpenTelemetry protocol is really slooow because of over-engineering.
It is better to write clear code for better maintanability in the future.
- Move Firehose parser from /lib/protoparser/firehose to lib/protoparser/opentelemetry/firehose,
since it is used only by opentelemetry parser.
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5893
This solves two issues:
1. The vm_backups_uploaded_bytes_total metric will grow more smoothly
2. This prevents from int overflow at metrics.Counter.Add() when uploading files bigger than 2GiB
This should simplify code maintenance by gradually converting to atomic.* types instead of calling atomic.* functions
on int and bool types.
See ea9e2b19a5
The issue has been introduced in bace9a2501
The improper fix was in the d4c0615dcd ,
since it fixed the issue just by an accident, because Go comiler aligned the rawRowsShards field
by 4-byte boundary inside partition struct.
The proper fix is to use atomic.Int64 field - this guarantees that the access to this field
won't result in unaligned 64-bit atomic operation. See https://github.com/golang/go/issues/50860
and https://github.com/golang/go/issues/19057
It has been appeared that there are VictoriaMetrics users, who rely on the fact that
VictoriaMetrics components were closing incoming connections to -httpListenAddr every 2 minutes
by default. So let's return back this value by default in order to fix the breaking change
made at d8c1db7953 .
See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1304#issuecomment-1961891450 .
Previously the (date, metricID) entries for dates older than the last 2 days were removed.
This could lead to slow check for the (date, metricID) entry in the indexdb during ingesting historical data (aka backfilling).
The issue has been introduced in 431aa16c8d
This commit returns back limits for these endpoints, which have been removed at 5d66ee88bd ,
since it has been appeared that missing limits result in high CPU usage, while the introduced concurrency limiter
results in failed lightweight requests to these endpoints because of timeout when heavyweight requests are executed.
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5055
Do not convert shard items to part when a shard becomes full. Instead, collect multiple
full shards and then convert them to a searchable part at once. This reduces
the number of searchable parts, which, in turn, should increase query performance,
since queries need to scan smaller number of parts.
* app/vmselect: adds milliseconds to the csv export response for rfc3339
* milliseconds is a standard prescion for VictoriaMetrics query request responses
https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5837
* app/victoria-metrics: adds tests for csv export/import
follow-up after 3541a8d0cf96dd4f8563624c4aab6816615d0756
---------
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Co-authored-by: hagen1778 <roman@victoriametrics.com>
Previously the interval between item addition and its conversion to searchable in-memory part
could vary significantly because of too coarse per-second precision. Switch from fasttime.UnixTimestamp()
to time.Now().UnixMilli() for millisecond precision. It is OK to use time.Now() for tracking
the time when buffered items must be converted to searchable in-memory parts, since time.Now()
calls aren't located in hot paths.
Increase the flush interval for converting buffered samples to searchable in-memory parts
from one second to two seconds. This should reduce the number of blocks, which are needed
to be processed during high-frequency alerting queries. This, in turn, should reduce CPU usage.
While at it, hardcode the maximum size of rawRows shard to 8Mb, since this size gives the optimal
data ingestion pefromance according to load tests. This reduces memory usage and CPU usage on systems
with big amounts of RAM under high data ingestion rate.
The pooled rawRowsBlock objects occupies big amounts of memory between flushes,
and the flushes are relatively rare. So it is better to don't use the pool
and to allocate rawRow blocks on demand. This should reduce the average
memory usage between flushes.
The buffer can be quite big under high ingestion rate (e.g. more than 100MB).
This leads to increased memory usage between buffer flushes.
So it is better to re-create the buffer on every flush in order to reduce memory usage
between buffer flushes.
* [lib/promutils, lib/httputils] fixed floating-point error when parsing time in RFC3339 format (#5801)
* fixed tests
* fixed test
* Revert "fixed test"
This reverts commit 8a29764806.
* Revert "fixed tests"
This reverts commit 9ce13d1042.
* Revert "[lib/promutils, lib/httputils] fixed floating-point error when parsing time in RFC3339 format (#5801)"
This reverts commit a7a04bd4
* [lib/httputils] fixed floating-point error when parsing time in RFC3339 format (#5801)
---------
Co-authored-by: Nikolay <nik@victoriametrics.com>
- Consistently return the first `limit` log entries if the total size of found log entries doesn't exceed 1Mb.
See app/vlselect/logsql/sort_writer.go . Previously random log entries could be returned with each request.
- Document the change at docs/VictoriaLogs/CHANGELOG.md
- Document the `limit` query arg at docs/VictoriaLogs/querying/README.md
- Make the change less intrusive.
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5674
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5778
Instead, log a sample of these long items once per 5 seconds into error log,
so users could notice and fix the issue with too long labels or too many labels.
Previously this panic could occur in production when ingesting samples with too long labels.
The 3c246cdf00 added an optimization where the previous metaindexRow
could be saved to disk when the current block header couldn't be added indexBlock because the resulting
indexBlock size became too big. This could result in an empty metaindexRow.firstItem for the next metaindexRow.
This allows removing importing unneeded command-line flags into binaries, which import lib/storage,
which, in turn, was importing lib/snapshot in order to use Time, Validate and NewName functions.
This is a follow-up for 83e55456e2
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5738
Closing client connections every 2 minutes doesn't help load balancing -
this just leads to "jumpy" connections between multiple backend servers,
e.g. the load isn't spread evenly among backend servers, and instead jumps
between the servers every 2 minutes.
It is still possible periodically closing client connections by specifying non-zero -http.connTimeout command-line flag.
This should help with https://github.com/VictoriaMetrics/VictoriaMetrics/issues/1304#issuecomment-1636997037
This is a follow-up for d387da142e
There is no sense in storing commonPrefix for blockHeader containing only a single item,
since this only increases blockHeader size without any benefits.
This panic could occur when samples with too long label values are ingested into VictoriaMetrics.
This could result in too long fistItem and commonPrefix values at blockHeader (up to 64kb each).
This may inflate the maximum index block size by 4 * maxIndexBlockSize.
For example, -fooDuration=',10s,' is now supported - it sets three command-line flag values:
- the first and the last one are set to the default value for `-fooDuration`
- the second one is set to 10s
This should significantly reduce the number of open ReaderAt files
on VictoriaMetrics and VictoriaLogs startup.
The open files can be tracked via vm_fs_readers metric
GOGC can be already set via environment variable. There is no need in adding
new approaches for setting the GOGC (such as command-line flag), since they complicate operations.
Remove temporary file before closing it in order to signal the OS that it shouldn't
store the file contents from page cache to disk when the file is closed.
Gracefully handle the case when the file cannot be removed before being closed -
in this case remove the file after closing it. This allows working on Windows.
Also remove superflouos opening of temporary file for reading - re-use already opened file handle for writing.
This is a follow-up for 9b1e002287
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/4020
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/70
The easyproto-based marshaler is 2x slower than the previous custom marshaler,
so let's stick with it. This improves the performance for sending data to remote storage at vmagent
and reduces CPU usage to pre-v1.97.0 levels.
* adding support for username_file in basic_auth of scrape config
Signed-off-by: Syed Nihal <syed.nihal@nokia.com>
* adding support for username_file in basic_auth of scrape config. See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5511
Signed-off-by: Syed Nihal <syed.nihal@nokia.com>
* adding support for username_file in basic_auth of scrape config
Signed-off-by: Syed Nihal <syed.nihal@nokia.com>
* adding support for username_file in basic_auth of scrape config
Signed-off-by: Syed Nihal <syed.nihal@nokia.com>
* adding support for username_file in basic_auth of scrape config
Signed-off-by: Syed Nihal <syed.nihal@nokia.com>
---------
Signed-off-by: Syed Nihal <syed.nihal@nokia.com>
Entries for the previous dates is usually not used, so there is little sense in keeping them in memory.
This should reduce the size of storage/date_metricID cache, which can be monitored
via vm_cache_entries{type="storage/date_metricID"} metric.
This limit has little sense for these APIs, since:
- Thses APIs frequently result in scanning of all the time series on the given time range.
For example, if extra_filters={datacenter="some_dc"} .
- Users expect these APIs shouldn't hit the -search.maxUniqueTimeseries limit,
which is intended for limiting resource usage at /api/v1/query and /api/v1/query_range requests.
Also limit the concurrency for /api/v1/labels, /api/v1/label/.../values
and /api/v1/series requests in order to limit the maximum memory usage and CPU usage for these API.
This limit shouldn't affect typical use cases for these APIs:
- Grafana dashboard load when dashboard labels should be loaded
- Auto-suggestion list load when editing the query in Grafana or vmui
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5055
* Optimize the performance of data merge: decimal.CalibrateScale() from 49633 ns/op to 9146 ns/op
* Optimize the performance of data merge: decimal.CalibrateScale()
Sending unfinished aggregate states tend to produce unexpected anomalies with lower values than expected.
The old behavior can be restored by specifying `flush_on_shutdown: true` setting in streaming aggregation config
Signed-off-by: hagen1778 <roman@victoriametrics.com>
- Maintain a separate worker pool per each part type (in-memory, file, big and small).
Previously a shared pool was used for merging all the part types.
A single merge worker could merge parts with mixed types at once. For example,
it could merge simultaneously an in-memory part plus a big file part.
Such a merge could take hours for big file part. During the duration of this merge
the in-memory part was pinned in memory and couldn't be persisted to disk
under the configured -inmemoryDataFlushInterval .
Another common issue, which could happen when parts with mixed types are merged,
is uncontrolled growth of in-memory parts or small parts when all the merge workers
were busy with merging big files. Such growth could lead to significant performance
degradataion for queries, since every query needs to check ever growing list of parts.
This could also slow down the registration of new time series, since VictoriaMetrics
searches for the internal series_id in the indexdb for every new time series.
The third issue is graceful shutdown duration, which could be very long when a background
merge is running on in-memory parts plus big file parts. This merge couldn't be interrupted,
since it merges in-memory parts.
A separate pool of merge workers per every part type elegantly resolves both issues:
- In-memory parts are merged to file-based parts in a timely manner, since the maximum
size of in-memory parts is limited.
- Long-running merges for big parts do not block merges for in-memory parts and small parts.
- Graceful shutdown duration is now limited by the time needed for flushing in-memory parts to files.
Merging for file parts is instantly canceled on graceful shutdown now.
- Deprecate -smallMergeConcurrency command-line flag, since the new background merge algorithm
should automatically self-tune according to the number of available CPU cores.
- Deprecate -finalMergeDelay command-line flag, since it wasn't working correctly.
It is better to run forced merge when needed - https://docs.victoriametrics.com/#forced-merge
- Tune the number of shards for pending rows and items before the data goes to in-memory parts
and becomes visible for search. This improves the maximum data ingestion rate and the maximum rate
for registration of new time series. This should reduce the duration of data ingestion slowdown
in VictoriaMetrics cluster on e.g. re-routing events, when some of vmstorage nodes become temporarily
unavailable.
- Prevent from possible "sync: WaitGroup misuse" panic on graceful shutdown.
This is a follow-up for fa566c68a6 .
Thanks @misutoth to for the inspiration at https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5212
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5190
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3790
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3551
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3337
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3425
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3647
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3641
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/648
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/291
The maxFileParts usage has been accidentally removed in fa566c68a6
While at it, add Count suffix to *AssistedMerges counter names in order to make them less misleading.
Previously their names were falsely suggesting that these are gauges, which show the number of concurrently
executed assisted merges.
It has been appeared that the registration of new time series slows down linearly
with the number of indexdb parts, since VictoriaMetrics needs to check every indexdb part
when it searches for TSID by newly ingested metric name.
The number of in-memory parts grows when new time series are registered
at high rate. The number of in-memory parts grows faster on systems with big number
of CPU cores, because the mergeset maintains per-CPU buffers with newly added entries
for the indexdb, and every such entry is transformed eventually into a separate in-memory part.
The solution has been suggested in https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5212
by @misutoth - to limit the number of in-memory parts with buffered channel.
This solution is implemented in this commit. Additionally, this commit merges per-CPU parts
into a single part before adding it to the list of in-memory parts. This reduces CPU load
when searching for TSID by newly ingested metric name.
The https://github.com/VictoriaMetrics/VictoriaMetrics/pull/5212 recommends setting the limit on the number
of in-memory parts to 100, but my internal testing shows that much lower limit 15 works with the same efficiency
on a system with 16 CPU cores while reducing memory usage for `indexdb/dataBlocks` cache by up to 50%.
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5190
This allows reducing the indexdb/tagFiltersToMetricIDs cache size by 8 on average.
The cache size can be checked via vm_cache_size_bytes{type="indexdb/tagFiltersToMetricIDs"} metric exposed at /metrics page.
Measuring read / write duration from / to in-memory buffers has little sense,
since it will be always fast. It is better to measure read / write duration from / to
real files at vm_filestream_write_duration_seconds_total and vm_filestream_read_duration_seconds_total metrics.
This also reduces overhead on time.Now() and Histogram.UpdateDuration() calls
per each filestream.Reader.Read() and filestream.Writer.Write() call when the data is read / written from / to in-memory buffers.
This is a follow-up for 2f63dec2e3
* lib/promscrape: respect `0` value for `series_limit` param
Respect `0` value for `series_limit` param in `scrape_config`
even if global limit was set via `-promscrape.seriesLimitPerTarget`.
Previously, `0` value will be ignored in favor of `-promscrape.seriesLimitPerTarget`.
This behavior aligns with possibility to override `series_limit` value via
relabeling with `__series_limit__` label.
Signed-off-by: hagen1778 <roman@victoriametrics.com>
* Update docs/CHANGELOG.md
---------
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Co-authored-by: Aliaksandr Valialkin <valyala@victoriametrics.com>
This should smooth CPU and RAM usage spikes related to these periodic tasks,
by reducing the probability that multiple concurrent periodic tasks are performed at the same time.
The dateMetricIDCache puts recently registered (date, metricID) entries into mutable cache protected by the mutex.
The dateMetricIDCache.Has() checks for the entry in the mutable cache when it isn't found in the immutable cache.
Access to the mutable cache is protected by the mutex. This means this access is slow on systems with many CPU cores.
The mutabe cache was merged into immutable cache every 10 seconds in order to avoid slow access to mutable cache.
This means that ingestion of new time series to VictoriaMetrics could result in significant slowdown for up to 10 seconds
because of bottleneck at the mutex.
Fix this by merging the mutable cache into immutable cache after len(cacheItems) / 2
cache hits under the mutex, e.g. when the entry is found in the mutable cache.
This should automatically adjust intervals between merges depending on the addition rate
for new time series (aka churn rate):
- The interval will be much smaller than 10 seconds under high churn rate.
This should reduce the mutex contention for mutable cache.
- The interval will be bigger than 10 seconds under low churn rate.
This should reduce the uneeded work on merging of mutable cache into immutable cache.
