This allows verifying how the benchmark performance scales with the number of available CPU cores
and makes the results of the benchmark consistent with other BenchmarkScrapeWorkScrapeInternal* benchmarks.
Also reduce the amounts of memory allocations inside generateScrape() function in order to reduce
measurement noise during the BenchmarkScrapeWorkScrapeInternalStreamBigData run.
This is a follow-up after c05ffa906d
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/8515
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/8159
It is always better to run benchmarks in parallel on all the available CPU cores
in order to see how their performance scales with the number of CPU cores (GOMAXPROCS).
The commit also performs the following modifications:
- Removes the dependency of on the scrapeWork from getLabelsHash() function.
- Makes sure that the benchmark cannot be optimized out by the compiler, by introducing a dependency
on a global Sink variable. Previously the getLabelsHash() function call could be optimized out
by the compiler, since this call has no side effects, and the returned result is ignored.
- Reduces the amounts of memory allocations inside the BenchmarkScrapeWorkGetLabelsHash
when preparing the labels for the benchmark. This should reduce measurements' noise during the benchmark.
This is a follow-up for c05ffa906d
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/8515
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/8159
Previously the buffer was increased by 30% after it became 50% full.
For example, if more than 5MB of data is read into 10MB buffer, then its' size
was increased to 13MB, leading to 13MB-5MB = 8MB of waste.
This translates to 8MB/5MB = 160% waste in the worst case.
The updated algorithm increases the buffer by 30% after it becomes ~94% full.
This means that if more than 9.4MB of data is read into 10MB buffer,
then its' size is increased to 13MB, leading to 13MB-9.4MB = 3.6MB of waste.
This translates to 3.6MB / 9.4MB = ~38% waste in the worst case.
This should reduce memory usage when vmagent reads big responses from scrape targets.
While at it, properly append the data to buffer if it already has more than 4KiB of data.
Previously the data over 4KiB in the buffer was lost after ReadFrom call.
This is a follow-up for f28f496a9d
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/6761
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/6759
It is faster to read the whole data and then decompress it in one go for zstd and snappy encodings.
This reduces the number of potential read() syscalls and decompress CGO calls needed
for reading and decompressing the data.
### Describe Your Changes
`its'` -> `its`
### Checklist
The following checks are **mandatory**:
- [x] My change adheres [VictoriaMetrics contributing
guidelines](https://docs.victoriametrics.com/contributing/).
(cherry picked from commit 0a49d8c930)
- Move lib/httputil.Transport to lib/promauth.NewTLSTransport. Remove the first arg to this function (URL),
since it has zero relation to the created transport.
- Move lib/httputil.TLSConfig to lib/promauth.NewTLSConfig. Re-use the existing functionality
from lib/promauth.Config for creating TLS config. This enables the following features:
- Ability to load key, cert and CA files from http urls.
- Ability to change the key, cert and CA files without the need to restart the service.
It automatically re-loads the new files after they change.
- Avoid a data race when multiple goroutines access and update roundTripper.trBase inside roundTripper.getTransport().
The way to go is to make sure the roundTripper.trBase is updated only during roundTripper creation,
and then can be only read without updating.
- Use the http.DefaultTransport for http2 client connections at Kubernetes service discovery.
Previously golang.org/x/net/http2.Transport was used there. This had the following issues:
- An additional dependency on golang.org/x/net/http2.
- Missing initialization of Transport.DialContext with netutil.Dialer.DialContext for http2 client.
- Missing initialization of Transport.TLSHandshakeTimeout for http2 client.
- Introduction of the lib/promauth.Config.NewRoundTripperFromGetter() method, which is hard to use properly.
- Unnecessary complications of the lib/promauth.roundTripper, which led to the data race described above.
- Avoid a data race when multiple goroutines access and update tls config shared between multiple
net/http.Transport instances at the TLSClientConfig field. The way to go is to always make a copy of the tls config
before assigning it to the net/http.Transport.TLSClientConfig field.
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5971
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/7114
It is much better to panic instead of returning an error on programming error (aka BUG),
since this significantly increases chances that the bug will be noticed, reported and fixed ASAP.
The returned error can be ignored, even if it is logged, while panic is much harder to ignore.
The code must always panic instead of returning errors when any programming error (aka unexpected state) is detected.
This is a follow-up for the commit 9feee15493
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/6783
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/6771
Previously, performance of stream.Parse could be limited by mutex.Lock on callback function. It used shared writeContext. With complicated relabeling rules and any slowness at pushData function, it could significantly decrease parsed rows processing performance.
This commit removes locks and makes parsed rows processing lock-free in the same manner as `stream.Parse` processing implemented at push ingestion processing.
Implementation details:
- Removing global lock around stream.Parse callback.
- Using atomic operations for counters
- Creating write contexts per callback instead of sharing
- Improving series limit checking with sync.Once
- Optimizing labels hash calculation with buffer pooling
- Adding comprehensive tests for concurrency correctness
Benchmark performance:
```
# before
BenchmarkScrapeWorkScrapeInternalStreamBigData-10 13 81973945 ns/op 37.68 MB/s 18947868 B/op 197 allocs/op
# after
goos: darwin
goarch: arm64
pkg: github.com/VictoriaMetrics/VictoriaMetrics/lib/promscrape
cpu: Apple M1 Pro
BenchmarkScrapeWorkScrapeInternalStreamBigData-10 74 15761331 ns/op 195.98 MB/s 15487399 B/op 148 allocs/op
PASS
ok github.com/VictoriaMetrics/VictoriaMetrics/lib/promscrape 1.806s
```
Related issue:
https://github.com/VictoriaMetrics/VictoriaMetrics/issues/8159
---------
Signed-off-by: Maksim Kotlyar <kotlyar.maksim@gmail.com>
Co-authored-by: Roman Khavronenko <hagen1778@gmail.com>
These errors could be caused by intermittent network issues, especially
in case of using proxies when accessing S3 storage. Previously, such
error would abort backup/restore process and require manual intervention
to ensure backups consistency.
This commit adds automatic retries to handle this to improve backups
reliability and resilience to network issues.
### Describe Your Changes
during initial flush with deduplication and windows enabled lower
timestamps threshold is set to an upper bound of the next deduplication
interval, which leads to ignoring all samples on subsequent intervals
### Checklist
The following checks are **mandatory**:
- [ ] My change adheres [VictoriaMetrics contributing
guidelines](https://docs.victoriametrics.com/contributing/).
---------
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Co-authored-by: hagen1778 <roman@victoriametrics.com>
(cherry picked from commit 511517f491)
Previously, vmagent may incorrectly store partial scrape response
in case of scrapping error. It may happen if `sw.ReadData` call fetched
some chunked response and store it at buffer. And later context deadline
exceed error happened.
As a result, at the next scrape iteration this partial response could
be forwarded to the `sw.sendStaleSeries(lastScrape...)` function call
and lead to `Prometheus line` parsing error.
This commit properly set response body to the empty value in case of
scrapping error. It prevents storing partial scrape response body. And
it no longer sends partial staleness markers to the remote storage.
Related issue:
https://github.com/VictoriaMetrics/VictoriaMetrics/issues/8528
### Describe Your Changes
`prasedRelabelConfig` -> `parsedRelabelConfig`
### Checklist
The following checks are **mandatory**:
- [x] My change adheres [VictoriaMetrics contributing
guidelines](https://docs.victoriametrics.com/contributing/).
(cherry picked from commit 127d4f37b8)
### Describe Your Changes
Fix many spelling errors and some grammar, including misspellings in
filenames.
The change also fixes a typo in metric `vm_mmaped_files` to `vm_mmapped_files`.
While this is a breaking change, this metric isn't used in alerts or dashboards.
So it seems to have low impact on users.
The change also deprecates `cspell` as it is much heavier and less usable.
---------
Co-authored-by: Andrii Chubatiuk <achubatiuk@victoriametrics.com>
Co-authored-by: Andrii Chubatiuk <andrew.chubatiuk@gmail.com>
(cherry picked from commit 76d205feae)
Signed-off-by: hagen1778 <roman@victoriametrics.com>
This reduces the size of LogRows.streamTagCanonicals by 1/3 because of the eliminated `cap` field
in the slice header (reflect.SliceHeader) compared to the string header (reflect.StringHeader).
This commit adds lib/chunkedbuffer.Buffer - an in-memory chunked buffer
optimized for random access via MustReadAt() function.
It is better than bytesutil.ByteBuffer for storing large volumes of data,
since it stores the data in chunks of a fixed size (4KiB at the moment)
instead of using a contiguous memory region. This has the following benefits over bytesutil.ByteBuffer:
- reduced memory fragmentation
- reduced memory re-allocations when new data is written to the buffer
- reduced memory usage, since the allocated chunks can be re-used
by other Buffer instances after Buffer.Reset() call
Performance tests show up to 2x memory reduction for VictoriaLogs
when ingesting logs with big number of fields (aka wide events) under high speed.
Pre-allocate the needed slice of strings and then assign items to it by index
instead of appending them. This reduces the number of memory allocations
and improves performance a bit.
…an 64KB at Reset
This commit reverts
b58e2ab214
as it has negative impacts when ByteBuffer is used for workloads that
always exceed 64KiB size. This significantly slows down affected
components because:
* buffers aren't beign reused;
* growing new buffers to >64KiB is very slow.
https://github.com/VictoriaMetrics/VictoriaMetrics/issues/8501
### Describe Your Changes
Please provide a brief description of the changes you made. Be as
specific as possible to help others understand the purpose and impact of
your modifications.
### Checklist
The following checks are **mandatory**:
- [ ] My change adheres [VictoriaMetrics contributing
guidelines](https://docs.victoriametrics.com/contributing/).
Signed-off-by: hagen1778 <roman@victoriametrics.com>
- Properly decode protobuf-encoded Loki request if it has no Content-Encoding header.
Protobuf Loki message is snappy-encoded by default, so snappy decoding must be used
when Content-Encoding header is missing.
- Return back the previous signatures of parseJSONRequest and parseProtobufRequest functions.
This eliminates the churn in tests for these functions. This also fixes broken
benchmarks BenchmarkParseJSONRequest and BenchmarkParseProtobufRequest, which consume
the whole request body on the first iteration and do nothing on subsequent iterations.
- Put the CHANGELOG entries into correct places, since they were incorrectly put into already released
versions of VictoriaMetrics and VictoriaLogs.
- Add support for reading zstd-compressed data ingestion requests into the remaining protocols
at VictoriaLogs and VictoriaMetrics.
- Remove the `encoding` arg from PutUncompressedReader() - it has enough information about
the passed reader arg in order to properly deal with it.
- Add ReadUncompressedData to lib/protoparser/common for reading uncompressed data from the reader until EOF.
This allows removing repeated code across request-based protocol parsers without streaming mode.
- Consistently limit data ingestion request sizes, which can be read by ReadUncompressedData function.
Previously this wasn't the case for all the supported protocols.
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/pull/8416
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/8380
Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/8300
