VictoriaMetrics/lib/logstorage/index_block_header.go

176 lines
5.9 KiB
Go
Raw Normal View History

package logstorage
import (
"fmt"
"io"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/encoding"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
)
// indexBlockHeader contains index information about multiple blocks.
//
// It allows locating the block by streamID and/or by time range.
type indexBlockHeader struct {
// streamID is the minimum streamID covered by the indexBlockHeader
streamID streamID
// minTimestamp is the mimumum timestamp seen across blocks covered by the indexBlockHeader
minTimestamp int64
// maxTimestamp is the maximum timestamp seen across blocks covered by the indexBlockHeader
maxTimestamp int64
// indexBlockOffset is an offset of the linked index block at indexFilename
indexBlockOffset uint64
// indexBlockSize is the size of the linked index block at indexFilename
indexBlockSize uint64
}
// reset resets ih for subsequent re-use.
func (ih *indexBlockHeader) reset() {
ih.streamID.reset()
ih.minTimestamp = 0
ih.maxTimestamp = 0
ih.indexBlockOffset = 0
ih.indexBlockSize = 0
}
// mustWriteIndexBlock writes data with the given additioanl args to sw and updates ih accordingly.
func (ih *indexBlockHeader) mustWriteIndexBlock(data []byte, sidFirst streamID, minTimestamp, maxTimestamp int64, sw *streamWriters) {
ih.streamID = sidFirst
ih.minTimestamp = minTimestamp
ih.maxTimestamp = maxTimestamp
bb := longTermBufPool.Get()
bb.B = encoding.CompressZSTDLevel(bb.B[:0], data, 1)
ih.indexBlockOffset = sw.indexWriter.bytesWritten
ih.indexBlockSize = uint64(len(bb.B))
sw.indexWriter.MustWrite(bb.B)
longTermBufPool.Put(bb)
}
// mustReadNextIndexBlock reads the next index block associated with ih from src, appends it to dst and returns the result.
func (ih *indexBlockHeader) mustReadNextIndexBlock(dst []byte, sr *streamReaders) []byte {
indexReader := &sr.indexReader
indexBlockSize := ih.indexBlockSize
if indexBlockSize > maxIndexBlockSize {
logger.Panicf("FATAL: %s: indexBlockHeader.indexBlockSize=%d cannot exceed %d bytes", indexReader.Path(), indexBlockSize, maxIndexBlockSize)
}
if ih.indexBlockOffset != indexReader.bytesRead {
logger.Panicf("FATAL: %s: indexBlockHeader.indexBlockOffset=%d must equal to %d", indexReader.Path(), ih.indexBlockOffset, indexReader.bytesRead)
}
bbCompressed := longTermBufPool.Get()
bbCompressed.B = bytesutil.ResizeNoCopyMayOverallocate(bbCompressed.B, int(indexBlockSize))
indexReader.MustReadFull(bbCompressed.B)
// Decompress bbCompressed to dst
var err error
dst, err = encoding.DecompressZSTD(dst, bbCompressed.B)
longTermBufPool.Put(bbCompressed)
if err != nil {
logger.Panicf("FATAL: %s: cannot decompress indexBlock read at offset %d with size %d: %s", indexReader.Path(), ih.indexBlockOffset, indexBlockSize, err)
}
return dst
}
// marshal appends marshaled ih to dst and returns the result.
func (ih *indexBlockHeader) marshal(dst []byte) []byte {
dst = ih.streamID.marshal(dst)
dst = encoding.MarshalUint64(dst, uint64(ih.minTimestamp))
dst = encoding.MarshalUint64(dst, uint64(ih.maxTimestamp))
dst = encoding.MarshalUint64(dst, ih.indexBlockOffset)
dst = encoding.MarshalUint64(dst, ih.indexBlockSize)
return dst
}
// unmarshal unmarshals ih from src and returns the tail left.
func (ih *indexBlockHeader) unmarshal(src []byte) ([]byte, error) {
srcOrig := src
// unmarshal ih.streamID
tail, err := ih.streamID.unmarshal(src)
if err != nil {
return srcOrig, fmt.Errorf("cannot unmarshal streamID: %w", err)
}
src = tail
// unmarshal the rest of indexBlockHeader fields
if len(src) < 32 {
return srcOrig, fmt.Errorf("cannot unmarshal indexBlockHeader from %d bytes; need at least 32 bytes", len(src))
}
ih.minTimestamp = int64(encoding.UnmarshalUint64(src))
ih.maxTimestamp = int64(encoding.UnmarshalUint64(src[8:]))
ih.indexBlockOffset = encoding.UnmarshalUint64(src[16:])
ih.indexBlockSize = encoding.UnmarshalUint64(src[24:])
return src[32:], nil
}
lib/logstorage: refactor storage format to be more efficient for querying wide events It has been appeared that VictoriaLogs is frequently used for collecting logs with tens of fields. For example, standard Kuberntes setup on top of Filebeat generates more than 20 fields per each log. Such logs are also known as "wide events". The previous storage format was optimized for logs with a few fields. When at least a single field was referenced in the query, then the all the meta-information about all the log fields was unpacked and parsed per each scanned block during the query. This could require a lot of additional disk IO and CPU time when logs contain many fields. Resolve this issue by providing an (field -> metainfo_offset) index per each field in every data block. This index allows reading and extracting only the needed metainfo for fields used in the query. This index is stored in columnsHeaderIndexFilename ( columns_header_index.bin ). This allows increasing performance for queries over wide events by 10x and more. Another issue was that the data for bloom filters and field values across all the log fields except of _msg was intermixed in two files - fieldBloomFilename ( field_bloom.bin ) and fieldValuesFilename ( field_values.bin ). This could result in huge disk read IO overhead when some small field was referred in the query, since the Operating System usually reads more data than requested. It reads the data from disk in at least 4KiB blocks (usually the block size is much bigger in the range 64KiB - 512KiB). So, if 512-byte bloom filter or values' block is read from the file, then the Operating System reads up to 512KiB of data from disk, which results in 1000x disk read IO overhead. This overhead isn't visible for recently accessed data, since this data is usually stored in RAM (aka Operating System page cache), but this overhead may become very annoying when performing the query over large volumes of data which isn't present in OS page cache. The solution for this issue is to split bloom filters and field values across multiple shards. This reduces the worst-case disk read IO overhead by at least Nx where N is the number of shards, while the disk read IO overhead is completely removed in best case when the number of columns doesn't exceed N. Currently the number of shards is 8 - see bloomValuesShardsCount . This solution increases performance for queries over large volumes of newly ingested data by up to 1000x. The new storage format is versioned as v1, while the old storage format is version as v0. It is stored in the partHeader.FormatVersion. Parts with the old storage format are converted into parts with the new storage format during background merge. It is possible to force merge by querying /internal/force_merge HTTP endpoint - see https://docs.victoriametrics.com/victorialogs/#forced-merge .
2024-10-16 14:18:28 +00:00
// mustWriteIndexBlockHeaders writes metaindexData to w.
func mustWriteIndexBlockHeaders(w *writerWithStats, metaindexData []byte) {
bb := longTermBufPool.Get()
bb.B = encoding.CompressZSTDLevel(bb.B[:0], metaindexData, 1)
w.MustWrite(bb.B)
if len(bb.B) < 1024*1024 {
longTermBufPool.Put(bb)
}
}
// mustReadIndexBlockHeaders reads indexBlockHeader entries from r, appends them to dst and returns the result.
func mustReadIndexBlockHeaders(dst []indexBlockHeader, r *readerWithStats) []indexBlockHeader {
data, err := io.ReadAll(r)
if err != nil {
lib/logstorage: refactor storage format to be more efficient for querying wide events It has been appeared that VictoriaLogs is frequently used for collecting logs with tens of fields. For example, standard Kuberntes setup on top of Filebeat generates more than 20 fields per each log. Such logs are also known as "wide events". The previous storage format was optimized for logs with a few fields. When at least a single field was referenced in the query, then the all the meta-information about all the log fields was unpacked and parsed per each scanned block during the query. This could require a lot of additional disk IO and CPU time when logs contain many fields. Resolve this issue by providing an (field -> metainfo_offset) index per each field in every data block. This index allows reading and extracting only the needed metainfo for fields used in the query. This index is stored in columnsHeaderIndexFilename ( columns_header_index.bin ). This allows increasing performance for queries over wide events by 10x and more. Another issue was that the data for bloom filters and field values across all the log fields except of _msg was intermixed in two files - fieldBloomFilename ( field_bloom.bin ) and fieldValuesFilename ( field_values.bin ). This could result in huge disk read IO overhead when some small field was referred in the query, since the Operating System usually reads more data than requested. It reads the data from disk in at least 4KiB blocks (usually the block size is much bigger in the range 64KiB - 512KiB). So, if 512-byte bloom filter or values' block is read from the file, then the Operating System reads up to 512KiB of data from disk, which results in 1000x disk read IO overhead. This overhead isn't visible for recently accessed data, since this data is usually stored in RAM (aka Operating System page cache), but this overhead may become very annoying when performing the query over large volumes of data which isn't present in OS page cache. The solution for this issue is to split bloom filters and field values across multiple shards. This reduces the worst-case disk read IO overhead by at least Nx where N is the number of shards, while the disk read IO overhead is completely removed in best case when the number of columns doesn't exceed N. Currently the number of shards is 8 - see bloomValuesShardsCount . This solution increases performance for queries over large volumes of newly ingested data by up to 1000x. The new storage format is versioned as v1, while the old storage format is version as v0. It is stored in the partHeader.FormatVersion. Parts with the old storage format are converted into parts with the new storage format during background merge. It is possible to force merge by querying /internal/force_merge HTTP endpoint - see https://docs.victoriametrics.com/victorialogs/#forced-merge .
2024-10-16 14:18:28 +00:00
logger.Panicf("FATAL: %s: cannot read indexBlockHeader entries: %s", r.Path(), err)
}
bb := longTermBufPool.Get()
bb.B, err = encoding.DecompressZSTD(bb.B[:0], data)
if err != nil {
lib/logstorage: refactor storage format to be more efficient for querying wide events It has been appeared that VictoriaLogs is frequently used for collecting logs with tens of fields. For example, standard Kuberntes setup on top of Filebeat generates more than 20 fields per each log. Such logs are also known as "wide events". The previous storage format was optimized for logs with a few fields. When at least a single field was referenced in the query, then the all the meta-information about all the log fields was unpacked and parsed per each scanned block during the query. This could require a lot of additional disk IO and CPU time when logs contain many fields. Resolve this issue by providing an (field -> metainfo_offset) index per each field in every data block. This index allows reading and extracting only the needed metainfo for fields used in the query. This index is stored in columnsHeaderIndexFilename ( columns_header_index.bin ). This allows increasing performance for queries over wide events by 10x and more. Another issue was that the data for bloom filters and field values across all the log fields except of _msg was intermixed in two files - fieldBloomFilename ( field_bloom.bin ) and fieldValuesFilename ( field_values.bin ). This could result in huge disk read IO overhead when some small field was referred in the query, since the Operating System usually reads more data than requested. It reads the data from disk in at least 4KiB blocks (usually the block size is much bigger in the range 64KiB - 512KiB). So, if 512-byte bloom filter or values' block is read from the file, then the Operating System reads up to 512KiB of data from disk, which results in 1000x disk read IO overhead. This overhead isn't visible for recently accessed data, since this data is usually stored in RAM (aka Operating System page cache), but this overhead may become very annoying when performing the query over large volumes of data which isn't present in OS page cache. The solution for this issue is to split bloom filters and field values across multiple shards. This reduces the worst-case disk read IO overhead by at least Nx where N is the number of shards, while the disk read IO overhead is completely removed in best case when the number of columns doesn't exceed N. Currently the number of shards is 8 - see bloomValuesShardsCount . This solution increases performance for queries over large volumes of newly ingested data by up to 1000x. The new storage format is versioned as v1, while the old storage format is version as v0. It is stored in the partHeader.FormatVersion. Parts with the old storage format are converted into parts with the new storage format during background merge. It is possible to force merge by querying /internal/force_merge HTTP endpoint - see https://docs.victoriametrics.com/victorialogs/#forced-merge .
2024-10-16 14:18:28 +00:00
logger.Panicf("FATAL: %s: cannot decompress indexBlockHeader entries: %s", r.Path(), err)
}
dst, err = unmarshalIndexBlockHeaders(dst, bb.B)
if len(bb.B) < 1024*1024 {
longTermBufPool.Put(bb)
}
if err != nil {
lib/logstorage: refactor storage format to be more efficient for querying wide events It has been appeared that VictoriaLogs is frequently used for collecting logs with tens of fields. For example, standard Kuberntes setup on top of Filebeat generates more than 20 fields per each log. Such logs are also known as "wide events". The previous storage format was optimized for logs with a few fields. When at least a single field was referenced in the query, then the all the meta-information about all the log fields was unpacked and parsed per each scanned block during the query. This could require a lot of additional disk IO and CPU time when logs contain many fields. Resolve this issue by providing an (field -> metainfo_offset) index per each field in every data block. This index allows reading and extracting only the needed metainfo for fields used in the query. This index is stored in columnsHeaderIndexFilename ( columns_header_index.bin ). This allows increasing performance for queries over wide events by 10x and more. Another issue was that the data for bloom filters and field values across all the log fields except of _msg was intermixed in two files - fieldBloomFilename ( field_bloom.bin ) and fieldValuesFilename ( field_values.bin ). This could result in huge disk read IO overhead when some small field was referred in the query, since the Operating System usually reads more data than requested. It reads the data from disk in at least 4KiB blocks (usually the block size is much bigger in the range 64KiB - 512KiB). So, if 512-byte bloom filter or values' block is read from the file, then the Operating System reads up to 512KiB of data from disk, which results in 1000x disk read IO overhead. This overhead isn't visible for recently accessed data, since this data is usually stored in RAM (aka Operating System page cache), but this overhead may become very annoying when performing the query over large volumes of data which isn't present in OS page cache. The solution for this issue is to split bloom filters and field values across multiple shards. This reduces the worst-case disk read IO overhead by at least Nx where N is the number of shards, while the disk read IO overhead is completely removed in best case when the number of columns doesn't exceed N. Currently the number of shards is 8 - see bloomValuesShardsCount . This solution increases performance for queries over large volumes of newly ingested data by up to 1000x. The new storage format is versioned as v1, while the old storage format is version as v0. It is stored in the partHeader.FormatVersion. Parts with the old storage format are converted into parts with the new storage format during background merge. It is possible to force merge by querying /internal/force_merge HTTP endpoint - see https://docs.victoriametrics.com/victorialogs/#forced-merge .
2024-10-16 14:18:28 +00:00
logger.Panicf("FATAL: %s: cannot parse indexBlockHeader entries: %s", r.Path(), err)
}
lib/logstorage: refactor storage format to be more efficient for querying wide events It has been appeared that VictoriaLogs is frequently used for collecting logs with tens of fields. For example, standard Kuberntes setup on top of Filebeat generates more than 20 fields per each log. Such logs are also known as "wide events". The previous storage format was optimized for logs with a few fields. When at least a single field was referenced in the query, then the all the meta-information about all the log fields was unpacked and parsed per each scanned block during the query. This could require a lot of additional disk IO and CPU time when logs contain many fields. Resolve this issue by providing an (field -> metainfo_offset) index per each field in every data block. This index allows reading and extracting only the needed metainfo for fields used in the query. This index is stored in columnsHeaderIndexFilename ( columns_header_index.bin ). This allows increasing performance for queries over wide events by 10x and more. Another issue was that the data for bloom filters and field values across all the log fields except of _msg was intermixed in two files - fieldBloomFilename ( field_bloom.bin ) and fieldValuesFilename ( field_values.bin ). This could result in huge disk read IO overhead when some small field was referred in the query, since the Operating System usually reads more data than requested. It reads the data from disk in at least 4KiB blocks (usually the block size is much bigger in the range 64KiB - 512KiB). So, if 512-byte bloom filter or values' block is read from the file, then the Operating System reads up to 512KiB of data from disk, which results in 1000x disk read IO overhead. This overhead isn't visible for recently accessed data, since this data is usually stored in RAM (aka Operating System page cache), but this overhead may become very annoying when performing the query over large volumes of data which isn't present in OS page cache. The solution for this issue is to split bloom filters and field values across multiple shards. This reduces the worst-case disk read IO overhead by at least Nx where N is the number of shards, while the disk read IO overhead is completely removed in best case when the number of columns doesn't exceed N. Currently the number of shards is 8 - see bloomValuesShardsCount . This solution increases performance for queries over large volumes of newly ingested data by up to 1000x. The new storage format is versioned as v1, while the old storage format is version as v0. It is stored in the partHeader.FormatVersion. Parts with the old storage format are converted into parts with the new storage format during background merge. It is possible to force merge by querying /internal/force_merge HTTP endpoint - see https://docs.victoriametrics.com/victorialogs/#forced-merge .
2024-10-16 14:18:28 +00:00
return dst
}
// unmarshalIndexBlockHeaders appends unmarshaled from src indexBlockHeader entries to dst and returns the result.
func unmarshalIndexBlockHeaders(dst []indexBlockHeader, src []byte) ([]indexBlockHeader, error) {
dstOrig := dst
for len(src) > 0 {
if len(dst) < cap(dst) {
dst = dst[:len(dst)+1]
} else {
dst = append(dst, indexBlockHeader{})
}
ih := &dst[len(dst)-1]
tail, err := ih.unmarshal(src)
if err != nil {
return dstOrig, fmt.Errorf("cannot unmarshal indexBlockHeader %d: %w", len(dst)-len(dstOrig), err)
}
src = tail
}
if err := validateIndexBlockHeaders(dst[len(dstOrig):]); err != nil {
return dstOrig, err
}
return dst, nil
}
func validateIndexBlockHeaders(ihs []indexBlockHeader) error {
for i := 1; i < len(ihs); i++ {
if ihs[i].streamID.less(&ihs[i-1].streamID) {
return fmt.Errorf("unexpected indexBlockHeader with smaller streamID=%s after bigger streamID=%s", &ihs[i].streamID, &ihs[i-1].streamID)
}
}
return nil
}