Like Prometheus, but for logs.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
loki/pkg/dataobj/section.go

184 lines
6.7 KiB

refactor(dataobj): invert dependency between dataobj and sections (#17762) Originally, the dataobj package was a higher-level API around sections. This design caused it to become a bottleneck: * Implementing any new public behaviour for a section required bubbling it up to the dataobj API for it to be exposed, making it tedious to add new sections or update existing ones. * The `dataobj.Builder` pattern was focused on constructing dataobjs for storing log data, which will cause friction as we build objects around other use cases. This PR builds on top of the foundation laid out by #17704 and #17708, fully inverting the dependency between dataobj and sections: * The `dataobj` package has no knowledge of what sections exist, and can now be used for writing and reading generic sections. Section packages now create higher-level APIs around the abstractions provided by `dataobj`. * Section packages are now public, and callers interact directly with these packages for writing and reading section-specific data. * All logic for a section (encoding, decoding, buffering, reading) is now fully self-contained inside the section package. Previously, the implementation of each section was spread across three packages (`pkg/dataobj/internal/encoding`, `pkg/dataobj/internal/sections/SECTION`, `pkg/dataobj`). * Cutting a section is now a decision made by the caller rather than the section implementation. Previously, the logs section builder would create multiple sections. For the most part, this change is a no-op, with two exceptions: 1. Section cutting is now performed by the caller; however, this shouldn't result in any issues. 2. Removing the high-level `dataobj.Stream` and `dataobj.Record` types will temporarily reduce the allocation gains from #16988. I will address this after this PR is merged.
8 months ago
package dataobj
import (
"context"
"io"
"iter"
"strconv"
refactor(dataobj): invert dependency between dataobj and sections (#17762) Originally, the dataobj package was a higher-level API around sections. This design caused it to become a bottleneck: * Implementing any new public behaviour for a section required bubbling it up to the dataobj API for it to be exposed, making it tedious to add new sections or update existing ones. * The `dataobj.Builder` pattern was focused on constructing dataobjs for storing log data, which will cause friction as we build objects around other use cases. This PR builds on top of the foundation laid out by #17704 and #17708, fully inverting the dependency between dataobj and sections: * The `dataobj` package has no knowledge of what sections exist, and can now be used for writing and reading generic sections. Section packages now create higher-level APIs around the abstractions provided by `dataobj`. * Section packages are now public, and callers interact directly with these packages for writing and reading section-specific data. * All logic for a section (encoding, decoding, buffering, reading) is now fully self-contained inside the section package. Previously, the implementation of each section was spread across three packages (`pkg/dataobj/internal/encoding`, `pkg/dataobj/internal/sections/SECTION`, `pkg/dataobj`). * Cutting a section is now a decision made by the caller rather than the section implementation. Previously, the logs section builder would create multiple sections. For the most part, this change is a no-op, with two exceptions: 1. Section cutting is now performed by the caller; however, this shouldn't result in any issues. 2. Removing the high-level `dataobj.Stream` and `dataobj.Record` types will temporarily reduce the allocation gains from #16988. I will address this after this PR is merged.
8 months ago
)
// A Sections is a slice of [Section].
type Sections []*Section
// Filter returns an iterator over sections that pass some predicate. The index
// field is the number of the section that passed the predicate.
func (s Sections) Filter(predicate func(*Section) bool) iter.Seq2[int, *Section] {
return func(yield func(int, *Section) bool) {
var matches int
for _, sec := range s {
if !predicate(sec) {
continue
} else if !yield(matches, sec) {
return
}
matches++
}
}
}
// Count returns the number of sections that pass some predicate.
func (s Sections) Count(predicate func(*Section) bool) int {
var count int
for range s.Filter(predicate) {
count++
}
return count
}
// A Section is a subset of an [Object] that holds a specific type of data. Use
// section packages for higher-level abstractions around sections.
type Section struct {
Type SectionType // The type denoting the kind of data held in a section.
Reader SectionReader // The low-level reader for a Section.
// Tenant specifies the tenant that owns this section. Tenant is required
// for sections which wholly contain tenant-specific data.
Tenant string
refactor(dataobj): invert dependency between dataobj and sections (#17762) Originally, the dataobj package was a higher-level API around sections. This design caused it to become a bottleneck: * Implementing any new public behaviour for a section required bubbling it up to the dataobj API for it to be exposed, making it tedious to add new sections or update existing ones. * The `dataobj.Builder` pattern was focused on constructing dataobjs for storing log data, which will cause friction as we build objects around other use cases. This PR builds on top of the foundation laid out by #17704 and #17708, fully inverting the dependency between dataobj and sections: * The `dataobj` package has no knowledge of what sections exist, and can now be used for writing and reading generic sections. Section packages now create higher-level APIs around the abstractions provided by `dataobj`. * Section packages are now public, and callers interact directly with these packages for writing and reading section-specific data. * All logic for a section (encoding, decoding, buffering, reading) is now fully self-contained inside the section package. Previously, the implementation of each section was spread across three packages (`pkg/dataobj/internal/encoding`, `pkg/dataobj/internal/sections/SECTION`, `pkg/dataobj`). * Cutting a section is now a decision made by the caller rather than the section implementation. Previously, the logs section builder would create multiple sections. For the most part, this change is a no-op, with two exceptions: 1. Section cutting is now performed by the caller; however, this shouldn't result in any issues. 2. Removing the high-level `dataobj.Stream` and `dataobj.Record` types will temporarily reduce the allocation gains from #16988. I will address this after this PR is merged.
8 months ago
}
// SectionType uniquely identifies a [Section] type.
type SectionType struct {
Namespace string // A namesapce for the section (e.g., "github.com/grafana/loki").
Kind string // The kind of section, scoped to the namespace (e.g., "logs").
// Version is an optional section-specified value denoting an encoding
// version of the section.
Version uint32
refactor(dataobj): invert dependency between dataobj and sections (#17762) Originally, the dataobj package was a higher-level API around sections. This design caused it to become a bottleneck: * Implementing any new public behaviour for a section required bubbling it up to the dataobj API for it to be exposed, making it tedious to add new sections or update existing ones. * The `dataobj.Builder` pattern was focused on constructing dataobjs for storing log data, which will cause friction as we build objects around other use cases. This PR builds on top of the foundation laid out by #17704 and #17708, fully inverting the dependency between dataobj and sections: * The `dataobj` package has no knowledge of what sections exist, and can now be used for writing and reading generic sections. Section packages now create higher-level APIs around the abstractions provided by `dataobj`. * Section packages are now public, and callers interact directly with these packages for writing and reading section-specific data. * All logic for a section (encoding, decoding, buffering, reading) is now fully self-contained inside the section package. Previously, the implementation of each section was spread across three packages (`pkg/dataobj/internal/encoding`, `pkg/dataobj/internal/sections/SECTION`, `pkg/dataobj`). * Cutting a section is now a decision made by the caller rather than the section implementation. Previously, the logs section builder would create multiple sections. For the most part, this change is a no-op, with two exceptions: 1. Section cutting is now performed by the caller; however, this shouldn't result in any issues. 2. Removing the high-level `dataobj.Stream` and `dataobj.Record` types will temporarily reduce the allocation gains from #16988. I will address this after this PR is merged.
8 months ago
}
// Equals returns true if o has the same namespace and kind as ty. The Version
// field is not checked.
func (ty SectionType) Equals(o SectionType) bool {
return ty.Namespace == o.Namespace && ty.Kind == o.Kind
}
refactor(dataobj): invert dependency between dataobj and sections (#17762) Originally, the dataobj package was a higher-level API around sections. This design caused it to become a bottleneck: * Implementing any new public behaviour for a section required bubbling it up to the dataobj API for it to be exposed, making it tedious to add new sections or update existing ones. * The `dataobj.Builder` pattern was focused on constructing dataobjs for storing log data, which will cause friction as we build objects around other use cases. This PR builds on top of the foundation laid out by #17704 and #17708, fully inverting the dependency between dataobj and sections: * The `dataobj` package has no knowledge of what sections exist, and can now be used for writing and reading generic sections. Section packages now create higher-level APIs around the abstractions provided by `dataobj`. * Section packages are now public, and callers interact directly with these packages for writing and reading section-specific data. * All logic for a section (encoding, decoding, buffering, reading) is now fully self-contained inside the section package. Previously, the implementation of each section was spread across three packages (`pkg/dataobj/internal/encoding`, `pkg/dataobj/internal/sections/SECTION`, `pkg/dataobj`). * Cutting a section is now a decision made by the caller rather than the section implementation. Previously, the logs section builder would create multiple sections. For the most part, this change is a no-op, with two exceptions: 1. Section cutting is now performed by the caller; however, this shouldn't result in any issues. 2. Removing the high-level `dataobj.Stream` and `dataobj.Record` types will temporarily reduce the allocation gains from #16988. I will address this after this PR is merged.
8 months ago
func (ty SectionType) String() string {
base := ty.Namespace + "/" + ty.Kind
base += "@v" + strconv.FormatUint(uint64(ty.Version), 10)
return base
refactor(dataobj): invert dependency between dataobj and sections (#17762) Originally, the dataobj package was a higher-level API around sections. This design caused it to become a bottleneck: * Implementing any new public behaviour for a section required bubbling it up to the dataobj API for it to be exposed, making it tedious to add new sections or update existing ones. * The `dataobj.Builder` pattern was focused on constructing dataobjs for storing log data, which will cause friction as we build objects around other use cases. This PR builds on top of the foundation laid out by #17704 and #17708, fully inverting the dependency between dataobj and sections: * The `dataobj` package has no knowledge of what sections exist, and can now be used for writing and reading generic sections. Section packages now create higher-level APIs around the abstractions provided by `dataobj`. * Section packages are now public, and callers interact directly with these packages for writing and reading section-specific data. * All logic for a section (encoding, decoding, buffering, reading) is now fully self-contained inside the section package. Previously, the implementation of each section was spread across three packages (`pkg/dataobj/internal/encoding`, `pkg/dataobj/internal/sections/SECTION`, `pkg/dataobj`). * Cutting a section is now a decision made by the caller rather than the section implementation. Previously, the logs section builder would create multiple sections. For the most part, this change is a no-op, with two exceptions: 1. Section cutting is now performed by the caller; however, this shouldn't result in any issues. 2. Removing the high-level `dataobj.Stream` and `dataobj.Record` types will temporarily reduce the allocation gains from #16988. I will address this after this PR is merged.
8 months ago
}
// SectionReader is a low-level interface to read data ranges and metadata from
// a section.
//
// Section packages provider higher-level abstractions around [Section] using
// this interface.
type SectionReader interface {
// ExtensionData returns optional encoded information about the section
// stored at the file level, provided through the [SectionWriter]. Sections
// can use this for retrieving critical information that must be known
// without needing to read the metadata first.
//
// ExtensionData will be nil if no extension data is available.
ExtensionData() []byte
refactor(dataobj): invert dependency between dataobj and sections (#17762) Originally, the dataobj package was a higher-level API around sections. This design caused it to become a bottleneck: * Implementing any new public behaviour for a section required bubbling it up to the dataobj API for it to be exposed, making it tedious to add new sections or update existing ones. * The `dataobj.Builder` pattern was focused on constructing dataobjs for storing log data, which will cause friction as we build objects around other use cases. This PR builds on top of the foundation laid out by #17704 and #17708, fully inverting the dependency between dataobj and sections: * The `dataobj` package has no knowledge of what sections exist, and can now be used for writing and reading generic sections. Section packages now create higher-level APIs around the abstractions provided by `dataobj`. * Section packages are now public, and callers interact directly with these packages for writing and reading section-specific data. * All logic for a section (encoding, decoding, buffering, reading) is now fully self-contained inside the section package. Previously, the implementation of each section was spread across three packages (`pkg/dataobj/internal/encoding`, `pkg/dataobj/internal/sections/SECTION`, `pkg/dataobj`). * Cutting a section is now a decision made by the caller rather than the section implementation. Previously, the logs section builder would create multiple sections. For the most part, this change is a no-op, with two exceptions: 1. Section cutting is now performed by the caller; however, this shouldn't result in any issues. 2. Removing the high-level `dataobj.Stream` and `dataobj.Record` types will temporarily reduce the allocation gains from #16988. I will address this after this PR is merged.
8 months ago
// DataRange opens a reader of length bytes from the data region of a
// section. The offset argument determines where in the data region reading
// should start.
//
// DataRange returns an error if the read fails or if offset+length goes
// beyond the readable data region. The returned reader is only valid as long
// as the provided ctx is not canceled.
DataRange(ctx context.Context, offset, length int64) (io.ReadCloser, error)
// MetadataRange opens a reader of length bytes from the metadata region of
// a section. The offset argument determines where in the metadata region
// reading should start.
//
// MetadataRange returns an error if the read fails or if offset+length goes
// beyond the readable metadata region. The returned reader is only valid as long
// as the provided ctx is not canceled.
MetadataRange(ctx context.Context, offset, length int64) (io.ReadCloser, error)
// DataSize returns the total size of the data region of a section. DataSize
// returns 0 for sections with no data region.
DataSize() int64
// MetadataSize returns the total size of the metadata region of a section.
// MetadataSize returns 0 for sections with no metadata region.
MetadataSize() int64
refactor(dataobj): invert dependency between dataobj and sections (#17762) Originally, the dataobj package was a higher-level API around sections. This design caused it to become a bottleneck: * Implementing any new public behaviour for a section required bubbling it up to the dataobj API for it to be exposed, making it tedious to add new sections or update existing ones. * The `dataobj.Builder` pattern was focused on constructing dataobjs for storing log data, which will cause friction as we build objects around other use cases. This PR builds on top of the foundation laid out by #17704 and #17708, fully inverting the dependency between dataobj and sections: * The `dataobj` package has no knowledge of what sections exist, and can now be used for writing and reading generic sections. Section packages now create higher-level APIs around the abstractions provided by `dataobj`. * Section packages are now public, and callers interact directly with these packages for writing and reading section-specific data. * All logic for a section (encoding, decoding, buffering, reading) is now fully self-contained inside the section package. Previously, the implementation of each section was spread across three packages (`pkg/dataobj/internal/encoding`, `pkg/dataobj/internal/sections/SECTION`, `pkg/dataobj`). * Cutting a section is now a decision made by the caller rather than the section implementation. Previously, the logs section builder would create multiple sections. For the most part, this change is a no-op, with two exceptions: 1. Section cutting is now performed by the caller; however, this shouldn't result in any issues. 2. Removing the high-level `dataobj.Stream` and `dataobj.Record` types will temporarily reduce the allocation gains from #16988. I will address this after this PR is merged.
8 months ago
}
// A SectionBuilder accumulates data for a single in-progress section.
//
// Each section package provides an implementation of SectionBuilder that
// includes utilities to buffer data into that section. Callers should use
// Bytes or EstimatedSize to determine when enough data has been accumulated
// into a section.
type SectionBuilder interface {
// Type returns the SectionType representing the section being built.
// Implementations are responsible for guaranteeing that two no
// SectionBuilders return the same SectionType for different encodings.
//
// The returned Type is encoded directly into data objects. Implementations
// that change SectionType values should be careful to continue supporting
// old values for backwards compatibility.
Type() SectionType
// Flush encodes and flushes the section to w. Encodings that rely on byte
// offsets should be relative to the first byte of the section's data.
//
// Flush returns the number of bytes written to w, and any error encountered
// while encoding or flushing.
//
// After Flush is called, the SectionBuilder is reset to a fresh state and
// can be reused.
Flush(w SectionWriter) (n int64, err error)
// Reset resets the SectionBuilder to a fresh state.
Reset()
}
// SectionWriter writes data object sections to an underlying stream, such as a
// data object.
type SectionWriter interface {
// WriteSection writes a section to the underlying data stream, partitioned
// by section data and section metadata. It returns the sum of bytes written
// from both input slices (0 <= n <= len(data)+len(metadata)) and any error
// encountered that caused the write to stop early.
//
// The opts argument provides additional information about the section being
// written. If opts is nil, the section is written without any additional
// context.
//
refactor(dataobj): invert dependency between dataobj and sections (#17762) Originally, the dataobj package was a higher-level API around sections. This design caused it to become a bottleneck: * Implementing any new public behaviour for a section required bubbling it up to the dataobj API for it to be exposed, making it tedious to add new sections or update existing ones. * The `dataobj.Builder` pattern was focused on constructing dataobjs for storing log data, which will cause friction as we build objects around other use cases. This PR builds on top of the foundation laid out by #17704 and #17708, fully inverting the dependency between dataobj and sections: * The `dataobj` package has no knowledge of what sections exist, and can now be used for writing and reading generic sections. Section packages now create higher-level APIs around the abstractions provided by `dataobj`. * Section packages are now public, and callers interact directly with these packages for writing and reading section-specific data. * All logic for a section (encoding, decoding, buffering, reading) is now fully self-contained inside the section package. Previously, the implementation of each section was spread across three packages (`pkg/dataobj/internal/encoding`, `pkg/dataobj/internal/sections/SECTION`, `pkg/dataobj`). * Cutting a section is now a decision made by the caller rather than the section implementation. Previously, the logs section builder would create multiple sections. For the most part, this change is a no-op, with two exceptions: 1. Section cutting is now performed by the caller; however, this shouldn't result in any issues. 2. Removing the high-level `dataobj.Stream` and `dataobj.Record` types will temporarily reduce the allocation gains from #16988. I will address this after this PR is merged.
8 months ago
// Implementations of WriteSection:
//
// - Must return an error if the write stops early.
// - Must not modify the slices passed to it, even temporarily.
// - Must not retain references to slices after WriteSection returns.
//
// The physical layout of data and metadata is not defined: they may be
// written non-contiguously, interleaved, or in any order.
WriteSection(opts *WriteSectionOptions, data, metadata []byte) (n int64, err error)
}
// WriteSectionOptions provides additional options when writing sections.
type WriteSectionOptions struct {
// Tenant that owns the written data and metadata. Tenant must be set for
// sections that are wholly owned by a single tenant.
Tenant string
// ExtensionData is an optional field for section information to store at
// the file level. To minimize the cost of opening data objects, sections
// should only use this field for information that's required to start
// reading section metadata and to keep the payload as small as possible.
//
// ExtensionData does not impact the return value of n in
// [SectionWriter.WriteSection].
//
// Implementations of [SectionWriter] must not retain references to this
// slice after WriteSection returns.
ExtensionData []byte
refactor(dataobj): invert dependency between dataobj and sections (#17762) Originally, the dataobj package was a higher-level API around sections. This design caused it to become a bottleneck: * Implementing any new public behaviour for a section required bubbling it up to the dataobj API for it to be exposed, making it tedious to add new sections or update existing ones. * The `dataobj.Builder` pattern was focused on constructing dataobjs for storing log data, which will cause friction as we build objects around other use cases. This PR builds on top of the foundation laid out by #17704 and #17708, fully inverting the dependency between dataobj and sections: * The `dataobj` package has no knowledge of what sections exist, and can now be used for writing and reading generic sections. Section packages now create higher-level APIs around the abstractions provided by `dataobj`. * Section packages are now public, and callers interact directly with these packages for writing and reading section-specific data. * All logic for a section (encoding, decoding, buffering, reading) is now fully self-contained inside the section package. Previously, the implementation of each section was spread across three packages (`pkg/dataobj/internal/encoding`, `pkg/dataobj/internal/sections/SECTION`, `pkg/dataobj`). * Cutting a section is now a decision made by the caller rather than the section implementation. Previously, the logs section builder would create multiple sections. For the most part, this change is a no-op, with two exceptions: 1. Section cutting is now performed by the caller; however, this shouldn't result in any issues. 2. Removing the high-level `dataobj.Stream` and `dataobj.Record` types will temporarily reduce the allocation gains from #16988. I will address this after this PR is merged.
8 months ago
}