Documentation
¶
Overview ¶
Package postings is the inverted index: for each (name, value) attribute — identified by **interned symbol ids**, not strings — it keeps the sorted list of [signal.SeriesID]s that carry it, and composes those lists with lazy set-op iterators (Intersect/Merge/Without) to resolve label matchers to series.
Keys are uint32 symbol ids (from index/symbols), so the index is zero-alloc and memory-compact, and it preserves attribute types: the value id is interned from the value's type-tagged encoding, so int 5, "5" and 5.0 are distinct buckets — not a stringified collapse. Matching is **callback-based and typed at the edges**: MemPostings.Select hands a predicate the candidate value id, and the caller decodes it to a signal.Value and applies any rule (regexp, numeric range, exact). Storage knows nothing about a query language's operators.
Index ¶
- func ToSlice(p Postings) ([]signal.SeriesID, error)
- type Matcher
- type MemPostings
- func (p *MemPostings) Add(series signal.SeriesID, nameID, valueID uint32)
- func (p *MemPostings) All() Postings
- func (p *MemPostings) EnsureSorted()
- func (p *MemPostings) ForName(nameID uint32) Postings
- func (p *MemPostings) Get(nameID, valueID uint32) Postings
- func (p *MemPostings) LabelValues(nameID uint32) []uint32
- func (p *MemPostings) Resolve(ms ...Matcher) Postings
- func (p *MemPostings) Select(nameID uint32, match func(valueID uint32) bool) Postings
- func (p *MemPostings) Sorted() bool
- func (p *MemPostings) WithoutName(nameID uint32) Postings
- type Postings
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
Types ¶
type Matcher ¶
Matcher is a single label condition: the value ids of NameID that satisfy the Match predicate. The callback form keeps storage language- and type-neutral — the caller decodes the value id to a typed value and supplies the rule.
type MemPostings ¶
type MemPostings struct {
// contains filtered or unexported fields
}
MemPostings is the in-memory inverted index: nameID → valueID → sorted series list, plus the set of all series. The zero value is not usable; create one with NewMemPostings. Not safe for concurrent use; callers own synchronization (the first read sorts the lists in place).
func (*MemPostings) Add ¶
func (p *MemPostings) Add(series signal.SeriesID, nameID, valueID uint32)
Add records that series carries the attribute nameID=valueID (both interned symbol ids). Re-adding the same triple is fine; duplicates are removed on the first read.
func (*MemPostings) All ¶
func (p *MemPostings) All() Postings
All returns a Postings over every series in the index.
func (*MemPostings) EnsureSorted ¶ added in v0.2.0
func (p *MemPostings) EnsureSorted()
EnsureSorted sorts and deduplicates the index in place (idempotent; a no-op once sorted). It is the exported form of the read-triggered lazy sort: a caller holding the index's **exclusive** lock calls it after writes so that subsequent **concurrent** reads (which only inspect the already-sorted lists) never trigger the in-place mutation. See MemPostings.Sorted.
func (*MemPostings) ForName ¶
func (p *MemPostings) ForName(nameID uint32) Postings
ForName returns the series that carry the attribute nameID with any value.
func (*MemPostings) Get ¶
func (p *MemPostings) Get(nameID, valueID uint32) Postings
Get returns the series carrying exactly nameID=valueID.
func (*MemPostings) LabelValues ¶
func (p *MemPostings) LabelValues(nameID uint32) []uint32
LabelValues returns, sorted, the distinct value ids seen for nameID.
func (*MemPostings) Resolve ¶
func (p *MemPostings) Resolve(ms ...Matcher) Postings
Resolve returns the series matching all matchers (their intersection). With no matchers it returns every series. Negation and absent-label semantics are composed by the caller with Without / MemPostings.WithoutName.
func (*MemPostings) Select ¶
func (p *MemPostings) Select(nameID uint32, match func(valueID uint32) bool) Postings
Select returns the union of the series whose value id for nameID satisfies match. The predicate receives the candidate value id; the caller resolves it to a typed signal.Value (via the symbol table) and applies any rule, so storage stays free of query-language operators. For exact equality prefer MemPostings.Get (an O(1) lookup).
func (*MemPostings) Sorted ¶ added in v0.2.0
func (p *MemPostings) Sorted() bool
Sorted reports whether the index is already sorted (no read will mutate it). A caller that owns synchronization checks this under a read lock to decide whether it must upgrade to an exclusive lock and call MemPostings.EnsureSorted before issuing concurrent reads.
func (*MemPostings) WithoutName ¶
func (p *MemPostings) WithoutName(nameID uint32) Postings
WithoutName returns the series that do not carry the attribute nameID at all.
type Postings ¶
type Postings interface {
Next() bool
Seek(v signal.SeriesID) bool
At() signal.SeriesID
Err() error
}
Postings is a forward iterator over a sorted, deduplicated stream of signal.SeriesID. Next advances; Seek jumps to the first id ≥ v; At returns the current id; Err reports a terminal error. The set-op constructors (Intersect, Merge, Without) compose Postings lazily, so resolving a matcher never materializes intermediate lists.
Source Files