Documentation
¶
Overview ¶
Package metrics is GoGraph's optional observability surface.
The package exposes lightweight counters and latency observers that public blocking APIs can populate. Two backends are wired up:
- the default no-op backend: zero overhead, used when no external metrics system is configured.
- a Prometheus-compatible backend (opt-in via SetBackend): once installed, latency histograms and counters are exported through the standard prometheus.Registry mechanism.
The Prometheus client_golang import is intentionally NOT a dependency of this package: keeping it out of go.mod means every consumer that doesn't want Prometheus pays no module-graph cost. Callers that want the Prometheus backend implement the Backend interface in their own code; the [Prometheus] helper lives in a separate internal/metrics/prometheus subpackage when added later.
Wire-up. Every public blocking API in search/, search/centrality/, search/community/, search/flow/, search/extern/, graph/io/{csv,graphml,dot,jsonl}, and store/{wal,snapshot,txn,checkpoint,recovery,bulk} emits a latency observation under the name "<package-path>.<ExportedSymbol>" and increments a paired "<package-path>.<ExportedSymbol>.errors" counter on the error path. The authoritative inventory of every wired metric is in docs/metrics.md; the wireup smoke test in internal/metrics/wireup_test.go fails loudly if a wired symbol stops emitting its expected name.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func IncCounter ¶
IncCounter increments the named counter by delta on the current backend.
func ObserveLatency ¶
ObserveLatency records a latency sample on the current backend.
func SetBackend ¶
func SetBackend(b Backend)
SetBackend swaps the global metrics sink. Pass nil to restore the no-op default. The function is safe to call from any goroutine at any time; in-flight events on the previous backend complete against the previous pointer.
func Time ¶
func Time(name string) func()
Time is a convenience helper that observes the elapsed time from invocation until the returned function is called. Usage:
defer metrics.Time("search.dijkstra")()
On the no-op backend the overhead is two atomic loads + a time.Now pair (~50 ns), payable once per call site.
Types ¶
type Backend ¶
type Backend interface {
// IncCounter increments the named counter by delta.
IncCounter(name string, delta uint64)
// ObserveLatency records a single latency sample under name.
ObserveLatency(name string, d time.Duration)
}
Backend is the interface every metrics sink implements. It is intentionally tiny so the no-op default is zero overhead.
Source Files
Directories
¶
| Path | Synopsis |
|---|---|
|
Package prometheus provides a [metrics.Backend] implementation that produces Prometheus-compatible text exposition output — with no dependency on github.com/prometheus/client_golang.
|
Package prometheus provides a [metrics.Backend] implementation that produces Prometheus-compatible text exposition output — with no dependency on github.com/prometheus/client_golang. |