storage

func ConfigFromEnv ¶

func ConfigFromEnv() Config

ConfigFromEnv builds a Config from MAPING_CLICKHOUSE_DSN, falling back to the local dev default.

func DownstreamForEndpoint ¶

func DownstreamForEndpoint(
	ctx context.Context,
	conn driver.Conn,
	tenantID tenant.ID,
	service, method, route string,
	from, to time.Time,
) (DownstreamStat, error)

DownstreamForEndpoint returns the self-vs-downstream time split for the given endpoint for tenant over [from, to). method and route may be empty to aggregate across all methods / all route templates of the service. It answers "how much of this endpoint's latency is its own work vs waiting on a dependency?".

func QueryEndpointDetail ¶

func QueryEndpointDetail(
	ctx context.Context,
	conn driver.Conn,
	tenantID tenant.ID,
	service, method, route string,
	from, to time.Time,
) (EndpointDetail, error)

QueryEndpointDetail returns the full aggregate for one endpoint over [from, to): the merged sketch as a latency histogram, the per-class breakdown, the exact status-code map, and p50/p95/p99 computed in Go from the same merged sketch via QuantileFromBuckets — so the percentiles and the histogram can never disagree. Returns a zero-count EndpointDetail (no error) when the endpoint has no data in the window. Named QueryEndpointDetail (not EndpointDetail) because the return type already owns that identifier; the QueryService method is EndpointDetail, since methods have their own namespace.

func Endpoints ¶

func Endpoints(
	ctx context.Context,
	conn driver.Conn,
	tenantID tenant.ID,
	service string,
	from, to time.Time,
) ([]EndpointStat, error)

Endpoints returns one aggregate EndpointStat per (method, route_template) of service for tenant over [from, to), ordered by Count descending.

func ErrorClassesForEndpoint ¶

func ErrorClassesForEndpoint(
	ctx context.Context,
	conn driver.Conn,
	tenantID tenant.ID,
	service, method, route string,
	from, to time.Time,
) ([]ErrorClassStat, error)

ErrorClassesForEndpoint returns the top error-class labels attached to the given endpoint for tenant over [from, to), ordered by request count descending and capped at maxErrorClassesPerEndpoint. method and route may be empty to aggregate across all methods / all route templates of the service.

func ExemplarsForEndpoint ¶

func ExemplarsForEndpoint(
	ctx context.Context,
	conn driver.Conn,
	tenantID tenant.ID,
	service, method, route string,
	from, to time.Time,
) ([]ExemplarRow, error)

ExemplarsForEndpoint returns up to maxExemplarsPerEndpoint real-request breadcrumbs for the given endpoint for tenant over [from, to), ordered by latency descending, so a user can pivot from a p99 / error spike to an actual request to open in a tracing tool or logs. method and route may be empty to aggregate across all methods / all route templates of the service.

It reads the RAW summaries tier directly (NOT via selectTier): exemplars live only in the finest tier under its short TTL, so this query is only meaningful for recent windows within the raw retention. Older windows return no exemplars even though the RED aggregates remain available on the rollups.

func InstanceResourcesForService ¶

func InstanceResourcesForService(
	ctx context.Context,
	conn driver.Conn,
	tenantID tenant.ID,
	service string,
	from, to time.Time,
) ([]InstanceResourceStat, error)

InstanceResourcesForService returns one InstanceResourceStat per instance of the service for tenant over [from, to), ordered by instance. It is the USE side of the debug view: saturation gauges to sit next to the RED metrics so a latency rise can be attributed to GC or goroutine growth without a release.

func InstancesForEndpoint ¶

func InstancesForEndpoint(
	ctx context.Context,
	conn driver.Conn,
	tenantID tenant.ID,
	service, method, route string,
	from, to time.Time,
) ([]InstanceStat, error)

InstancesForEndpoint returns one InstanceStat per instance (replica) serving the given endpoint for tenant over [from, to), ordered by instance. method and route may be empty to aggregate across all methods / all route templates of the service. This is the flagship outlier query: it answers whether a degradation is confined to one replica or is fleet-wide. Percentiles use the frozen DDSketch convention; the error rate is (4xx+5xx+no_status)/total.

func MemoryTrendForService ¶

func MemoryTrendForService(
	ctx context.Context,
	conn driver.Conn,
	tenantID tenant.ID,
	service string,
	from, to time.Time,
	step time.Duration,
) ([]MemoryTrendPoint, error)

MemoryTrendForService returns the per-bucket fleet memory trend for a service over [from, to), bucketed by step, ordered by time. It reads the same [from, to, step] window the detail-page timeline uses so heap and latency line up. step is floored to one second; the caller passes the timeline step.

func NoStatusReasonsForEndpoint ¶

func NoStatusReasonsForEndpoint(
	ctx context.Context,
	conn driver.Conn,
	tenantID tenant.ID,
	service, method, route string,
	from, to time.Time,
) ([]NoStatusReasonStat, error)

NoStatusReasonsForEndpoint returns the NO_STATUS reason breakdown for the given endpoint for tenant over [from, to), ordered by reason. method and route may be empty to aggregate across all methods / all route templates of the service.

func PerformanceStats ¶

func PerformanceStats(
	ctx context.Context,
	conn driver.Conn,
	tenantID tenant.ID,
	from, to time.Time,
) (PerformanceStat, error)

PerformanceStats measures, for tenant over [from, to), the total requests the stored summaries represent (sum of the per-summary count) and the number of shipped summary rows, both from the raw summaries tier. It then estimates the on-disk size of those summaries from the measured average summary-row size (system.parts, best-effort — a failure falls back to a constant rather than failing the page). Their ratio is the compression the page reports; the caller projects the raw-event-pipeline size from Requests and a documented per-event assumption (mAPI-ng stores no raw events, so that side cannot be measured).

func NewQueryService ¶

func NewQueryService(conn driver.Conn) *QueryService

NewQueryService builds a QueryService over an open connection.

func NewRow ¶

func NewRow(
	tenantID tenant.ID,
	service, instance, method, routeTemplate, statusClass string,
	windowStart, windowEnd time.Time,
	count, sumDurationNs, reqBytes, respBytes uint64,
	sketch map[int32]uint64,
	statusCodes map[uint32]uint64,
	deployVersion, deployID, environment, region string,
	instanceStart time.Time,
	maxDurationNs uint64,
	exemplars []Exemplar,
	errorClasses map[string]uint64,
	noStatusReasons map[uint32]uint64,
	sumDownstreamNs uint64,
) Row

NewRow builds a Row from an already-mapped sketch and status-code maps, sorting both by key so the ClickHouse map column ordering is deterministic across inserts.

func Services ¶

func Services(
	ctx context.Context,
	conn driver.Conn,
	tenantID tenant.ID,
	from, to time.Time,
) ([]ServiceStat, error)

Services returns one aggregate ServiceStat per service for tenant over [from, to), ordered by Count descending. Percentiles use the frozen DDSketch convention; the error rate is the CONTEXT (4xx+5xx+no_status)/total.

func Usage ¶ added in v0.11.0

func Usage(ctx context.Context, conn driver.Conn, tenantID tenant.ID, now time.Time) (TenantUsage, error)

Usage assembles the operator volumetry for tenantID as of now. It runs three cheap aggregates: liveness from the 1d tier (longest retention, so a churned tenant still shows its last-ingest date), cardinality and 30-day requests from the 1m tier, and the disk estimate from the raw tier via PerformanceStats. now is a parameter (not time.Now()) so the windows are deterministic in tests.

func SeriesOverTime ¶

func SeriesOverTime(
	ctx context.Context,
	conn driver.Conn,
	tenantID tenant.ID,
	service, method, route string,
	from, to time.Time,
	step time.Duration,
) ([]TimePoint, error)

SeriesOverTime returns the per-time-bucket RED distribution for a series over [from, to), bucketed by step. method and route may be empty to aggregate across all methods / all route templates of the service. Percentiles use the frozen DDSketch convention above; the same math runs client-side, which is the whole correctness point.

func VersionsForEndpoint ¶

func VersionsForEndpoint(
	ctx context.Context,
	conn driver.Conn,
	tenantID tenant.ID,
	service, method, route string,
	from, to time.Time,
) ([]VersionStat, error)

VersionsForEndpoint returns one VersionStat per deploy_version serving the given endpoint for tenant over [from, to), ordered by deploy_version. method and route may be empty to aggregate across all methods / all route templates of the service. It answers whether a release regressed an endpoint by putting each version's RED metrics side by side. Percentiles use the frozen DDSketch convention; the error rate is (4xx+5xx+no_status)/total.

func NewWriter ¶

func NewWriter(ctx context.Context, cfg Config, log *slog.Logger) (*Writer, error)

NewWriter opens a ClickHouse connection from cfg and starts the batcher goroutine. The caller owns the returned Writer and must call Close.