feat(query-api): add richer stats and retention

This commit is contained in:
William Valentin
2026-03-26 11:22:34 -07:00
parent fdfcb50e80
commit 43877a5448
10 changed files with 583 additions and 85 deletions
+101 -17
View File
@@ -16,14 +16,22 @@ type Summary struct {
RunsToday int `json:"runs_today"`
ToolCallsToday int `json:"tool_calls_today"`
ErrorsToday int `json:"errors_today"`
TokensToday int64 `json:"tokens_today"`
CostToday float64 `json:"cost_today"`
AvgDurationMS float64 `json:"avg_duration_ms"`
ByFramework map[string]FrameworkStats `json:"by_framework"`
}
type TimeseriesBucket struct {
TS time.Time `json:"ts"`
Runs int `json:"runs"`
Tools int `json:"tools"`
Errors int `json:"errors"`
TS time.Time `json:"ts"`
Runs int `json:"runs"`
Tools int `json:"tools"`
Errors int `json:"errors"`
Tokens int64 `json:"tokens"`
InputTokens int64 `json:"input_tokens"`
OutputTokens int64 `json:"output_tokens"`
Cost float64 `json:"cost"`
AvgDurationMS float64 `json:"avg_duration_ms"`
}
type TimeseriesResult struct {
@@ -36,21 +44,23 @@ func (d *DB) GetSummary(ctx context.Context) (*Summary, error) {
now := time.Now()
midnight := time.Date(now.Year(), now.Month(), now.Day(), 0, 0, 0, 0, now.Location())
// Active sessions: sessions with a session.start but no session.end (ever)
// Active sessions: sessions with a session.start but no session.end (last 7 days)
activeQ := `
SELECT COUNT(DISTINCT session_id)
FROM events
WHERE type = 'session.start'
AND session_id IS NOT NULL
AND session_id NOT IN (
SELECT DISTINCT session_id
FROM events
WHERE type = 'session.end'
AND session_id IS NOT NULL
SELECT COUNT(DISTINCT e.session_id)
FROM events e
WHERE e.type = 'session.start'
AND e.session_id IS NOT NULL
AND e.ts >= $1
AND NOT EXISTS (
SELECT 1
FROM events e2
WHERE e2.type = 'session.end'
AND e2.session_id = e.session_id
)
`
var activeSessions int
if err := d.sql.QueryRowContext(ctx, activeQ).Scan(&activeSessions); err != nil {
activeSessionsSince := time.Now().Add(-7 * 24 * time.Hour)
if err := d.sql.QueryRowContext(ctx, activeQ, activeSessionsSince).Scan(&activeSessions); err != nil {
return nil, err
}
@@ -64,6 +74,7 @@ func (d *DB) GetSummary(ctx context.Context) (*Summary, error) {
COUNT(*) FILTER (WHERE type = 'error') AS errors
FROM events
WHERE ts >= $1
AND type IN ('run.start', 'span.end', 'error')
GROUP BY source_framework
`
rows, err := d.sql.QueryContext(ctx, fwQ, midnight)
@@ -90,11 +101,30 @@ func (d *DB) GetSummary(ctx context.Context) (*Summary, error) {
return nil, err
}
// Usage stats for today (tokens, cost, avg latency)
usageQ := `
SELECT
COALESCE(SUM((payload->'payload'->'usage'->>'total_tokens')::bigint), 0),
COALESCE(SUM((payload->'payload'->'usage'->>'total_cost')::float8), 0),
COALESCE(AVG((payload->'payload'->>'duration_ms')::float8), 0)
FROM events
WHERE type = 'run.end'
AND ts >= $1
`
var tokensToday int64
var costToday, avgDurationMS float64
if err := d.sql.QueryRowContext(ctx, usageQ, midnight).Scan(&tokensToday, &costToday, &avgDurationMS); err != nil {
return nil, err
}
return &Summary{
ActiveSessions: activeSessions,
RunsToday: totalRuns,
ToolCallsToday: totalTools,
ErrorsToday: totalErrors,
TokensToday: tokensToday,
CostToday: costToday,
AvgDurationMS: avgDurationMS,
ByFramework: byFramework,
}, nil
}
@@ -104,6 +134,11 @@ type TopTool struct {
Count int `json:"count"`
}
type TopModel struct {
Name string `json:"name"`
Count int `json:"count"`
}
func (d *DB) GetTopTools(ctx context.Context, limit int) ([]TopTool, error) {
if limit <= 0 {
limit = 10
@@ -141,6 +176,43 @@ func (d *DB) GetTopTools(ctx context.Context, limit int) ([]TopTool, error) {
return out, rows.Err()
}
func (d *DB) GetTopModels(ctx context.Context, limit int) ([]TopModel, error) {
if limit <= 0 {
limit = 10
}
now := time.Now()
midnight := time.Date(now.Year(), now.Month(), now.Day(), 0, 0, 0, 0, now.Location())
q := `
SELECT
payload->'payload'->>'model' AS model_name,
COUNT(*) AS cnt
FROM events
WHERE type = 'run.end'
AND payload->'payload'->>'model' IS NOT NULL
AND payload->'payload'->>'model' <> ''
AND ts >= $1
GROUP BY model_name
ORDER BY cnt DESC
LIMIT $2
`
rows, err := d.sql.QueryContext(ctx, q, midnight, limit)
if err != nil {
return nil, err
}
defer rows.Close()
var out []TopModel
for rows.Next() {
var m TopModel
if err := rows.Scan(&m.Name, &m.Count); err != nil {
return nil, err
}
out = append(out, m)
}
return out, rows.Err()
}
func bucketForWindow(window string) string {
switch window {
case "1h":
@@ -191,9 +263,20 @@ func (d *DB) GetTimeseries(ctx context.Context, window string) (*TimeseriesResul
COUNT(*) FILTER (WHERE type = 'run.start') AS runs,
COUNT(*) FILTER (WHERE type = 'span.end'
AND payload->'attributes'->>'span_kind' = 'tool') AS tools,
COUNT(*) FILTER (WHERE type = 'error') AS errors
COUNT(*) FILTER (WHERE type = 'error') AS errors,
COALESCE(SUM((payload->'payload'->'usage'->>'total_tokens')::bigint)
FILTER (WHERE type = 'run.end'), 0) AS tokens,
COALESCE(SUM((payload->'payload'->'usage'->>'input_tokens')::bigint)
FILTER (WHERE type = 'run.end'), 0) AS input_tokens,
COALESCE(SUM((payload->'payload'->'usage'->>'output_tokens')::bigint)
FILTER (WHERE type = 'run.end'), 0) AS output_tokens,
COALESCE(SUM((payload->'payload'->'usage'->>'total_cost')::float8)
FILTER (WHERE type = 'run.end'), 0) AS cost,
COALESCE(AVG((payload->'payload'->>'duration_ms')::float8)
FILTER (WHERE type = 'run.end'), 0) AS avg_duration_ms
FROM events
WHERE ts >= $2
AND type IN ('run.start', 'run.end', 'span.end', 'error')
GROUP BY bucket_ts
ORDER BY bucket_ts ASC
`
@@ -207,7 +290,8 @@ func (d *DB) GetTimeseries(ctx context.Context, window string) (*TimeseriesResul
var series []TimeseriesBucket
for rows.Next() {
var b TimeseriesBucket
if err := rows.Scan(&b.TS, &b.Runs, &b.Tools, &b.Errors); err != nil {
if err := rows.Scan(&b.TS, &b.Runs, &b.Tools, &b.Errors,
&b.Tokens, &b.InputTokens, &b.OutputTokens, &b.Cost, &b.AvgDurationMS); err != nil {
return nil, err
}
series = append(series, b)