will/agentmon
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(metrics): surface tool-span latency in stats and dashboard

Browse Source 
Tool spans already carry duration_ms and status, but the metrics layer
only counted them. Expose that data:

- GetTopTools now returns avg/p95 duration and error count per tool.
- Timeseries buckets gain tool_avg_ms / tool_p95_ms (filtered
  percentile_cont over tool spans).
- Dashboard Top Tools shows avg latency per tool; the Latency panel,
  previously always empty (it read run-level duration that is never
  emitted), now plots real tool-span latency (min/avg/p95).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
William Valentin
2026-06-23 11:16:23 -07:00
parent c44e7fe72e
commit 5014d89258
2 changed files with 48 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files
internal/store/postgres/stats.go
+23 -6
View File
@@ -33,6 +33,8 @@ type TimeseriesBucket struct {
OutputTokens int64 `json:"output_tokens"`
Cost float64 `json:"cost"`
AvgDurationMS float64 `json:"avg_duration_ms"`
ToolAvgMS float64 `json:"tool_avg_ms"`
ToolP95MS float64 `json:"tool_p95_ms"`
}
type TimeseriesResult struct {
@@ -157,8 +159,11 @@ func (d *DB) GetSummary(ctx context.Context) (*Summary, error) {
}
type TopTool struct {
Name string `json:"name"`
Count int `json:"count"`
Name string `json:"name"`
Count int `json:"count"`
AvgMS float64 `json:"avg_ms"`
P95MS float64 `json:"p95_ms"`
Errors int `json:"errors"`
}
type TopModel struct {
@@ -176,7 +181,11 @@ func (d *DB) GetTopTools(ctx context.Context, limit int) ([]TopTool, error) {
q := `
SELECT
payload->'attributes'->>'name' AS tool_name,
COUNT(*) AS cnt
COUNT(*) AS cnt,
COALESCE(AVG((payload->'payload'->>'duration_ms')::float8), 0) AS avg_ms,
COALESCE(percentile_cont(0.95) WITHIN GROUP (
ORDER BY (payload->'payload'->>'duration_ms')::float8), 0) AS p95_ms,
COUNT(*) FILTER (WHERE payload->'payload'->>'status' = 'error') AS errors
FROM events
WHERE type = 'span.end'
AND payload->'attributes'->>'span_kind' = 'tool'
@@ -195,7 +204,7 @@ func (d *DB) GetTopTools(ctx context.Context, limit int) ([]TopTool, error) {
var out []TopTool
for rows.Next() {
var t TopTool
if err := rows.Scan(&t.Name, &t.Count); err != nil {
if err := rows.Scan(&t.Name, &t.Count, &t.AvgMS, &t.P95MS, &t.Errors); err != nil {
return nil, err
}
out = append(out, t)
@@ -300,7 +309,14 @@ func (d *DB) GetTimeseries(ctx context.Context, window string) (*TimeseriesResul
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
FILTER (WHERE type = 'run.end'), 0) AS avg_duration_ms,
COALESCE(AVG((payload->'payload'->>'duration_ms')::float8)
FILTER (WHERE type = 'span.end'
AND payload->'attributes'->>'span_kind' = 'tool'), 0) AS tool_avg_ms,
COALESCE(percentile_cont(0.95) WITHIN GROUP (
ORDER BY (payload->'payload'->>'duration_ms')::float8)
FILTER (WHERE type = 'span.end'
AND payload->'attributes'->>'span_kind' = 'tool'), 0) AS tool_p95_ms
FROM events
WHERE ts >= $2
AND type IN ('run.start', 'run.end', 'span.end', 'error')
@@ -318,7 +334,8 @@ func (d *DB) GetTimeseries(ctx context.Context, window string) (*TimeseriesResul
for rows.Next() {
var b TimeseriesBucket
if err := rows.Scan(&b.TS, &b.Runs, &b.Tools, &b.Errors,
&b.Tokens, &b.InputTokens, &b.OutputTokens, &b.Cost, &b.AvgDurationMS); err != nil {
&b.Tokens, &b.InputTokens, &b.OutputTokens, &b.Cost, &b.AvgDurationMS,
&b.ToolAvgMS, &b.ToolP95MS); err != nil {
return nil, err
}
series = append(series, b)