stats: per-bucket turns-by-model chart
each turn_stats row already records the model; roll it up per bucket so the /stats page can show which model ran when. model choice greatly affects token cost, so the new stacked-bar chart sits right under the cost chart for eyeball correlation across the window. Snapshot gains a sorted `models` series list; each Bucket carries a `model_counts` map.
This commit is contained in:
müde
parent
24b10becc9
commit
f13c3dff8f
3 changed files with 71 additions and 9 deletions
|
|
@ -80,6 +80,7 @@
|
|||
<div class="card wide"><h3>turn duration (ms) — p50 / p95 / avg</h3><div class="chart-wrap"><canvas id="chart-duration"></canvas></div></div>
|
||||
<div class="card wide"><h3>context tokens (last inference per turn) — avg / max</h3><div class="chart-wrap"><canvas id="chart-ctx"></canvas></div></div>
|
||||
<div class="card wide"><h3>token cost per bucket (sum across inferences)</h3><div class="chart-wrap"><canvas id="chart-cost"></canvas></div></div>
|
||||
<div class="card wide"><h3>turns by model per bucket — model drives token cost</h3><div class="chart-wrap"><canvas id="chart-model"></canvas></div></div>
|
||||
<div class="card"><h3>top tools</h3><div class="chart-wrap"><canvas id="chart-tools"></canvas></div></div>
|
||||
<div class="card"><h3>wake source mix</h3><div class="chart-wrap"><canvas id="chart-wake"></canvas></div></div>
|
||||
<div class="card"><h3>result mix</h3><div class="chart-wrap"><canvas id="chart-result"></canvas></div></div>
|
||||
|
|
|
|||
|
|
@ -226,6 +226,34 @@
|
|||
});
|
||||
}
|
||||
|
||||
function renderModelChart(s) {
|
||||
const id = 'chart-model';
|
||||
destroy(id);
|
||||
const models = s.models || [];
|
||||
if (!models.length) { paintEmpty(id, 'no turns in window'); return; }
|
||||
const labels = s.buckets.map((b) => bucketLabel(b.ts, s.bucket_seconds));
|
||||
// One stacked series per model. Model choice drives token cost,
|
||||
// so this lines up against the cost chart above it.
|
||||
const datasets = models.map((m, i) => ({
|
||||
label: m,
|
||||
data: s.buckets.map((b) => (b.model_counts && b.model_counts[m]) || 0),
|
||||
backgroundColor: wheel[i % wheel.length],
|
||||
}));
|
||||
charts[id] = new Chart(document.getElementById(id), {
|
||||
type: 'bar',
|
||||
data: { labels, datasets },
|
||||
options: {
|
||||
responsive: true, maintainAspectRatio: false,
|
||||
plugins: { legend: { position: 'top', labels: { boxWidth: 12 } } },
|
||||
scales: {
|
||||
x: { stacked: true, grid: { color: palette.border } },
|
||||
y: { stacked: true, beginAtZero: true,
|
||||
grid: { color: palette.border }, ticks: { precision: 0 } },
|
||||
},
|
||||
},
|
||||
});
|
||||
}
|
||||
|
||||
function renderKeyCount(canvasId, items, emptyMsg) {
|
||||
destroy(canvasId);
|
||||
if (!items || items.length === 0) {
|
||||
|
|
@ -252,6 +280,7 @@
|
|||
paintEmpty('chart-duration', 'no turns in window');
|
||||
paintEmpty('chart-ctx', 'no turns in window');
|
||||
paintEmpty('chart-cost', 'no turns in window');
|
||||
paintEmpty('chart-model', 'no turns in window');
|
||||
paintEmpty('chart-tools', 'no tool calls');
|
||||
paintEmpty('chart-wake', 'no wakes');
|
||||
paintEmpty('chart-result', 'no results');
|
||||
|
|
@ -261,6 +290,7 @@
|
|||
renderDurationChart(s);
|
||||
renderCtxChart(s);
|
||||
renderCostChart(s);
|
||||
renderModelChart(s);
|
||||
renderKeyCount('chart-tools', s.tool_breakdown, 'no tool calls');
|
||||
renderKeyCount('chart-wake', s.wake_mix, 'no wakes');
|
||||
renderKeyCount('chart-result', s.result_mix, 'no results');
|
||||
|
|
|
|||
|
|
@ -7,7 +7,7 @@
|
|||
//! propagating — the stats page is decorative, not authoritative, and
|
||||
//! a missing db on a brand-new agent shouldn't 500 the route.
|
||||
|
||||
use std::collections::HashMap;
|
||||
use std::collections::{HashMap, HashSet};
|
||||
use std::path::{Path, PathBuf};
|
||||
|
||||
use anyhow::{Context, Result};
|
||||
|
|
@ -73,6 +73,10 @@ pub struct Snapshot {
|
|||
pub tool_breakdown: Vec<KeyCount>,
|
||||
pub wake_mix: Vec<KeyCount>,
|
||||
pub result_mix: Vec<KeyCount>,
|
||||
/// Distinct models seen in the window, sorted. Each bucket's
|
||||
/// `model_counts` keys into this set; the stats page uses it as
|
||||
/// the stacked-bar series list (stable order + colours).
|
||||
pub models: Vec<String>,
|
||||
/// Across-window p50 / p95 / avg of `duration_ms`. Same numbers
|
||||
/// as the per-bucket fields but aggregated over the whole window
|
||||
/// for the headline summary chips.
|
||||
|
|
@ -97,6 +101,10 @@ pub struct Bucket {
|
|||
/// size at turn-end — useful for spotting drift toward compaction).
|
||||
pub avg_ctx_tokens: f64,
|
||||
pub max_ctx_tokens: u64,
|
||||
/// Turn count per model in this bucket. Model choice greatly
|
||||
/// affects token cost, so this lets the operator line model usage
|
||||
/// up against the cost series over time.
|
||||
pub model_counts: HashMap<String, u64>,
|
||||
}
|
||||
|
||||
#[derive(Debug, Serialize)]
|
||||
|
|
@ -142,6 +150,7 @@ fn empty_snapshot(window: Window) -> Snapshot {
|
|||
tool_breakdown: Vec::new(),
|
||||
wake_mix: Vec::new(),
|
||||
result_mix: Vec::new(),
|
||||
models: Vec::new(),
|
||||
duration_summary: DurationSummary::default(),
|
||||
}
|
||||
}
|
||||
|
|
@ -162,7 +171,7 @@ fn snapshot(path: &Path, window: Window) -> Result<Snapshot> {
|
|||
cache_read_input_tokens, cache_creation_input_tokens,
|
||||
last_input_tokens,
|
||||
tool_call_breakdown_json,
|
||||
wake_from, result_kind
|
||||
wake_from, result_kind, model
|
||||
FROM turn_stats
|
||||
WHERE started_at >= ?1
|
||||
ORDER BY started_at ASC",
|
||||
|
|
@ -179,6 +188,7 @@ fn snapshot(path: &Path, window: Window) -> Result<Snapshot> {
|
|||
tool_breakdown_json: row.get::<_, Option<String>>(7)?,
|
||||
wake_from: row.get::<_, String>(8)?,
|
||||
result_kind: row.get::<_, String>(9)?,
|
||||
model: row.get::<_, String>(10)?,
|
||||
})
|
||||
})?;
|
||||
|
||||
|
|
@ -186,6 +196,7 @@ fn snapshot(path: &Path, window: Window) -> Result<Snapshot> {
|
|||
let mut tool_totals: HashMap<String, u64> = HashMap::new();
|
||||
let mut wake_totals: HashMap<String, u64> = HashMap::new();
|
||||
let mut result_totals: HashMap<String, u64> = HashMap::new();
|
||||
let mut model_set: HashSet<String> = HashSet::new();
|
||||
let mut all_durations: Vec<i64> = Vec::new();
|
||||
let mut turn_count: u64 = 0;
|
||||
|
||||
|
|
@ -206,10 +217,12 @@ fn snapshot(path: &Path, window: Window) -> Result<Snapshot> {
|
|||
.saturating_add(r.cache_creation_input_tokens);
|
||||
acc.ctx_sum = acc.ctx_sum.saturating_add(r.last_input_tokens);
|
||||
acc.ctx_max = acc.ctx_max.max(r.last_input_tokens);
|
||||
*acc.model_counts.entry(r.model.clone()).or_insert(0) += 1;
|
||||
|
||||
all_durations.push(r.duration_ms.max(0));
|
||||
*wake_totals.entry(r.wake_from).or_insert(0) += 1;
|
||||
*result_totals.entry(r.result_kind).or_insert(0) += 1;
|
||||
model_set.insert(r.model);
|
||||
|
||||
if let Some(json) = r.tool_breakdown_json
|
||||
&& let Ok(map) = serde_json::from_str::<HashMap<String, u64>>(&json)
|
||||
|
|
@ -222,6 +235,8 @@ fn snapshot(path: &Path, window: Window) -> Result<Snapshot> {
|
|||
|
||||
let buckets = fill_buckets(from, now, bucket_secs, &by_bucket);
|
||||
let duration_summary = summarize_durations(&mut all_durations);
|
||||
let mut models: Vec<String> = model_set.into_iter().collect();
|
||||
models.sort_unstable();
|
||||
|
||||
Ok(Snapshot {
|
||||
window: window.label(),
|
||||
|
|
@ -233,6 +248,7 @@ fn snapshot(path: &Path, window: Window) -> Result<Snapshot> {
|
|||
tool_breakdown: top_n(tool_totals, 10),
|
||||
wake_mix: top_n(wake_totals, 20),
|
||||
result_mix: top_n(result_totals, 20),
|
||||
models,
|
||||
duration_summary,
|
||||
})
|
||||
}
|
||||
|
|
@ -248,6 +264,7 @@ struct Row {
|
|||
tool_breakdown_json: Option<String>,
|
||||
wake_from: String,
|
||||
result_kind: String,
|
||||
model: String,
|
||||
}
|
||||
|
||||
#[derive(Default)]
|
||||
|
|
@ -260,6 +277,7 @@ struct BucketAcc {
|
|||
cache_creation_input_tokens: u64,
|
||||
ctx_sum: u64,
|
||||
ctx_max: u64,
|
||||
model_counts: HashMap<String, u64>,
|
||||
}
|
||||
|
||||
fn fill_buckets(
|
||||
|
|
@ -307,6 +325,7 @@ fn fill_buckets(
|
|||
cache_creation_input_tokens: acc.cache_creation_input_tokens,
|
||||
avg_ctx_tokens: avg_ctx,
|
||||
max_ctx_tokens: acc.ctx_max,
|
||||
model_counts: acc.model_counts.clone(),
|
||||
}
|
||||
} else {
|
||||
Bucket {
|
||||
|
|
@ -321,6 +340,7 @@ fn fill_buckets(
|
|||
cache_creation_input_tokens: 0,
|
||||
avg_ctx_tokens: 0.0,
|
||||
max_ctx_tokens: 0,
|
||||
model_counts: HashMap::new(),
|
||||
}
|
||||
};
|
||||
out.push(bucket);
|
||||
|
|
@ -392,7 +412,7 @@ mod tests {
|
|||
std::env::temp_dir().join(format!("hyperhive-stats-test-{pid}-{n}.sqlite"))
|
||||
}
|
||||
|
||||
fn seed_db(path: &Path, rows: &[(i64, i64, &str, &str, &str)]) {
|
||||
fn seed_db(path: &Path, rows: &[(i64, i64, &str, &str, &str, &str)]) {
|
||||
let conn = Connection::open(path).unwrap();
|
||||
conn.execute_batch(
|
||||
"CREATE TABLE turn_stats (
|
||||
|
|
@ -419,13 +439,13 @@ mod tests {
|
|||
);",
|
||||
)
|
||||
.unwrap();
|
||||
for (started, dur, wake, result, tools_json) in rows {
|
||||
for (started, dur, model, wake, result, tools_json) in rows {
|
||||
conn.execute(
|
||||
"INSERT INTO turn_stats
|
||||
(started_at, ended_at, duration_ms, model, wake_from,
|
||||
last_input_tokens, tool_call_breakdown_json, result_kind)
|
||||
VALUES (?1, ?2, ?3, 'm', ?4, 1000, ?5, ?6)",
|
||||
params![started, started + dur / 1000, dur, wake, tools_json, result],
|
||||
VALUES (?1, ?2, ?3, ?4, ?5, 1000, ?6, ?7)",
|
||||
params![started, started + dur / 1000, dur, model, wake, tools_json, result],
|
||||
)
|
||||
.unwrap();
|
||||
}
|
||||
|
|
@ -439,9 +459,9 @@ mod tests {
|
|||
seed_db(
|
||||
&db,
|
||||
&[
|
||||
(now - 600, 5_000, "recv", "ok", r#"{"Read":2,"Bash":1}"#),
|
||||
(now - 300, 10_000, "recv", "ok", r#"{"Read":3}"#),
|
||||
(now - 100, 20_000, "operator", "failed", "{}"),
|
||||
(now - 600, 5_000, "opus", "recv", "ok", r#"{"Read":2,"Bash":1}"#),
|
||||
(now - 300, 10_000, "opus", "recv", "ok", r#"{"Read":3}"#),
|
||||
(now - 100, 20_000, "sonnet", "operator", "failed", "{}"),
|
||||
],
|
||||
);
|
||||
let s = snapshot(&db, Window::Day).unwrap();
|
||||
|
|
@ -469,6 +489,17 @@ mod tests {
|
|||
.collect();
|
||||
assert_eq!(result_map.get("ok").copied(), Some(2));
|
||||
assert_eq!(result_map.get("failed").copied(), Some(1));
|
||||
// Model breakdown: 2 opus + 1 sonnet, all in the same hour
|
||||
// bucket given the 24h window.
|
||||
assert_eq!(s.models, vec!["opus".to_string(), "sonnet".to_string()]);
|
||||
let mut model_totals: HashMap<String, u64> = HashMap::new();
|
||||
for b in &s.buckets {
|
||||
for (k, v) in &b.model_counts {
|
||||
*model_totals.entry(k.clone()).or_insert(0) += v;
|
||||
}
|
||||
}
|
||||
assert_eq!(model_totals.get("opus").copied(), Some(2));
|
||||
assert_eq!(model_totals.get("sonnet").copied(), Some(1));
|
||||
// Durations: [5000, 10000, 20000] → avg ≈ 11666.67, p50 = 10000, p95 ~ 20000
|
||||
assert!((s.duration_summary.avg_ms - 11_666.666_666_666_666).abs() < 1.0);
|
||||
assert!((s.duration_summary.p50_ms - 10_000.0).abs() < 1.0);
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue