i/Aether
1
0
Fork 0
mirror of https://github.com/fawney19/Aether.git synced 2026-10-08 18:37:46 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat: improve admin performance analysis

Browse Source
This commit is contained in:
fawney19
2026-05-06 02:01:32 +08:00
parent f1358dd845
commit cf27773582
9 changed files with 833 additions and 205 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/aether-data/src/repository/usage/memory.rs
+66 -4
View File
@@ -618,6 +618,36 @@ fn usage_matches_provider_performance_query(
{
return None;
}
if let Some(provider_id) = query.provider_id.as_deref() {
if item.provider_id.as_deref() != Some(provider_id) {
return None;
}
}
if let Some(model) = query.model.as_deref() {
if item.model != model {
return None;
}
}
if let Some(api_format) = query.api_format.as_deref() {
if item.api_format.as_deref() != Some(api_format) {
return None;
}
}
if let Some(endpoint_kind) = query.endpoint_kind.as_deref() {
if item.endpoint_kind.as_deref() != Some(endpoint_kind) {
return None;
}
}
if let Some(is_stream) = query.is_stream {
if item.is_stream != is_stream {
return None;
}
}
if let Some(has_format_conversion) = query.has_format_conversion {
if item.has_format_conversion != has_format_conversion {
return None;
}
}
usage_provider_performance_identity(item)
}
@@ -1703,12 +1733,19 @@ impl UsageReadRepository for InMemoryUsageReadRepository {
response_time_sample_count: u64,
response_times: Vec<u64>,
first_byte_times: Vec<u64>,
slow_request_count: u64,
}
impl ProviderPerformanceBucket {
fn add(&mut self, item: &StoredRequestUsageAudit) {
fn add(&mut self, item: &StoredRequestUsageAudit, slow_threshold_ms: u64) {
self.request_count = self.request_count.saturating_add(1);
self.output_tokens = self.output_tokens.saturating_add(item.output_tokens);
if item
.response_time_ms
.is_some_and(|value| value >= slow_threshold_ms)
{
self.slow_request_count = self.slow_request_count.saturating_add(1);
}
if !usage_is_success(item) {
return;
}
@@ -1771,14 +1808,16 @@ impl UsageReadRepository for InMemoryUsageReadRepository {
else {
continue;
};
summary_bucket.add(item);
summary_bucket.add(item, query.slow_threshold_ms);
let bucket = grouped.entry(provider_id).or_default();
if bucket.provider.is_empty() {
bucket.provider = provider;
}
bucket.add(item);
bucket.add(item, query.slow_threshold_ms);
}
let mut summary_response_times = summary_bucket.response_times.clone();
let mut summary_first_byte_times = summary_bucket.first_byte_times.clone();
let summary = StoredUsageProviderPerformanceSummary {
request_count: summary_bucket.request_count,
success_count: summary_bucket.success_count,
@@ -1794,14 +1833,25 @@ impl UsageReadRepository for InMemoryUsageReadRepository {
summary_bucket.response_time_ms_sum,
summary_bucket.response_time_sample_count,
),
p90_response_time_ms: usage_percentile_cont(&mut summary_response_times, 0.9),
p99_response_time_ms: usage_percentile_cont(&mut summary_response_times, 0.99),
p90_first_byte_time_ms: usage_percentile_cont(&mut summary_first_byte_times, 0.9),
p99_first_byte_time_ms: usage_percentile_cont(&mut summary_first_byte_times, 0.99),
tps_sample_count: summary_bucket.tps_sample_count,
response_time_sample_count: summary_bucket.response_time_sample_count,
first_byte_sample_count: summary_bucket.first_byte_sample_count,
slow_request_count: summary_bucket.slow_request_count,
};
let mut providers = grouped
.into_iter()
.map(|(provider_id, mut bucket)| {
let p90_response_time_ms = usage_percentile_cont(&mut bucket.response_times, 0.9);
let p99_response_time_ms = usage_percentile_cont(&mut bucket.response_times, 0.99);
let p90_first_byte_time_ms =
usage_percentile_cont(&mut bucket.first_byte_times, 0.9);
let p99_first_byte_time_ms =
usage_percentile_cont(&mut bucket.first_byte_times, 0.99);
StoredUsageProviderPerformanceProviderRow {
provider_id,
provider: bucket.provider,
@@ -1821,9 +1871,13 @@ impl UsageReadRepository for InMemoryUsageReadRepository {
bucket.response_time_sample_count,
),
p90_response_time_ms,
p99_response_time_ms,
p90_first_byte_time_ms,
p99_first_byte_time_ms,
tps_sample_count: bucket.tps_sample_count,
response_time_sample_count: bucket.response_time_sample_count,
first_byte_sample_count: bucket.first_byte_sample_count,
slow_request_count: bucket.slow_request_count,
}
})
.collect::<Vec<_>>();
@@ -1860,7 +1914,7 @@ impl UsageReadRepository for InMemoryUsageReadRepository {
if bucket.provider.is_empty() {
bucket.provider = provider;
}
bucket.add(item);
bucket.add(item, query.slow_threshold_ms);
}
let timeline = timeline_grouped
@@ -1885,6 +1939,7 @@ impl UsageReadRepository for InMemoryUsageReadRepository {
bucket.response_time_ms_sum,
bucket.response_time_sample_count,
),
slow_request_count: bucket.slow_request_count,
},
)
.collect();
@@ -4846,6 +4901,13 @@ mod tests {
granularity: UsageTimeSeriesGranularity::Hour,
tz_offset_minutes: 0,
limit: 1,
provider_id: None,
model: None,
api_format: None,
endpoint_kind: None,
is_stream: None,
has_format_conversion: None,
slow_threshold_ms: 10_000,
})
.await
.expect("provider performance should summarize");