fix(usage): 修复缓存创建 token 展示并在总量中计入缓存组件

- gateway/admin 用量载荷在 cache_creation_input_tokens 为 0 时回填 ephemeral_5m/1h 合计
- 标准化用量写入时将 cache_creation/cache_read token 计入 total_tokens
- 前端列表与轮询同步新增分类字段,修复缓存 token 列显示
This commit is contained in:
fawney19
2026-04-19 02:46:28 +08:00
parent e82c9c5104
commit 2f487fda66
11 changed files with 316 additions and 19 deletions

View File

@@ -457,6 +457,7 @@ pub fn admin_usage_record_json(
let output_price_per_1m = item.settlement_output_price_per_1m();
let cache_creation_price_per_1m = item.settlement_cache_creation_price_per_1m();
let cache_read_price_per_1m = item.settlement_cache_read_price_per_1m();
let cache_creation_input_tokens = admin_usage_cache_creation_tokens(item);
let is_free_tier = item.settlement_is_free_tier();
let user = item
.user_id
@@ -485,7 +486,7 @@ pub fn admin_usage_record_json(
"input_tokens": item.input_tokens,
"effective_input_tokens": admin_usage_effective_input_tokens(item),
"output_tokens": item.output_tokens,
"cache_creation_input_tokens": item.cache_creation_input_tokens,
"cache_creation_input_tokens": cache_creation_input_tokens,
"cache_creation_ephemeral_5m_input_tokens": item.cache_creation_ephemeral_5m_input_tokens,
"cache_creation_ephemeral_1h_input_tokens": item.cache_creation_ephemeral_1h_input_tokens,
"cache_read_input_tokens": item.cache_read_input_tokens,
@@ -1418,13 +1419,14 @@ pub fn build_admin_usage_active_requests_response(
let provider_key_name = admin_usage_provider_key_name(item, provider_key_names);
let api_key_name =
admin_usage_api_key_name(item, api_key_names, auth_api_key_reader_available);
let cache_creation_input_tokens = admin_usage_cache_creation_tokens(item);
let mut value = json!({
"id": item.id,
"status": item.status,
"input_tokens": item.input_tokens,
"effective_input_tokens": admin_usage_effective_input_tokens(item),
"output_tokens": item.output_tokens,
"cache_creation_input_tokens": item.cache_creation_input_tokens,
"cache_creation_input_tokens": cache_creation_input_tokens,
"cache_creation_ephemeral_5m_input_tokens": item.cache_creation_ephemeral_5m_input_tokens,
"cache_creation_ephemeral_1h_input_tokens": item.cache_creation_ephemeral_1h_input_tokens,
"cache_read_input_tokens": item.cache_read_input_tokens,
@@ -1923,6 +1925,30 @@ mod tests {
assert_eq!(payload["cache_read_price_per_1m"], 0.30);
}
#[test]
fn admin_usage_record_rehydrates_cache_creation_total_from_classified_fields() {
let item = StoredRequestUsageAudit {
cache_creation_input_tokens: 0,
cache_creation_ephemeral_5m_input_tokens: 12,
cache_creation_ephemeral_1h_input_tokens: 8,
..sample_usage("completed", Some(200), None)
};
let payload = admin_usage_record_json(
&item,
&BTreeMap::new(),
&BTreeMap::new(),
false,
false,
Some("primary"),
);
assert_eq!(payload["cache_creation_input_tokens"], 20);
assert_eq!(payload["cache_creation_ephemeral_5m_input_tokens"], 12);
assert_eq!(payload["cache_creation_ephemeral_1h_input_tokens"], 8);
assert_eq!(payload["total_tokens"], 50);
}
#[test]
fn detail_payload_exposes_typed_settlement_section() {
let item = StoredRequestUsageAudit {

View File

@@ -1366,10 +1366,7 @@ fn apply_standardized_usage(
if usage.cache_read_tokens > 0 {
data.cache_read_input_tokens = Some(usage.cache_read_tokens as u64);
}
let total_tokens = usage
.input_tokens
.saturating_add(usage.output_tokens)
.max(0) as u64;
let total_tokens = standardized_usage_total_tokens(&usage);
if total_tokens > 0 {
data.total_tokens = Some(total_tokens);
}
@@ -1396,15 +1393,23 @@ fn apply_standardized_usage_seed(usage: &StandardizedUsage, data: &mut UsageEven
if usage.cache_read_tokens > 0 {
data.cache_read_input_tokens = Some(usage.cache_read_tokens as u64);
}
let total_tokens = usage
.input_tokens
.saturating_add(usage.output_tokens)
.max(0) as u64;
let total_tokens = standardized_usage_total_tokens(usage);
if total_tokens > 0 {
data.total_tokens = Some(total_tokens);
}
}
fn standardized_usage_total_tokens(usage: &StandardizedUsage) -> u64 {
positive_usage_component(usage.input_tokens)
.saturating_add(positive_usage_component(usage.output_tokens))
.saturating_add(positive_usage_component(usage.cache_creation_tokens))
.saturating_add(positive_usage_component(usage.cache_read_tokens))
}
fn positive_usage_component(value: i64) -> u64 {
value.max(0) as u64
}
fn headers_to_json(headers: &BTreeMap<String, String>) -> Value {
Value::Object(Map::from_iter(headers.iter().map(|(key, value)| {
(key.clone(), Value::String(mask_header_value(key, value)))
@@ -1661,10 +1666,20 @@ fn extract_token_counts_from_json(value: &Value) -> Option<(u64, u64, u64)> {
.or_else(|| usage.get("completion_tokens"))
.and_then(Value::as_u64)
.unwrap_or_default();
let total = usage
let raw_total = usage
.get("total_tokens")
.and_then(Value::as_u64)
.unwrap_or(input + output);
let cache_creation = extract_cache_creation_tokens_from_usage_object(usage);
let cache_read = extract_cache_read_tokens_from_usage_object(usage);
let total = if cache_creation > 0 || cache_read > 0 {
input
.saturating_add(output)
.saturating_add(cache_creation)
.saturating_add(cache_read)
} else {
raw_total
};
return Some((input, output, total));
}
@@ -1677,10 +1692,19 @@ fn extract_token_counts_from_json(value: &Value) -> Option<(u64, u64, u64)> {
.get("candidatesTokenCount")
.and_then(Value::as_u64)
.unwrap_or_default();
let total = usage
let raw_total = usage
.get("totalTokenCount")
.and_then(Value::as_u64)
.unwrap_or(input + output);
let cache_read = usage
.get("cachedContentTokenCount")
.and_then(Value::as_u64)
.unwrap_or_default();
let total = if cache_read > 0 {
input.saturating_add(output).saturating_add(cache_read)
} else {
raw_total
};
return Some((input, output, total));
}
@@ -1703,6 +1727,59 @@ fn extract_token_counts_from_json(value: &Value) -> Option<(u64, u64, u64)> {
None
}
fn extract_cache_creation_tokens_from_usage_object(usage: &serde_json::Map<String, Value>) -> u64 {
let direct = usage
.get("cache_creation_input_tokens")
.and_then(Value::as_u64)
.unwrap_or_default();
if direct > 0 {
return direct;
}
let breakdown = usage
.get("cache_creation")
.and_then(Value::as_object)
.map(|cache_creation| {
cache_creation
.get("ephemeral_5m_input_tokens")
.and_then(Value::as_u64)
.unwrap_or_default()
.saturating_add(
cache_creation
.get("ephemeral_1h_input_tokens")
.and_then(Value::as_u64)
.unwrap_or_default(),
)
})
.unwrap_or_default();
if breakdown > 0 {
return breakdown;
}
usage
.get("input_tokens_details")
.or_else(|| usage.get("prompt_tokens_details"))
.and_then(Value::as_object)
.and_then(|details| details.get("cached_creation_tokens"))
.and_then(Value::as_u64)
.unwrap_or_default()
}
fn extract_cache_read_tokens_from_usage_object(usage: &serde_json::Map<String, Value>) -> u64 {
usage
.get("cache_read_input_tokens")
.and_then(Value::as_u64)
.or_else(|| {
usage
.get("input_tokens_details")
.or_else(|| usage.get("prompt_tokens_details"))
.and_then(Value::as_object)
.and_then(|details| details.get("cached_tokens"))
.and_then(Value::as_u64)
})
.unwrap_or_default()
}
fn empty_to_none(value: Option<String>) -> Option<String> {
value
.map(|value| value.trim().to_string())
@@ -1746,6 +1823,39 @@ mod tests {
assert_eq!(tokens, (3, 5, 8));
}
#[test]
fn extracts_openai_usage_tokens_with_cache_components() {
let tokens = extract_token_counts_from_json(&json!({
"usage": {
"input_tokens": 3,
"output_tokens": 5,
"total_tokens": 8,
"input_tokens_details": {
"cached_tokens": 2,
"cached_creation_tokens": 1
}
}
}))
.expect("tokens should exist");
assert_eq!(tokens, (3, 5, 11));
}
#[test]
fn extracts_claude_usage_tokens_with_cache_components() {
let tokens = extract_token_counts_from_json(&json!({
"usage": {
"input_tokens": 6,
"output_tokens": 20,
"cache_creation_input_tokens": 41857,
"cache_read_input_tokens": 0
}
}))
.expect("tokens should exist");
assert_eq!(tokens, (6, 20, 41883));
}
#[test]
fn builds_upsert_record_from_terminal_event() {
let record = build_upsert_usage_record_from_event(&UsageEvent {
@@ -2117,7 +2227,7 @@ mod tests {
assert_eq!(event.data.input_tokens, Some(3));
assert_eq!(event.data.output_tokens, Some(5));
assert_eq!(event.data.total_tokens, Some(8));
assert_eq!(event.data.total_tokens, Some(11));
assert_eq!(event.data.cache_creation_input_tokens, Some(1));
assert_eq!(event.data.cache_read_input_tokens, Some(2));
assert_eq!(
@@ -2531,7 +2641,7 @@ mod tests {
assert_eq!(event.data.input_tokens, Some(3));
assert_eq!(event.data.output_tokens, Some(5));
assert_eq!(event.data.total_tokens, Some(8));
assert_eq!(event.data.total_tokens, Some(11));
assert_eq!(event.data.cache_creation_input_tokens, Some(1));
assert_eq!(event.data.cache_read_input_tokens, Some(2));
assert_eq!(