Track lifetime spending for each user and model (#16137)

Release Notes:

- N/A

Co-authored-by: Marshall <marshall@zed.dev>
This commit is contained in:
Max Brunsfeld 2024-08-12 17:15:26 -07:00 committed by GitHub
parent 68ae347077
commit dbcd06642c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
8 changed files with 121 additions and 11 deletions

View file

@ -0,0 +1,9 @@
create table lifetime_usages (
id serial primary key,
user_id integer not null,
model_id integer not null references models (id) on delete cascade,
input_tokens bigint not null default 0,
output_tokens bigint not null default 0
);
create unique index uix_lifetime_usages_on_user_id_model_id on lifetime_usages (user_id, model_id);

View file

@ -4,8 +4,8 @@ mod telemetry;
mod token;
use crate::{
api::CloudflareIpCountryHeader, build_clickhouse_client, executor::Executor, Config, Error,
Result,
api::CloudflareIpCountryHeader, build_clickhouse_client, db::UserId, executor::Executor,
Config, Error, Result,
};
use anyhow::{anyhow, Context as _};
use authorization::authorize_access_to_language_model;
@ -396,7 +396,12 @@ async fn check_usage_limit(
let model = state.db.model(provider, model_name)?;
let usage = state
.db
.get_usage(claims.user_id as i32, provider, model_name, Utc::now())
.get_usage(
UserId::from_proto(claims.user_id),
provider,
model_name,
Utc::now(),
)
.await?;
let active_users = state.get_active_user_count().await?;
@ -523,7 +528,7 @@ impl<S> Drop for TokenCountingStream<S> {
let usage = state
.db
.record_usage(
claims.user_id as i32,
UserId::from_proto(claims.user_id),
claims.is_staff,
provider,
&model,
@ -555,6 +560,7 @@ impl<S> Drop for TokenCountingStream<S> {
input_tokens_this_month: usage.input_tokens_this_month as u64,
output_tokens_this_month: usage.output_tokens_this_month as u64,
spending_this_month: usage.spending_this_month as u64,
lifetime_spending: usage.lifetime_spending as u64,
},
)
.await

View file

@ -1,3 +1,4 @@
use crate::db::UserId;
use chrono::Duration;
use rpc::LanguageModelProvider;
use sea_orm::QuerySelect;
@ -14,6 +15,7 @@ pub struct Usage {
pub input_tokens_this_month: usize,
pub output_tokens_this_month: usize,
pub spending_this_month: usize,
pub lifetime_spending: usize,
}
#[derive(Clone, Copy, Debug, Default)]
@ -63,7 +65,7 @@ impl LlmDatabase {
pub async fn get_usage(
&self,
user_id: i32,
user_id: UserId,
provider: LanguageModelProvider,
model_name: &str,
now: DateTimeUtc,
@ -83,6 +85,18 @@ impl LlmDatabase {
.all(&*tx)
.await?;
let (lifetime_input_tokens, lifetime_output_tokens) = lifetime_usage::Entity::find()
.filter(
lifetime_usage::Column::UserId
.eq(user_id)
.and(lifetime_usage::Column::ModelId.eq(model.id)),
)
.one(&*tx)
.await?
.map_or((0, 0), |usage| {
(usage.input_tokens as usize, usage.output_tokens as usize)
});
let requests_this_minute =
self.get_usage_for_measure(&usages, now, UsageMeasure::RequestsPerMinute)?;
let tokens_this_minute =
@ -95,6 +109,8 @@ impl LlmDatabase {
self.get_usage_for_measure(&usages, now, UsageMeasure::OutputTokensPerMonth)?;
let spending_this_month =
calculate_spending(model, input_tokens_this_month, output_tokens_this_month);
let lifetime_spending =
calculate_spending(model, lifetime_input_tokens, lifetime_output_tokens);
Ok(Usage {
requests_this_minute,
@ -103,6 +119,7 @@ impl LlmDatabase {
input_tokens_this_month,
output_tokens_this_month,
spending_this_month,
lifetime_spending,
})
})
.await
@ -111,7 +128,7 @@ impl LlmDatabase {
#[allow(clippy::too_many_arguments)]
pub async fn record_usage(
&self,
user_id: i32,
user_id: UserId,
is_staff: bool,
provider: LanguageModelProvider,
model_name: &str,
@ -194,6 +211,50 @@ impl LlmDatabase {
let spending_this_month =
calculate_spending(model, input_tokens_this_month, output_tokens_this_month);
// Update lifetime usage
let lifetime_usage = lifetime_usage::Entity::find()
.filter(
lifetime_usage::Column::UserId
.eq(user_id)
.and(lifetime_usage::Column::ModelId.eq(model.id)),
)
.one(&*tx)
.await?;
let lifetime_usage = match lifetime_usage {
Some(usage) => {
lifetime_usage::Entity::update(lifetime_usage::ActiveModel {
id: ActiveValue::unchanged(usage.id),
input_tokens: ActiveValue::set(
usage.input_tokens + input_token_count as i64,
),
output_tokens: ActiveValue::set(
usage.output_tokens + output_token_count as i64,
),
..Default::default()
})
.exec(&*tx)
.await?
}
None => {
lifetime_usage::ActiveModel {
user_id: ActiveValue::set(user_id),
model_id: ActiveValue::set(model.id),
input_tokens: ActiveValue::set(input_token_count as i64),
output_tokens: ActiveValue::set(output_token_count as i64),
..Default::default()
}
.insert(&*tx)
.await?
}
};
let lifetime_spending = calculate_spending(
model,
lifetime_usage.input_tokens as usize,
lifetime_usage.output_tokens as usize,
);
Ok(Usage {
requests_this_minute,
tokens_this_minute,
@ -201,6 +262,7 @@ impl LlmDatabase {
input_tokens_this_month,
output_tokens_this_month,
spending_this_month,
lifetime_spending,
})
})
.await
@ -246,7 +308,7 @@ impl LlmDatabase {
#[allow(clippy::too_many_arguments)]
async fn update_usage_for_measure(
&self,
user_id: i32,
user_id: UserId,
is_staff: bool,
model_id: ModelId,
usages: &[usage::Model],

View file

@ -1,3 +1,4 @@
pub mod lifetime_usage;
pub mod model;
pub mod provider;
pub mod usage;

View file

@ -0,0 +1,18 @@
use crate::{db::UserId, llm::db::ModelId};
use sea_orm::entity::prelude::*;
#[derive(Clone, Debug, PartialEq, DeriveEntityModel)]
#[sea_orm(table_name = "lifetime_usages")]
pub struct Model {
#[sea_orm(primary_key)]
pub id: i32,
pub user_id: UserId,
pub model_id: ModelId,
pub input_tokens: i64,
pub output_tokens: i64,
}
#[derive(Copy, Clone, Debug, EnumIter, DeriveRelation)]
pub enum Relation {}
impl ActiveModelBehavior for ActiveModel {}

View file

@ -1,4 +1,7 @@
use crate::llm::db::{ModelId, UsageId, UsageMeasureId};
use crate::{
db::UserId,
llm::db::{ModelId, UsageId, UsageMeasureId},
};
use sea_orm::entity::prelude::*;
/// An LLM usage record.
@ -10,7 +13,7 @@ pub struct Model {
/// The ID of the Zed user.
///
/// Corresponds to the `users` table in the primary collab database.
pub user_id: i32,
pub user_id: UserId,
pub model_id: ModelId,
pub measure_id: UsageMeasureId,
pub timestamp: DateTime,

View file

@ -1,5 +1,9 @@
use crate::{
llm::db::{queries::providers::ModelParams, queries::usages::Usage, LlmDatabase},
db::UserId,
llm::db::{
queries::{providers::ModelParams, usages::Usage},
LlmDatabase,
},
test_llm_db,
};
use chrono::{Duration, Utc};
@ -26,7 +30,7 @@ async fn test_tracking_usage(db: &mut LlmDatabase) {
.unwrap();
let t0 = Utc::now();
let user_id = 123;
let user_id = UserId::from_proto(123);
let now = t0;
db.record_usage(user_id, false, provider, model, 1000, 0, now)
@ -48,6 +52,7 @@ async fn test_tracking_usage(db: &mut LlmDatabase) {
input_tokens_this_month: 3000,
output_tokens_this_month: 0,
spending_this_month: 0,
lifetime_spending: 0,
}
);
@ -62,6 +67,7 @@ async fn test_tracking_usage(db: &mut LlmDatabase) {
input_tokens_this_month: 3000,
output_tokens_this_month: 0,
spending_this_month: 0,
lifetime_spending: 0,
}
);
@ -80,6 +86,7 @@ async fn test_tracking_usage(db: &mut LlmDatabase) {
input_tokens_this_month: 6000,
output_tokens_this_month: 0,
spending_this_month: 0,
lifetime_spending: 0,
}
);
@ -95,6 +102,7 @@ async fn test_tracking_usage(db: &mut LlmDatabase) {
input_tokens_this_month: 6000,
output_tokens_this_month: 0,
spending_this_month: 0,
lifetime_spending: 0,
}
);
@ -112,6 +120,7 @@ async fn test_tracking_usage(db: &mut LlmDatabase) {
input_tokens_this_month: 10000,
output_tokens_this_month: 0,
spending_this_month: 0,
lifetime_spending: 0,
}
);
@ -127,6 +136,7 @@ async fn test_tracking_usage(db: &mut LlmDatabase) {
input_tokens_this_month: 9000,
output_tokens_this_month: 0,
spending_this_month: 0,
lifetime_spending: 0,
}
);
}

View file

@ -17,6 +17,7 @@ pub struct LlmUsageEventRow {
pub input_tokens_this_month: u64,
pub output_tokens_this_month: u64,
pub spending_this_month: u64,
pub lifetime_spending: u64,
}
#[derive(Serialize, Debug, clickhouse::Row)]