web3-proxy/web3_proxy/src/user_queries.rs

462 lines
15 KiB
Rust
Raw Normal View History

2022-10-20 07:44:33 +03:00
use anyhow::Context;
2022-10-20 09:17:20 +03:00
use axum::{
headers::{authorization::Bearer, Authorization},
TypedHeader,
};
use chrono::NaiveDateTime;
2022-10-20 07:44:33 +03:00
use entities::{rpc_accounting, user_keys};
2022-10-20 02:02:34 +03:00
use hashbrown::HashMap;
2022-10-20 07:44:33 +03:00
use migration::Expr;
use num::Zero;
2022-10-20 09:17:20 +03:00
use redis_rate_limiter::{redis::AsyncCommands, RedisConnection};
2022-10-20 00:34:05 +03:00
use sea_orm::{
2022-10-20 09:54:45 +03:00
ColumnTrait, Condition, EntityTrait, JoinType, PaginatorTrait, QueryFilter, QueryOrder,
QuerySelect, RelationTrait,
2022-10-20 00:34:05 +03:00
};
2022-10-20 07:44:33 +03:00
use tracing::trace;
2022-10-20 09:17:20 +03:00
use crate::app::Web3ProxyApp;
/// get the attached address from redis for the given auth_token.
/// 0 means all users
2022-10-20 09:54:45 +03:00
async fn get_user_id_from_params(
2022-10-20 09:17:20 +03:00
mut redis_conn: RedisConnection,
// this is a long type. should we strip it down?
bearer: Option<TypedHeader<Authorization<Bearer>>>,
params: &HashMap<String, String>,
) -> anyhow::Result<u64> {
match (bearer, params.get("user_id")) {
(Some(bearer), Some(user_id)) => {
// check for the bearer cache key
// TODO: move this to a helper function
let bearer_cache_key = format!("bearer:{}", bearer.token());
// get the user id that is attached to this bearer token
redis_conn
.get::<_, u64>(bearer_cache_key)
.await
// TODO: this should be a 403
.context("fetching user_key_id from redis with bearer_cache_key")
}
(_, None) => {
// they have a bearer token. we don't care about it on public pages
// 0 means all
Ok(0)
}
(None, Some(x)) => {
// they do not have a bearer token, but requested a specific id. block
// TODO: proper error code
// TODO: maybe instead of this sharp edged warn, we have a config value?
// TODO: check config for if we should deny or allow this
x.parse().context("Parsing user_id param")
}
}
}
/// only allow user_key to be set if user_id is also set.
/// this will keep people from reading someone else's keys.
/// 0 means none.
2022-10-26 00:10:05 +03:00
pub fn get_user_key_id_from_params(
2022-10-20 09:54:45 +03:00
user_id: u64,
params: &HashMap<String, String>,
) -> anyhow::Result<u64> {
2022-10-20 09:17:20 +03:00
if user_id > 0 {
2022-10-20 20:43:31 +03:00
params.get("user_key_id").map_or_else(
2022-10-20 09:17:20 +03:00
|| Ok(0),
|c| {
let c = c.parse()?;
Ok(c)
},
)
} else {
Ok(0)
}
}
2022-10-26 00:10:05 +03:00
pub fn get_chain_id_from_params(
2022-10-20 09:17:20 +03:00
app: &Web3ProxyApp,
params: &HashMap<String, String>,
) -> anyhow::Result<u64> {
params.get("chain_id").map_or_else(
|| Ok(app.config.chain_id),
|c| {
let c = c.parse()?;
Ok(c)
},
)
}
2022-10-26 00:10:05 +03:00
pub fn get_query_start_from_params(
2022-10-20 09:17:20 +03:00
params: &HashMap<String, String>,
) -> anyhow::Result<chrono::NaiveDateTime> {
params.get("query_start").map_or_else(
|| {
// no timestamp in params. set default
let x = chrono::Utc::now() - chrono::Duration::days(30);
Ok(x.naive_utc())
},
|x: &String| {
// parse the given timestamp
let x = x.parse::<i64>().context("parsing timestamp query param")?;
// TODO: error code 401
let x =
NaiveDateTime::from_timestamp_opt(x, 0).context("parsing timestamp query param")?;
Ok(x)
},
)
}
2022-10-26 00:10:05 +03:00
pub fn get_page_from_params(params: &HashMap<String, String>) -> anyhow::Result<u64> {
2022-10-25 06:41:59 +03:00
params.get("page").map_or_else::<anyhow::Result<u64>, _, _>(
|| {
// no page in params. set default
Ok(0)
},
|x: &String| {
// parse the given timestamp
// TODO: error code 401
let x = x.parse().context("parsing page query from params")?;
2022-10-20 09:17:20 +03:00
2022-10-25 06:41:59 +03:00
Ok(x)
},
)
2022-10-20 09:17:20 +03:00
}
2022-10-26 00:10:05 +03:00
pub fn get_query_window_seconds_from_params(
params: &HashMap<String, String>,
) -> anyhow::Result<u64> {
2022-10-20 09:17:20 +03:00
params.get("query_window_seconds").map_or_else(
|| {
// no page in params. set default
Ok(0)
},
|x: &String| {
// parse the given timestamp
// TODO: error code 401
let x = x
.parse()
.context("parsing query window seconds from params")?;
Ok(x)
},
)
}
2022-10-20 07:44:33 +03:00
/// stats aggregated across a large time period
2022-10-20 09:17:20 +03:00
pub async fn get_aggregate_rpc_stats_from_params(
app: &Web3ProxyApp,
bearer: Option<TypedHeader<Authorization<Bearer>>>,
params: HashMap<String, String>,
2022-10-20 07:44:33 +03:00
) -> anyhow::Result<HashMap<&str, serde_json::Value>> {
2022-10-20 09:17:20 +03:00
let db_conn = app.db_conn().context("connecting to db")?;
let redis_conn = app.redis_conn().await.context("connecting to redis")?;
2022-10-20 09:54:45 +03:00
let user_id = get_user_id_from_params(redis_conn, bearer, &params).await?;
2022-10-20 09:17:20 +03:00
let chain_id = get_chain_id_from_params(app, &params)?;
let query_start = get_query_start_from_params(&params)?;
let query_window_seconds = get_query_window_seconds_from_params(&params)?;
2022-10-20 09:54:45 +03:00
let page = get_page_from_params(&params)?;
2022-10-20 09:17:20 +03:00
// TODO: warn if unknown fields in params
// TODO: page size from config
let page_size = 200;
trace!(?chain_id, %query_start, ?user_id, "get_aggregate_stats");
2022-10-20 00:34:05 +03:00
2022-10-20 07:44:33 +03:00
// TODO: minimum query_start of 90 days?
let mut response = HashMap::new();
response.insert("page", serde_json::to_value(page)?);
response.insert("page_size", serde_json::to_value(page_size)?);
response.insert("chain_id", serde_json::to_value(chain_id)?);
2022-10-20 09:17:20 +03:00
response.insert(
"query_start",
2022-10-20 09:54:45 +03:00
serde_json::to_value(query_start.timestamp() as u64)?,
2022-10-20 09:17:20 +03:00
);
2022-10-20 07:44:33 +03:00
// TODO: how do we get count reverts compared to other errors? does it matter? what about http errors to our users?
// TODO: how do we count uptime?
let q = rpc_accounting::Entity::find()
.select_only()
.column_as(
rpc_accounting::Column::FrontendRequests.sum(),
"total_requests",
)
.column_as(
rpc_accounting::Column::CacheMisses.sum(),
"total_cache_misses",
)
.column_as(rpc_accounting::Column::CacheHits.sum(), "total_cache_hits")
.column_as(
rpc_accounting::Column::BackendRetries.sum(),
"total_backend_retries",
)
.column_as(
rpc_accounting::Column::SumResponseBytes.sum(),
"total_response_bytes",
)
.column_as(
// TODO: can we sum bools like this?
rpc_accounting::Column::ErrorResponse.sum(),
"total_error_responses",
)
.column_as(
rpc_accounting::Column::SumResponseMillis.sum(),
"total_response_millis",
2022-10-20 07:44:33 +03:00
)
.order_by_asc(rpc_accounting::Column::PeriodDatetime.min());
2022-10-20 09:54:45 +03:00
// TODO: DRYer
2022-10-20 09:17:20 +03:00
let q = if query_window_seconds != 0 {
/*
let query_start_timestamp: u64 = query_start
.timestamp()
.try_into()
.context("query_start to timestamp")?;
*/
2022-10-20 07:44:33 +03:00
// TODO: is there a better way to do this? how can we get "period_datetime" into this with types?
2022-10-20 09:17:20 +03:00
// TODO: how can we get the first window to start at query_start_timestamp
2022-10-20 07:44:33 +03:00
let expr = Expr::cust_with_values(
"FLOOR(UNIX_TIMESTAMP(rpc_accounting.period_datetime) / ?) * ?",
[query_window_seconds, query_window_seconds],
);
response.insert(
"query_window_seconds",
serde_json::to_value(query_window_seconds)?,
);
2022-10-20 22:01:07 +03:00
q.column_as(expr, "query_window_seconds")
.group_by(Expr::cust("query_window_seconds"))
2022-10-20 07:44:33 +03:00
} else {
// TODO: order by more than this?
2022-10-20 09:54:45 +03:00
// query_window_seconds is not set so we aggregate all records
2022-10-20 07:44:33 +03:00
q
};
let condition = Condition::all().add(rpc_accounting::Column::PeriodDatetime.gte(query_start));
let (condition, q) = if chain_id.is_zero() {
// fetch all the chains. don't filter
2022-10-20 00:34:05 +03:00
// TODO: wait. do we want chain id on the logs? we can get that by joining key
let q = q
.column(rpc_accounting::Column::ChainId)
.group_by(rpc_accounting::Column::ChainId);
(condition, q)
} else {
let condition = condition.add(rpc_accounting::Column::ChainId.eq(chain_id));
(condition, q)
};
let (condition, q) = if user_id.is_zero() {
// 0 means everyone. don't filter on user
(condition, q)
} else {
// TODO: authentication here? or should that be higher in the stack? here sems safest
// TODO: only join some columns
// TODO: are these joins correct?
// TODO: what about keys where they are the secondary users?
let q = q
.join(
JoinType::InnerJoin,
rpc_accounting::Relation::UserKeys.def(),
)
.column(user_keys::Column::UserId)
.group_by(user_keys::Column::UserId);
let condition = condition.add(user_keys::Column::UserId.eq(user_id));
(condition, q)
};
let q = q.filter(condition);
2022-10-20 00:34:05 +03:00
// TODO: enum between searching on user_key_id on user_id
// TODO: handle secondary users, too
// log query here. i think sea orm has a useful log level for this
2022-10-20 07:44:33 +03:00
let aggregate = q
.into_json()
2022-10-20 09:17:20 +03:00
.paginate(&db_conn, page_size)
2022-10-20 07:44:33 +03:00
.fetch_page(page)
.await?;
2022-10-20 09:17:20 +03:00
response.insert("aggregate", serde_json::Value::Array(aggregate));
2022-10-20 07:44:33 +03:00
Ok(response)
}
2022-10-20 02:02:34 +03:00
/// stats grouped by key_id and error_repsponse and method and key
pub async fn get_detailed_stats(
2022-10-20 09:17:20 +03:00
app: &Web3ProxyApp,
bearer: Option<TypedHeader<Authorization<Bearer>>>,
params: HashMap<String, String>,
2022-10-20 02:02:34 +03:00
) -> anyhow::Result<HashMap<&str, serde_json::Value>> {
2022-10-20 09:54:45 +03:00
let db_conn = app.db_conn().context("connecting to db")?;
let redis_conn = app.redis_conn().await.context("connecting to redis")?;
let user_id = get_user_id_from_params(redis_conn, bearer, &params).await?;
let user_key_id = get_user_key_id_from_params(user_id, &params)?;
2022-10-20 09:17:20 +03:00
let chain_id = get_chain_id_from_params(app, &params)?;
let query_start = get_query_start_from_params(&params)?;
2022-10-20 09:54:45 +03:00
let query_window_seconds = get_query_window_seconds_from_params(&params)?;
2022-10-20 09:17:20 +03:00
let page = get_page_from_params(&params)?;
2022-10-20 09:54:45 +03:00
// TODO: handle secondary users, too
2022-10-20 09:17:20 +03:00
// TODO: page size from config
let page_size = 200;
2022-10-20 02:02:34 +03:00
2022-10-20 07:44:33 +03:00
// TODO: minimum query_start of 90 days?
2022-10-20 02:02:34 +03:00
let mut response = HashMap::new();
2022-10-20 07:44:33 +03:00
response.insert("page", serde_json::to_value(page)?);
response.insert("page_size", serde_json::to_value(page_size)?);
2022-10-20 02:02:34 +03:00
response.insert("chain_id", serde_json::to_value(chain_id)?);
2022-10-20 09:17:20 +03:00
response.insert(
"query_start",
2022-10-20 09:54:45 +03:00
serde_json::to_value(query_start.timestamp() as u64)?,
2022-10-20 09:17:20 +03:00
);
2022-10-20 02:02:34 +03:00
// TODO: how do we get count reverts compared to other errors? does it matter? what about http errors to our users?
// TODO: how do we count uptime?
let q = rpc_accounting::Entity::find()
.select_only()
// groups
.column(rpc_accounting::Column::ErrorResponse)
.group_by(rpc_accounting::Column::ErrorResponse)
.column(rpc_accounting::Column::Method)
.group_by(rpc_accounting::Column::Method)
// aggregate columns
.column_as(
rpc_accounting::Column::FrontendRequests.sum(),
"total_requests",
)
.column_as(
rpc_accounting::Column::CacheMisses.sum(),
"total_cache_misses",
)
.column_as(rpc_accounting::Column::CacheHits.sum(), "total_cache_hits")
.column_as(
rpc_accounting::Column::BackendRetries.sum(),
"total_backend_retries",
)
.column_as(
rpc_accounting::Column::SumResponseBytes.sum(),
"total_response_bytes",
)
.column_as(
// TODO: can we sum bools like this?
rpc_accounting::Column::ErrorResponse.sum(),
"total_error_responses",
)
.column_as(
rpc_accounting::Column::SumResponseMillis.sum(),
"total_response_millis",
2022-10-20 07:44:33 +03:00
)
// TODO: order on method next?
.order_by_asc(rpc_accounting::Column::PeriodDatetime.min());
let condition = Condition::all().add(rpc_accounting::Column::PeriodDatetime.gte(query_start));
let (condition, q) = if chain_id.is_zero() {
// fetch all the chains. don't filter
// TODO: wait. do we want chain id on the logs? we can get that by joining key
let q = q
.column(rpc_accounting::Column::ChainId)
.group_by(rpc_accounting::Column::ChainId);
(condition, q)
} else {
let condition = condition.add(rpc_accounting::Column::ChainId.eq(chain_id));
(condition, q)
};
2022-10-20 09:54:45 +03:00
let (condition, q) = if user_id == 0 {
2022-10-20 07:44:33 +03:00
// 0 means everyone. don't filter on user
(condition, q)
} else {
// TODO: move authentication here?
// TODO: what about keys where this user is a secondary user?
let q = q
.join(
JoinType::InnerJoin,
rpc_accounting::Relation::UserKeys.def(),
)
.column(user_keys::Column::UserId)
2022-10-20 09:54:45 +03:00
.group_by(user_keys::Column::UserId);
2022-10-20 07:44:33 +03:00
let condition = condition.add(user_keys::Column::UserId.eq(user_id));
2022-10-20 09:54:45 +03:00
let q = if user_key_id == 0 {
q.column(user_keys::Column::UserId)
.group_by(user_keys::Column::UserId)
} else {
response.insert("user_key_id", serde_json::to_value(user_key_id)?);
// no need to group_by user_id when we are grouping by key_id
q.column(user_keys::Column::Id)
.group_by(user_keys::Column::Id)
};
2022-10-20 07:44:33 +03:00
(condition, q)
};
2022-10-20 22:01:07 +03:00
let q = if query_window_seconds != 0 {
/*
let query_start_timestamp: u64 = query_start
.timestamp()
.try_into()
.context("query_start to timestamp")?;
*/
// TODO: is there a better way to do this? how can we get "period_datetime" into this with types?
// TODO: how can we get the first window to start at query_start_timestamp
let expr = Expr::cust_with_values(
"FLOOR(UNIX_TIMESTAMP(rpc_accounting.period_datetime) / ?) * ?",
[query_window_seconds, query_window_seconds],
);
response.insert(
"query_window_seconds",
serde_json::to_value(query_window_seconds)?,
);
q.column_as(expr, "query_window_seconds")
.group_by(Expr::cust("query_window_seconds"))
} else {
// TODO: order by more than this?
// query_window_seconds is not set so we aggregate all records
q
};
2022-10-20 23:26:14 +03:00
let q = q.filter(condition);
2022-10-20 07:44:33 +03:00
// log query here. i think sea orm has a useful log level for this
// TODO: transform this into a nested hashmap instead of a giant table?
let r = q
.into_json()
2022-10-20 09:17:20 +03:00
.paginate(&db_conn, page_size)
2022-10-20 07:44:33 +03:00
.fetch_page(page)
.await?;
response.insert("detailed_aggregate", serde_json::Value::Array(r));
// number of keys
// number of secondary keys
// avg and max concurrent requests per second per api key
Ok(response)
}