A rust library to query Apache Pinot.
To install Pinot locally, please follow this Pinot Quickstart link to install and start Pinot batch quickstart locally.
bin/quick-start-batch.sh
Alternatively, the docker contained Pinot database ochestrated by this repository's docker-compose.yaml
file may be used.
make prepare-pinot
Check out Client library Github Repo
git clone [email protected]:yougov/pinot-client-rust.git
cd pinot-client-rust
Start up the docker contained pinot database
make prepare-pinot
Build and run an example application to query from Pinot
cargo run --example pql-query
cargo run --example sql-query-deserialize-to-data-row
cargo run --example sql-query-deserialize-to-struct
Pinot client could be initialized through:
- Zookeeper Path.
let client = pinot_client_rust::connection::client_from_zookeeper(
&pinot_client_rust::zookeeper::ZookeeperConfig::new(
vec!["localhost:2181".to_string()],
"/PinotCluster".to_string(),
),
None
);
- A list of broker addresses.
let client = pinot_client_rust::connection::client_from_broker_list(
vec!["localhost:8099".to_string()], None);
An asynchronous connection can be established with pinot_client_rust::async_connection::AsyncConnection
for
which exist equivalents to the above described synchronous instantiation methods.
Please see this example for your reference.
Code snippet:
fn main() {
let client = pinot_client_rust::connection::client_from_broker_list(
vec!["localhost:8099".to_string()], None).unwrap();
let broker_response = client.execute_sql::<pinot_client_rust::response::data::DataRow>(
"baseballStats",
"select count(*) as cnt, sum(homeRuns) as sum_homeRuns from baseballStats group by teamID limit 10"
).unwrap();
if let Some(stats) = broker_response.stats {
log::info!(
"Query Stats: response time - {} ms, scanned docs - {}, total docs - {}",
stats.time_used_ms,
stats.num_docs_scanned,
stats.total_docs,
);
}
}
Query Responses are defined by one of two broker response structures.
SQL queries return SqlResponse
, whose generic parameter is supported by all structs implementing the
FromRow
trait, whereas PQL queries return PqlResponse
.
SqlResponse
contains a Table
, the holder for SQL query data, whereas PqlResponse
contains
AggregationResults
and SelectionResults
, the holders for PQL query data.
Exceptions for a given request for both SqlResponse
and PqlResponse
are stored in the Exception
array.
Stats for a given request for both SqlResponse
and PqlResponse
are stored in ResponseStats
.
Exception
is defined as:
/// Pinot exception.
#[derive(Clone, Debug, Deserialize, Eq, PartialEq)]
pub struct PinotException {
#[serde(rename(deserialize = "errorCode"))]
pub error_code: i32,
pub message: String,
}
ResponseStats
is defined as:
/// ResponseStats carries all stats returned by a query.
#[derive(Clone, Debug, PartialEq)]
pub struct ResponseStats {
pub trace_info: HashMap<String, String>,
pub num_servers_queried: i32,
pub num_servers_responded: i32,
pub num_segments_queried: i32,
pub num_segments_processed: i32,
pub num_segments_matched: i32,
pub num_consuming_segments_queried: i32,
pub num_docs_scanned: i64,
pub num_entries_scanned_in_filter: i64,
pub num_entries_scanned_post_filter: i64,
pub num_groups_limit_reached: bool,
pub total_docs: i64,
pub time_used_ms: i32,
pub min_consuming_freshness_time_ms: i64,
}
PqlResponse
is defined as:
/// PqlResponse is the data structure for broker response to a PQL query.
#[derive(Clone, Debug, PartialEq)]
pub struct PqlResponse {
pub aggregation_results: Vec<AggregationResult>,
pub selection_results: Option<SelectionResults>,
pub stats: Option<ResponseStats>,
}
SqlResponse
is defined as:
/// SqlResponse is the data structure for a broker response to an SQL query.
#[derive(Clone, Debug, PartialEq)]
pub struct SqlResponse<T: FromRow> {
pub table: Option<Table<T>>,
pub stats: Option<ResponseStats>,
}
Table
is defined as:
/// Table is the holder for SQL queries.
#[derive(Clone, Debug, PartialEq)]
pub struct Table<T: FromRow> {
schema: Schema,
rows: Vec<T>,
}
Schema
is defined as:
/// Schema is response schema with a bimap to allow easy name <-> index retrieval
#[derive(Clone, Debug, Eq, PartialEq)]
pub struct Schema {
column_data_types: Vec<DataType>,
column_name_to_index: bimap::BiMap::<String, usize>,
}
There are multiple functions defined for Schema
, like:
fn get_column_count(&self) -> usize;
fn get_column_name(&self, column_index: usize) -> Result<&str>;
fn get_column_index(&self, column_name: &str) -> Result<usize>;
fn get_column_data_type(&self, column_index: usize) -> Result<DataType>;
fn get_column_data_type_by_name(&self, column_name: &str) -> Result<DataType>;
DataType
is defined as:
/// Pinot native types
#[derive(Clone, Debug, Eq, PartialEq)]
pub enum DataType {
Int,
Long,
Float,
Double,
Boolean,
Timestamp,
String,
Json,
Bytes,
IntArray,
LongArray,
FloatArray,
DoubleArray,
BooleanArray,
TimestampArray,
StringArray,
BytesArray,
}
FromRow
is defined as:
/// FromRow represents any structure which can deserialize
/// the Table.rows json field provided a `Schema`
pub trait FromRow: Sized {
fn from_row(
data_schema: &Schema,
row: Vec<Value>,
) -> std::result::Result<Self, serde_json::Error>;
}
In addition to being implemented by DataRow
, FromRow
is also implemented by all implementors
of serde::de::Deserialize
, which is achieved by first deserializing the response to json and then
before each row is deserialized into final form, a json map of column name to value is substituted.
Additionally, there are a number of serde deserializer functions provided to deserialize complex pinot types:
/// Converts Pinot timestamps into `Vec<DateTime<Utc>>` using `deserialize_timestamps_from_json()`.
fn deserialize_timestamps<'de, D>(deserializer: D) -> std::result::Result<Vec<DateTime<Utc>>, D::Error>...
/// Converts Pinot timestamps into `DateTime<Utc>` using `deserialize_timestamp_from_json()`.
pub fn deserialize_timestamp<'de, D>(deserializer: D) -> std::result::Result<DateTime<Utc>, D::Error>...
/// Converts Pinot hex strings into `Vec<Vec<u8>>` using `deserialize_bytes_array_from_json()`.
pub fn deserialize_bytes_array<'de, D>(deserializer: D) -> std::result::Result<Vec<Vec<u8>>, D::Error>...
/// Converts Pinot hex string into `Vec<u8>` using `deserialize_bytes_from_json()`.
pub fn deserialize_bytes<'de, D>(deserializer: D) -> std::result::Result<Vec<u8>, D::Error>...
/// Deserializes json potentially packaged into a string by calling `deserialize_json_from_json()`.
pub fn deserialize_json<'de, D>(deserializer: D) -> std::result::Result<Value, D::Error>
For example usage, please refer to this example
DataRow
is defined as:
/// A row of `Data`
#[derive(Clone, Debug, PartialEq)]
pub struct DataRow {
row: Vec<Data>,
}
Data
is defined as:
/// Typed Pinot data
#[derive(Clone, Debug, PartialEq)]
pub enum Data {
Int(i32),
Long(i64),
Float(f32),
Double(f64),
Boolean(bool),
Timestamp(DateTime<Utc>),
String(String),
Json(Value),
Bytes(Vec<u8>),
IntArray(Vec<i32>),
LongArray(Vec<i64>),
FloatArray(Vec<f32>),
DoubleArray(Vec<f64>),
BooleanArray(Vec<bool>),
TimestampArray(Vec<DateTime<Utc>>),
StringArray(Vec<String>),
BytesArray(Vec<Vec<u8>>),
Null(DataType),
}
There are multiple functions defined for Data
, like:
fn data_type(&self) -> DataType;
fn get_int(&self) -> Result<i32>;
fn get_long(&self) -> Result<i64>;
fn get_float(&self) -> Result<f32>;
fn get_double(&self) -> Result<f64>;
fn get_boolean(&self) -> Result<bool>;
fn get_timestamp(&self) -> Result<DateTime<Utc>>;
fn get_string(&self) -> Result<&str>;
fn get_json(&self) -> Result<&Value>;
fn get_bytes(&self) -> Result<&Vec<u8>>;
fn is_null(&self) -> bool;
In addition to row count, DataRow
also contains convenience counterparts to those above given a column index.