refactor: generic top-level flight implementation (#3882)

This allows us to implement flight for the NG querier by just
implementing a few traits and reuse all the existing glue code and
optimizations (like dictionary handling).
pull/24376/head
Marco Neumann 2022-03-01 14:33:08 +00:00 committed by GitHub
parent 43ada68f37
commit e5c45aeab6
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 600 additions and 565 deletions

View File

@ -11,6 +11,7 @@ use crate::influxdb_ioxd::{
serving_readiness::ServingReadiness,
};
pub(crate) mod flight;
pub(crate) mod testing;
/// Returns the name of the gRPC service S.

View File

@ -0,0 +1,587 @@
//! Implements the native gRPC IOx query API using Arrow Flight
use std::fmt::Debug;
use std::task::Poll;
use std::{pin::Pin, sync::Arc};
use arrow::{
array::{make_array, ArrayRef, MutableArrayData},
datatypes::{DataType, Field, Schema, SchemaRef},
error::ArrowError,
record_batch::RecordBatch,
};
use arrow_flight::{
flight_service_server::{FlightService as Flight, FlightServiceServer as FlightServer},
Action, ActionType, Criteria, Empty, FlightData, FlightDescriptor, FlightInfo,
HandshakeRequest, HandshakeResponse, PutResult, SchemaAsIpc, SchemaResult, Ticket,
};
use datafusion::physical_plan::ExecutionPlan;
use futures::{SinkExt, Stream, StreamExt};
use pin_project::{pin_project, pinned_drop};
use query::QueryDatabase;
use serde::Deserialize;
use snafu::{ResultExt, Snafu};
use tokio::task::JoinHandle;
use tonic::{Request, Response, Streaming};
use data_types::{DatabaseName, DatabaseNameError};
use observability_deps::tracing::{info, warn};
use query::exec::{ExecutionContextProvider, IOxExecutionContext};
use crate::influxdb_ioxd::planner::Planner;
#[allow(clippy::enum_variant_names)]
#[derive(Debug, Snafu)]
pub enum Error {
#[snafu(display("Invalid ticket. Error: {:?} Ticket: {:?}", source, ticket))]
InvalidTicket {
source: std::string::FromUtf8Error,
ticket: Vec<u8>,
},
#[snafu(display("Invalid query, could not parse '{}': {}", query, source))]
InvalidQuery {
query: String,
source: serde_json::Error,
},
#[snafu(display("Database {} not found", database_name))]
DatabaseNotFound { database_name: String },
#[snafu(display(
"Internal error reading points from database {}: {}",
database_name,
source
))]
Query {
database_name: String,
source: Box<dyn std::error::Error + Send + Sync>,
},
#[snafu(display("Invalid database name: {}", source))]
InvalidDatabaseName { source: DatabaseNameError },
#[snafu(display("Invalid RecordBatch: {}", source))]
InvalidRecordBatch { source: ArrowError },
#[snafu(display("Failed to hydrate dictionary: {}", source))]
DictionaryError { source: ArrowError },
#[snafu(display("Error while planning query: {}", source))]
Planning {
source: crate::influxdb_ioxd::planner::Error,
},
}
pub type Result<T, E = Error> = std::result::Result<T, E>;
impl From<Error> for tonic::Status {
/// Converts a result from the business logic into the appropriate tonic
/// status
fn from(err: Error) -> Self {
// An explicit match on the Error enum will ensure appropriate
// logging is handled for any new error variants.
let msg = "Error handling Flight gRPC request";
match err {
Error::DatabaseNotFound { .. }
| Error::InvalidTicket { .. }
| Error::InvalidQuery { .. }
// TODO(edd): this should be `debug`. Keeping at info whilst IOx still in early development
| Error::InvalidDatabaseName { .. } => info!(?err, msg),
Error::Query { .. } => info!(?err, msg),
Error::DictionaryError { .. }
| Error::InvalidRecordBatch { .. }
| Error::Planning { .. } => warn!(?err, msg),
}
err.to_status()
}
}
impl Error {
/// Converts a result from the business logic into the appropriate tonic
/// status
fn to_status(&self) -> tonic::Status {
use tonic::Status;
match &self {
Self::InvalidTicket { .. } => Status::invalid_argument(self.to_string()),
Self::InvalidQuery { .. } => Status::invalid_argument(self.to_string()),
Self::DatabaseNotFound { .. } => Status::not_found(self.to_string()),
Self::Query { .. } => Status::internal(self.to_string()),
Self::InvalidDatabaseName { .. } => Status::invalid_argument(self.to_string()),
Self::InvalidRecordBatch { .. } => Status::internal(self.to_string()),
Self::Planning { .. } => Status::invalid_argument(self.to_string()),
Self::DictionaryError { .. } => Status::internal(self.to_string()),
}
}
}
type TonicStream<T> = Pin<Box<dyn Stream<Item = Result<T, tonic::Status>> + Send + Sync + 'static>>;
#[derive(Deserialize, Debug)]
/// Body of the `Ticket` serialized and sent to the do_get endpoint; this should
/// be shared with the read API probably...
struct ReadInfo {
database_name: String,
sql_query: String,
}
pub trait QueryDatabaseProvider: std::fmt::Debug + Send + Sync + 'static {
type Db: ExecutionContextProvider + QueryDatabase;
fn db(&self, db_name: &DatabaseName<'_>) -> std::result::Result<Arc<Self::Db>, tonic::Status>;
}
/// Concrete implementation of the gRPC Arrow Flight Service API
#[derive(Debug)]
struct FlightService<S>
where
S: QueryDatabaseProvider,
{
server: Arc<S>,
}
pub fn make_server<S>(server: Arc<S>) -> FlightServer<impl Flight>
where
S: QueryDatabaseProvider,
{
FlightServer::new(FlightService { server })
}
#[tonic::async_trait]
impl<S> Flight for FlightService<S>
where
S: QueryDatabaseProvider,
{
type HandshakeStream = TonicStream<HandshakeResponse>;
type ListFlightsStream = TonicStream<FlightInfo>;
type DoGetStream = TonicStream<FlightData>;
type DoPutStream = TonicStream<PutResult>;
type DoActionStream = TonicStream<arrow_flight::Result>;
type ListActionsStream = TonicStream<ActionType>;
type DoExchangeStream = TonicStream<FlightData>;
async fn get_schema(
&self,
_request: Request<FlightDescriptor>,
) -> Result<Response<SchemaResult>, tonic::Status> {
Err(tonic::Status::unimplemented("Not yet implemented"))
}
async fn do_get(
&self,
request: Request<Ticket>,
) -> Result<Response<Self::DoGetStream>, tonic::Status> {
let span_ctx = request.extensions().get().cloned();
let ticket = request.into_inner();
let json_str = String::from_utf8(ticket.ticket.to_vec()).context(InvalidTicketSnafu {
ticket: ticket.ticket,
})?;
let read_info: ReadInfo =
serde_json::from_str(&json_str).context(InvalidQuerySnafu { query: &json_str })?;
let database =
DatabaseName::new(&read_info.database_name).context(InvalidDatabaseNameSnafu)?;
let db = self.server.db(&database)?;
let _query_completed_token = db.record_query("sql", Box::new(read_info.sql_query.clone()));
let ctx = db.new_query_context(span_ctx);
let physical_plan = Planner::new(&ctx)
.sql(&read_info.sql_query)
.await
.context(PlanningSnafu)?;
let output = GetStream::new(ctx, physical_plan, read_info.database_name).await?;
Ok(Response::new(Box::pin(output) as Self::DoGetStream))
}
async fn handshake(
&self,
request: Request<Streaming<HandshakeRequest>>,
) -> Result<Response<Self::HandshakeStream>, tonic::Status> {
let request = request.into_inner().message().await?.unwrap();
let response = HandshakeResponse {
protocol_version: request.protocol_version,
payload: request.payload,
};
let output = futures::stream::iter(std::iter::once(Ok(response)));
Ok(Response::new(Box::pin(output) as Self::HandshakeStream))
}
async fn list_flights(
&self,
_request: Request<Criteria>,
) -> Result<Response<Self::ListFlightsStream>, tonic::Status> {
Err(tonic::Status::unimplemented("Not yet implemented"))
}
async fn get_flight_info(
&self,
_request: Request<FlightDescriptor>,
) -> Result<Response<FlightInfo>, tonic::Status> {
Err(tonic::Status::unimplemented("Not yet implemented"))
}
async fn do_put(
&self,
_request: Request<Streaming<FlightData>>,
) -> Result<Response<Self::DoPutStream>, tonic::Status> {
Err(tonic::Status::unimplemented("Not yet implemented"))
}
async fn do_action(
&self,
_request: Request<Action>,
) -> Result<Response<Self::DoActionStream>, tonic::Status> {
Err(tonic::Status::unimplemented("Not yet implemented"))
}
async fn list_actions(
&self,
_request: Request<Empty>,
) -> Result<Response<Self::ListActionsStream>, tonic::Status> {
Err(tonic::Status::unimplemented("Not yet implemented"))
}
async fn do_exchange(
&self,
_request: Request<Streaming<FlightData>>,
) -> Result<Response<Self::DoExchangeStream>, tonic::Status> {
Err(tonic::Status::unimplemented("Not yet implemented"))
}
}
#[pin_project(PinnedDrop)]
struct GetStream {
#[pin]
rx: futures::channel::mpsc::Receiver<Result<FlightData, tonic::Status>>,
join_handle: JoinHandle<()>,
done: bool,
}
impl GetStream {
async fn new(
ctx: IOxExecutionContext,
physical_plan: Arc<dyn ExecutionPlan>,
database_name: String,
) -> Result<Self, tonic::Status> {
// setup channel
let (mut tx, rx) = futures::channel::mpsc::channel::<Result<FlightData, tonic::Status>>(1);
// get schema
let schema = Arc::new(optimize_schema(&physical_plan.schema()));
// setup stream
let options = arrow::ipc::writer::IpcWriteOptions::default();
let schema_flight_data = SchemaAsIpc::new(&schema, &options).into();
let mut stream_record_batches = ctx
.execute_stream(Arc::clone(&physical_plan))
.await
.map_err(|e| Box::new(e) as _)
.context(QuerySnafu {
database_name: &database_name,
})?;
let join_handle = tokio::spawn(async move {
if tx.send(Ok(schema_flight_data)).await.is_err() {
// receiver gone
return;
}
while let Some(batch_or_err) = stream_record_batches.next().await {
match batch_or_err {
Ok(batch) => {
match optimize_record_batch(&batch, Arc::clone(&schema)) {
Ok(batch) => {
let (flight_dictionaries, flight_batch) =
arrow_flight::utils::flight_data_from_arrow_batch(
&batch, &options,
);
for dict in flight_dictionaries {
if tx.send(Ok(dict)).await.is_err() {
// receiver is gone
return;
}
}
if tx.send(Ok(flight_batch)).await.is_err() {
// receiver is gone
return;
}
}
Err(e) => {
// failure sending here is OK because we're cutting the stream anyways
tx.send(Err(e.into())).await.ok();
// end stream
return;
}
}
}
Err(e) => {
// failure sending here is OK because we're cutting the stream anyways
tx.send(Err(Error::Query {
database_name: database_name.clone(),
source: Box::new(e),
}
.into()))
.await
.ok();
// end stream
return;
}
}
}
});
Ok(Self {
rx,
join_handle,
done: false,
})
}
}
#[pinned_drop]
impl PinnedDrop for GetStream {
fn drop(self: Pin<&mut Self>) {
self.join_handle.abort();
}
}
impl Stream for GetStream {
type Item = Result<FlightData, tonic::Status>;
fn poll_next(
self: Pin<&mut Self>,
cx: &mut std::task::Context<'_>,
) -> std::task::Poll<Option<Self::Item>> {
let this = self.project();
if *this.done {
Poll::Ready(None)
} else {
match this.rx.poll_next(cx) {
Poll::Ready(None) => {
*this.done = true;
Poll::Ready(None)
}
e @ Poll::Ready(Some(Err(_))) => {
*this.done = true;
e
}
other => other,
}
}
}
}
/// Some batches are small slices of the underlying arrays.
/// At this stage we only know the number of rows in the record batch
/// and the sizes in bytes of the backing buffers of the column arrays.
/// There is no straight-forward relationship between these two quantities,
/// since some columns can host variable length data such as strings.
///
/// However we can apply a quick&dirty heuristic:
/// if the backing buffer is two orders of magnitudes bigger
/// than the number of rows in the result set, we assume
/// that deep-copying the record batch is cheaper than the and transfer costs.
///
/// Possible improvements: take the type of the columns into consideration
/// and perhaps sample a few element sizes (taking care of not doing more work
/// than to always copying the results in the first place).
///
/// Or we just fix this upstream in
/// arrow_flight::utils::flight_data_from_arrow_batch and re-encode the array
/// into a smaller buffer while we have to copy stuff around anyway.
///
/// See rationale and discussions about future improvements on
/// <https://github.com/influxdata/influxdb_iox/issues/1133>
fn optimize_record_batch(batch: &RecordBatch, schema: SchemaRef) -> Result<RecordBatch, Error> {
let max_buf_len = batch
.columns()
.iter()
.map(|a| a.get_array_memory_size())
.max()
.unwrap_or_default();
let columns: Result<Vec<_>, _> = batch
.columns()
.iter()
.map(|column| {
if matches!(column.data_type(), DataType::Dictionary(_, _)) {
hydrate_dictionary(column)
} else if max_buf_len > batch.num_rows() * 100 {
Ok(deep_clone_array(column))
} else {
Ok(Arc::clone(column))
}
})
.collect();
RecordBatch::try_new(schema, columns?).context(InvalidRecordBatchSnafu)
}
fn deep_clone_array(array: &ArrayRef) -> ArrayRef {
let mut mutable = MutableArrayData::new(vec![array.data()], false, 0);
mutable.extend(0, 0, array.len());
make_array(mutable.freeze())
}
/// Convert dictionary types to underlying types
/// See hydrate_dictionary for more information
fn optimize_schema(schema: &Schema) -> Schema {
let fields = schema
.fields()
.iter()
.map(|field| match field.data_type() {
DataType::Dictionary(_, value_type) => Field::new(
field.name(),
value_type.as_ref().clone(),
field.is_nullable(),
),
_ => field.clone(),
})
.collect();
Schema::new(fields)
}
/// Hydrates a dictionary to its underlying type
///
/// An IPC response, streaming or otherwise, defines its schema up front
/// which defines the mapping from dictionary IDs. It then sends these
/// dictionaries over the wire.
///
/// This requires identifying the different dictionaries in use, assigning
/// them IDs, and sending new dictionaries, delta or otherwise, when needed
///
/// This is tracked by #1318
///
/// For now we just hydrate the dictionaries to their underlying type
fn hydrate_dictionary(array: &ArrayRef) -> Result<ArrayRef, Error> {
match array.data_type() {
DataType::Dictionary(_, value) => {
arrow::compute::cast(array, value).context(DictionarySnafu)
}
_ => unreachable!("not a dictionary"),
}
}
#[cfg(test)]
mod tests {
use std::sync::Arc;
use arrow::array::StringArray;
use arrow::{
array::{DictionaryArray, UInt32Array},
datatypes::{DataType, Int32Type},
};
use arrow_flight::utils::flight_data_to_arrow_batch;
use datafusion::physical_plan::limit::truncate_batch;
use super::*;
#[test]
fn test_deep_clone_array() {
let mut builder = UInt32Array::builder(1000);
builder.append_slice(&[1, 2, 3, 4, 5, 6]).unwrap();
let array: ArrayRef = Arc::new(builder.finish());
assert_eq!(array.len(), 6);
let sliced = array.slice(0, 2);
assert_eq!(sliced.len(), 2);
let deep_cloned = deep_clone_array(&sliced);
assert!(sliced.data().get_array_memory_size() > deep_cloned.data().get_array_memory_size());
}
#[test]
fn test_encode_flight_data() {
let options = arrow::ipc::writer::IpcWriteOptions::default();
let c1 = UInt32Array::from(vec![1, 2, 3, 4, 5, 6]);
let batch = RecordBatch::try_from_iter(vec![("a", Arc::new(c1) as ArrayRef)])
.expect("cannot create record batch");
let schema = batch.schema();
let (_, baseline_flight_batch) =
arrow_flight::utils::flight_data_from_arrow_batch(&batch, &options);
let big_batch = truncate_batch(&batch, batch.num_rows() - 1);
let optimized_big_batch =
optimize_record_batch(&big_batch, Arc::clone(&schema)).expect("failed to optimize");
let (_, optimized_big_flight_batch) =
arrow_flight::utils::flight_data_from_arrow_batch(&optimized_big_batch, &options);
assert_eq!(
baseline_flight_batch.data_body.len(),
optimized_big_flight_batch.data_body.len()
);
let small_batch = truncate_batch(&batch, 1);
let optimized_small_batch =
optimize_record_batch(&small_batch, Arc::clone(&schema)).expect("failed to optimize");
let (_, optimized_small_flight_batch) =
arrow_flight::utils::flight_data_from_arrow_batch(&optimized_small_batch, &options);
assert!(
baseline_flight_batch.data_body.len() > optimized_small_flight_batch.data_body.len()
);
}
#[test]
fn test_encode_flight_data_dictionary() {
let options = arrow::ipc::writer::IpcWriteOptions::default();
let c1 = UInt32Array::from(vec![1, 2, 3, 4, 5, 6]);
let c2: DictionaryArray<Int32Type> = vec![
Some("foo"),
Some("bar"),
None,
Some("fiz"),
None,
Some("foo"),
]
.into_iter()
.collect();
let batch =
RecordBatch::try_from_iter(vec![("a", Arc::new(c1) as ArrayRef), ("b", Arc::new(c2))])
.expect("cannot create record batch");
let original_schema = batch.schema();
let optimized_schema = Arc::new(optimize_schema(&original_schema));
let optimized_batch = optimize_record_batch(&batch, Arc::clone(&optimized_schema)).unwrap();
let (_, flight_data) =
arrow_flight::utils::flight_data_from_arrow_batch(&optimized_batch, &options);
let batch =
flight_data_to_arrow_batch(&flight_data, Arc::clone(&optimized_schema), &[None, None])
.unwrap();
// Should hydrate string dictionary for transport
assert_eq!(optimized_schema.field(1).data_type(), &DataType::Utf8);
let array = batch
.column(1)
.as_any()
.downcast_ref::<StringArray>()
.unwrap();
let expected = StringArray::from(vec![
Some("foo"),
Some("bar"),
None,
Some("fiz"),
None,
Some("foo"),
]);
assert_eq!(array, &expected)
}
}

View File

@ -1,577 +1,24 @@
//! Implements the native gRPC IOx query API using Arrow Flight
use std::fmt::Debug;
use std::task::Poll;
use std::{pin::Pin, sync::Arc};
use std::sync::Arc;
use arrow::{
array::{make_array, ArrayRef, MutableArrayData},
datatypes::{DataType, Field, Schema, SchemaRef},
error::ArrowError,
record_batch::RecordBatch,
use arrow_flight::flight_service_server::{
FlightService as Flight, FlightServiceServer as FlightServer,
};
use arrow_flight::{
flight_service_server::{FlightService as Flight, FlightServiceServer as FlightServer},
Action, ActionType, Criteria, Empty, FlightData, FlightDescriptor, FlightInfo,
HandshakeRequest, HandshakeResponse, PutResult, SchemaAsIpc, SchemaResult, Ticket,
};
use datafusion::physical_plan::ExecutionPlan;
use futures::{SinkExt, Stream, StreamExt};
use pin_project::{pin_project, pinned_drop};
use query::QueryDatabase;
use serde::Deserialize;
use snafu::{ResultExt, Snafu};
use tokio::task::JoinHandle;
use tonic::{Request, Response, Streaming};
use data_types::{DatabaseName, DatabaseNameError};
use observability_deps::tracing::{info, warn};
use query::exec::{ExecutionContextProvider, IOxExecutionContext};
use data_types::DatabaseName;
use db::Db;
use server::Server;
use crate::influxdb_ioxd::rpc::flight::{make_server as make_server_inner, QueryDatabaseProvider};
use super::error::default_server_error_handler;
use crate::influxdb_ioxd::planner::Planner;
#[allow(clippy::enum_variant_names)]
#[derive(Debug, Snafu)]
pub enum Error {
#[snafu(display("Invalid ticket. Error: {:?} Ticket: {:?}", source, ticket))]
InvalidTicket {
source: std::string::FromUtf8Error,
ticket: Vec<u8>,
},
#[snafu(display("Invalid query, could not parse '{}': {}", query, source))]
InvalidQuery {
query: String,
source: serde_json::Error,
},
impl QueryDatabaseProvider for Server {
type Db = Db;
#[snafu(display("Database {} not found", database_name))]
DatabaseNotFound { database_name: String },
#[snafu(display(
"Internal error reading points from database {}: {}",
database_name,
source
))]
Query {
database_name: String,
source: Box<dyn std::error::Error + Send + Sync>,
},
#[snafu(display("Invalid database name: {}", source))]
InvalidDatabaseName { source: DatabaseNameError },
#[snafu(display("Invalid RecordBatch: {}", source))]
InvalidRecordBatch { source: ArrowError },
#[snafu(display("Failed to hydrate dictionary: {}", source))]
DictionaryError { source: ArrowError },
#[snafu(display("Error while planning query: {}", source))]
Planning {
source: crate::influxdb_ioxd::planner::Error,
},
}
pub type Result<T, E = Error> = std::result::Result<T, E>;
impl From<Error> for tonic::Status {
/// Converts a result from the business logic into the appropriate tonic
/// status
fn from(err: Error) -> Self {
// An explicit match on the Error enum will ensure appropriate
// logging is handled for any new error variants.
let msg = "Error handling Flight gRPC request";
match err {
Error::DatabaseNotFound { .. }
| Error::InvalidTicket { .. }
| Error::InvalidQuery { .. }
// TODO(edd): this should be `debug`. Keeping at info whilst IOx still in early development
| Error::InvalidDatabaseName { .. } => info!(?err, msg),
Error::Query { .. } => info!(?err, msg),
Error::DictionaryError { .. }
| Error::InvalidRecordBatch { .. }
| Error::Planning { .. } => warn!(?err, msg),
fn db(&self, db_name: &DatabaseName<'_>) -> Result<Arc<Self::Db>, tonic::Status> {
self.db(db_name).map_err(default_server_error_handler)
}
err.to_status()
}
}
impl Error {
/// Converts a result from the business logic into the appropriate tonic
/// status
fn to_status(&self) -> tonic::Status {
use tonic::Status;
match &self {
Self::InvalidTicket { .. } => Status::invalid_argument(self.to_string()),
Self::InvalidQuery { .. } => Status::invalid_argument(self.to_string()),
Self::DatabaseNotFound { .. } => Status::not_found(self.to_string()),
Self::Query { .. } => Status::internal(self.to_string()),
Self::InvalidDatabaseName { .. } => Status::invalid_argument(self.to_string()),
Self::InvalidRecordBatch { .. } => Status::internal(self.to_string()),
Self::Planning { .. } => Status::invalid_argument(self.to_string()),
Self::DictionaryError { .. } => Status::internal(self.to_string()),
}
}
}
type TonicStream<T> = Pin<Box<dyn Stream<Item = Result<T, tonic::Status>> + Send + Sync + 'static>>;
#[derive(Deserialize, Debug)]
/// Body of the `Ticket` serialized and sent to the do_get endpoint; this should
/// be shared with the read API probably...
struct ReadInfo {
database_name: String,
sql_query: String,
}
/// Concrete implementation of the gRPC Arrow Flight Service API
#[derive(Debug)]
struct FlightService {
server: Arc<Server>,
}
pub fn make_server(server: Arc<Server>) -> FlightServer<impl Flight> {
FlightServer::new(FlightService { server })
}
#[tonic::async_trait]
impl Flight for FlightService {
type HandshakeStream = TonicStream<HandshakeResponse>;
type ListFlightsStream = TonicStream<FlightInfo>;
type DoGetStream = TonicStream<FlightData>;
type DoPutStream = TonicStream<PutResult>;
type DoActionStream = TonicStream<arrow_flight::Result>;
type ListActionsStream = TonicStream<ActionType>;
type DoExchangeStream = TonicStream<FlightData>;
async fn get_schema(
&self,
_request: Request<FlightDescriptor>,
) -> Result<Response<SchemaResult>, tonic::Status> {
Err(tonic::Status::unimplemented("Not yet implemented"))
}
async fn do_get(
&self,
request: Request<Ticket>,
) -> Result<Response<Self::DoGetStream>, tonic::Status> {
let span_ctx = request.extensions().get().cloned();
let ticket = request.into_inner();
let json_str = String::from_utf8(ticket.ticket.to_vec()).context(InvalidTicketSnafu {
ticket: ticket.ticket,
})?;
let read_info: ReadInfo =
serde_json::from_str(&json_str).context(InvalidQuerySnafu { query: &json_str })?;
let database =
DatabaseName::new(&read_info.database_name).context(InvalidDatabaseNameSnafu)?;
let db = self
.server
.db(&database)
.map_err(default_server_error_handler)?;
let _query_completed_token = db.record_query("sql", Box::new(read_info.sql_query.clone()));
let ctx = db.new_query_context(span_ctx);
let physical_plan = Planner::new(&ctx)
.sql(&read_info.sql_query)
.await
.context(PlanningSnafu)?;
let output = GetStream::new(ctx, physical_plan, read_info.database_name).await?;
Ok(Response::new(Box::pin(output) as Self::DoGetStream))
}
async fn handshake(
&self,
request: Request<Streaming<HandshakeRequest>>,
) -> Result<Response<Self::HandshakeStream>, tonic::Status> {
let request = request.into_inner().message().await?.unwrap();
let response = HandshakeResponse {
protocol_version: request.protocol_version,
payload: request.payload,
};
let output = futures::stream::iter(std::iter::once(Ok(response)));
Ok(Response::new(Box::pin(output) as Self::HandshakeStream))
}
async fn list_flights(
&self,
_request: Request<Criteria>,
) -> Result<Response<Self::ListFlightsStream>, tonic::Status> {
Err(tonic::Status::unimplemented("Not yet implemented"))
}
async fn get_flight_info(
&self,
_request: Request<FlightDescriptor>,
) -> Result<Response<FlightInfo>, tonic::Status> {
Err(tonic::Status::unimplemented("Not yet implemented"))
}
async fn do_put(
&self,
_request: Request<Streaming<FlightData>>,
) -> Result<Response<Self::DoPutStream>, tonic::Status> {
Err(tonic::Status::unimplemented("Not yet implemented"))
}
async fn do_action(
&self,
_request: Request<Action>,
) -> Result<Response<Self::DoActionStream>, tonic::Status> {
Err(tonic::Status::unimplemented("Not yet implemented"))
}
async fn list_actions(
&self,
_request: Request<Empty>,
) -> Result<Response<Self::ListActionsStream>, tonic::Status> {
Err(tonic::Status::unimplemented("Not yet implemented"))
}
async fn do_exchange(
&self,
_request: Request<Streaming<FlightData>>,
) -> Result<Response<Self::DoExchangeStream>, tonic::Status> {
Err(tonic::Status::unimplemented("Not yet implemented"))
}
}
#[pin_project(PinnedDrop)]
struct GetStream {
#[pin]
rx: futures::channel::mpsc::Receiver<Result<FlightData, tonic::Status>>,
join_handle: JoinHandle<()>,
done: bool,
}
impl GetStream {
async fn new(
ctx: IOxExecutionContext,
physical_plan: Arc<dyn ExecutionPlan>,
database_name: String,
) -> Result<Self, tonic::Status> {
// setup channel
let (mut tx, rx) = futures::channel::mpsc::channel::<Result<FlightData, tonic::Status>>(1);
// get schema
let schema = Arc::new(optimize_schema(&physical_plan.schema()));
// setup stream
let options = arrow::ipc::writer::IpcWriteOptions::default();
let schema_flight_data = SchemaAsIpc::new(&schema, &options).into();
let mut stream_record_batches = ctx
.execute_stream(Arc::clone(&physical_plan))
.await
.map_err(|e| Box::new(e) as _)
.context(QuerySnafu {
database_name: &database_name,
})?;
let join_handle = tokio::spawn(async move {
if tx.send(Ok(schema_flight_data)).await.is_err() {
// receiver gone
return;
}
while let Some(batch_or_err) = stream_record_batches.next().await {
match batch_or_err {
Ok(batch) => {
match optimize_record_batch(&batch, Arc::clone(&schema)) {
Ok(batch) => {
let (flight_dictionaries, flight_batch) =
arrow_flight::utils::flight_data_from_arrow_batch(
&batch, &options,
);
for dict in flight_dictionaries {
if tx.send(Ok(dict)).await.is_err() {
// receiver is gone
return;
}
}
if tx.send(Ok(flight_batch)).await.is_err() {
// receiver is gone
return;
}
}
Err(e) => {
// failure sending here is OK because we're cutting the stream anyways
tx.send(Err(e.into())).await.ok();
// end stream
return;
}
}
}
Err(e) => {
// failure sending here is OK because we're cutting the stream anyways
tx.send(Err(Error::Query {
database_name: database_name.clone(),
source: Box::new(e),
}
.into()))
.await
.ok();
// end stream
return;
}
}
}
});
Ok(Self {
rx,
join_handle,
done: false,
})
}
}
#[pinned_drop]
impl PinnedDrop for GetStream {
fn drop(self: Pin<&mut Self>) {
self.join_handle.abort();
}
}
impl Stream for GetStream {
type Item = Result<FlightData, tonic::Status>;
fn poll_next(
self: Pin<&mut Self>,
cx: &mut std::task::Context<'_>,
) -> std::task::Poll<Option<Self::Item>> {
let this = self.project();
if *this.done {
Poll::Ready(None)
} else {
match this.rx.poll_next(cx) {
Poll::Ready(None) => {
*this.done = true;
Poll::Ready(None)
}
e @ Poll::Ready(Some(Err(_))) => {
*this.done = true;
e
}
other => other,
}
}
}
}
/// Some batches are small slices of the underlying arrays.
/// At this stage we only know the number of rows in the record batch
/// and the sizes in bytes of the backing buffers of the column arrays.
/// There is no straight-forward relationship between these two quantities,
/// since some columns can host variable length data such as strings.
///
/// However we can apply a quick&dirty heuristic:
/// if the backing buffer is two orders of magnitudes bigger
/// than the number of rows in the result set, we assume
/// that deep-copying the record batch is cheaper than the and transfer costs.
///
/// Possible improvements: take the type of the columns into consideration
/// and perhaps sample a few element sizes (taking care of not doing more work
/// than to always copying the results in the first place).
///
/// Or we just fix this upstream in
/// arrow_flight::utils::flight_data_from_arrow_batch and re-encode the array
/// into a smaller buffer while we have to copy stuff around anyway.
///
/// See rationale and discussions about future improvements on
/// <https://github.com/influxdata/influxdb_iox/issues/1133>
fn optimize_record_batch(batch: &RecordBatch, schema: SchemaRef) -> Result<RecordBatch, Error> {
let max_buf_len = batch
.columns()
.iter()
.map(|a| a.get_array_memory_size())
.max()
.unwrap_or_default();
let columns: Result<Vec<_>, _> = batch
.columns()
.iter()
.map(|column| {
if matches!(column.data_type(), DataType::Dictionary(_, _)) {
hydrate_dictionary(column)
} else if max_buf_len > batch.num_rows() * 100 {
Ok(deep_clone_array(column))
} else {
Ok(Arc::clone(column))
}
})
.collect();
RecordBatch::try_new(schema, columns?).context(InvalidRecordBatchSnafu)
}
fn deep_clone_array(array: &ArrayRef) -> ArrayRef {
let mut mutable = MutableArrayData::new(vec![array.data()], false, 0);
mutable.extend(0, 0, array.len());
make_array(mutable.freeze())
}
/// Convert dictionary types to underlying types
/// See hydrate_dictionary for more information
fn optimize_schema(schema: &Schema) -> Schema {
let fields = schema
.fields()
.iter()
.map(|field| match field.data_type() {
DataType::Dictionary(_, value_type) => Field::new(
field.name(),
value_type.as_ref().clone(),
field.is_nullable(),
),
_ => field.clone(),
})
.collect();
Schema::new(fields)
}
/// Hydrates a dictionary to its underlying type
///
/// An IPC response, streaming or otherwise, defines its schema up front
/// which defines the mapping from dictionary IDs. It then sends these
/// dictionaries over the wire.
///
/// This requires identifying the different dictionaries in use, assigning
/// them IDs, and sending new dictionaries, delta or otherwise, when needed
///
/// This is tracked by #1318
///
/// For now we just hydrate the dictionaries to their underlying type
fn hydrate_dictionary(array: &ArrayRef) -> Result<ArrayRef, Error> {
match array.data_type() {
DataType::Dictionary(_, value) => {
arrow::compute::cast(array, value).context(DictionarySnafu)
}
_ => unreachable!("not a dictionary"),
}
}
#[cfg(test)]
mod tests {
use std::sync::Arc;
use arrow::array::StringArray;
use arrow::{
array::{DictionaryArray, UInt32Array},
datatypes::{DataType, Int32Type},
};
use arrow_flight::utils::flight_data_to_arrow_batch;
use datafusion::physical_plan::limit::truncate_batch;
use super::*;
#[test]
fn test_deep_clone_array() {
let mut builder = UInt32Array::builder(1000);
builder.append_slice(&[1, 2, 3, 4, 5, 6]).unwrap();
let array: ArrayRef = Arc::new(builder.finish());
assert_eq!(array.len(), 6);
let sliced = array.slice(0, 2);
assert_eq!(sliced.len(), 2);
let deep_cloned = deep_clone_array(&sliced);
assert!(sliced.data().get_array_memory_size() > deep_cloned.data().get_array_memory_size());
}
#[test]
fn test_encode_flight_data() {
let options = arrow::ipc::writer::IpcWriteOptions::default();
let c1 = UInt32Array::from(vec![1, 2, 3, 4, 5, 6]);
let batch = RecordBatch::try_from_iter(vec![("a", Arc::new(c1) as ArrayRef)])
.expect("cannot create record batch");
let schema = batch.schema();
let (_, baseline_flight_batch) =
arrow_flight::utils::flight_data_from_arrow_batch(&batch, &options);
let big_batch = truncate_batch(&batch, batch.num_rows() - 1);
let optimized_big_batch =
optimize_record_batch(&big_batch, Arc::clone(&schema)).expect("failed to optimize");
let (_, optimized_big_flight_batch) =
arrow_flight::utils::flight_data_from_arrow_batch(&optimized_big_batch, &options);
assert_eq!(
baseline_flight_batch.data_body.len(),
optimized_big_flight_batch.data_body.len()
);
let small_batch = truncate_batch(&batch, 1);
let optimized_small_batch =
optimize_record_batch(&small_batch, Arc::clone(&schema)).expect("failed to optimize");
let (_, optimized_small_flight_batch) =
arrow_flight::utils::flight_data_from_arrow_batch(&optimized_small_batch, &options);
assert!(
baseline_flight_batch.data_body.len() > optimized_small_flight_batch.data_body.len()
);
}
#[test]
fn test_encode_flight_data_dictionary() {
let options = arrow::ipc::writer::IpcWriteOptions::default();
let c1 = UInt32Array::from(vec![1, 2, 3, 4, 5, 6]);
let c2: DictionaryArray<Int32Type> = vec![
Some("foo"),
Some("bar"),
None,
Some("fiz"),
None,
Some("foo"),
]
.into_iter()
.collect();
let batch =
RecordBatch::try_from_iter(vec![("a", Arc::new(c1) as ArrayRef), ("b", Arc::new(c2))])
.expect("cannot create record batch");
let original_schema = batch.schema();
let optimized_schema = Arc::new(optimize_schema(&original_schema));
let optimized_batch = optimize_record_batch(&batch, Arc::clone(&optimized_schema)).unwrap();
let (_, flight_data) =
arrow_flight::utils::flight_data_from_arrow_batch(&optimized_batch, &options);
let batch =
flight_data_to_arrow_batch(&flight_data, Arc::clone(&optimized_schema), &[None, None])
.unwrap();
// Should hydrate string dictionary for transport
assert_eq!(optimized_schema.field(1).data_type(), &DataType::Utf8);
let array = batch
.column(1)
.as_any()
.downcast_ref::<StringArray>()
.unwrap();
let expected = StringArray::from(vec![
Some("foo"),
Some("bar"),
None,
Some("fiz"),
None,
Some("foo"),
]);
assert_eq!(array, &expected)
}
make_server_inner(server)
}