feat: frontend instance (#238)

* feat: frontend instance

* no need to carry column length in `Column` proto

* add more tests

* rebase develop

* create a new variant with already provisioned RecordBatches in Output

* resolve code review comments

* new frontend instance does not connect datanode grpc

* add more tests

* add more tests

* rebase develop

Co-authored-by: luofucong <luofucong@greptime.com>
This commit is contained in:
LFC
2022-09-13 17:10:22 +08:00
committed by GitHub
parent bdd5bdd917
commit ec99eb0cd0
71 changed files with 2324 additions and 362 deletions

View File

@@ -7,13 +7,21 @@ edition = "2021"
[dependencies]
api = { path = "../api" }
async-stream = "0.3"
catalog = { path = "../catalog" }
common-base = { path = "../common/base" }
common-error = { path = "../common/error" }
common-grpc = { path = "../common/grpc" }
common-recordbatch = { path = "../common/recordbatch" }
common-time = { path = "../common/time" }
datafusion = { git = "https://github.com/apache/arrow-datafusion.git", branch = "arrow2", features = ["simd"] }
datatypes = { path = "../datatypes" }
query = { path = "../query" }
snafu = { version = "0.7", features = ["backtraces"] }
tonic = "0.8"
[dev-dependencies]
datanode = { path = "../datanode" }
tokio = { version = "1.0", features = ["full"] }
tracing = "0.1"
tracing-subscriber = { version = "0.3", features = ["env-filter"] }

View File

@@ -13,7 +13,13 @@ async fn run() {
let client = Client::connect("http://127.0.0.1:3001").await.unwrap();
let db = Database::new("greptime", client);
db.insert("demo", insert_batches()).await.unwrap();
let expr = InsertExpr {
table_name: "demo".to_string(),
expr: Some(insert_expr::Expr::Values(insert_expr::Values {
values: insert_batches(),
})),
};
db.insert(expr).await.unwrap();
}
fn insert_batches() -> Vec<Vec<u8>> {
@@ -37,6 +43,7 @@ fn insert_batches() -> Vec<Vec<u8>> {
semantic_type: SEMANTIC_TAG,
values: Some(host_vals),
null_mask: vec![0],
..Default::default()
};
let cpu_vals = column::Values {
@@ -48,6 +55,7 @@ fn insert_batches() -> Vec<Vec<u8>> {
semantic_type: SEMANTIC_FEILD,
values: Some(cpu_vals),
null_mask: vec![2],
..Default::default()
};
let mem_vals = column::Values {
@@ -59,6 +67,7 @@ fn insert_batches() -> Vec<Vec<u8>> {
semantic_type: SEMANTIC_FEILD,
values: Some(mem_vals),
null_mask: vec![4],
..Default::default()
};
let ts_vals = column::Values {
@@ -70,6 +79,7 @@ fn insert_batches() -> Vec<Vec<u8>> {
semantic_type: SEMANTIC_TS,
values: Some(ts_vals),
null_mask: vec![0],
..Default::default()
};
let insert_batch = InsertBatch {

View File

@@ -1,4 +1,6 @@
use api::v1::*;
use common_error::prelude::StatusCode;
use query::Output;
use snafu::prelude::*;
use crate::database::PROTOCOL_VERSION;
@@ -20,6 +22,10 @@ impl Admin {
}
}
pub async fn start(&mut self, url: impl Into<String>) -> Result<()> {
self.client.start(url).await
}
pub async fn create(&self, expr: CreateExpr) -> Result<AdminResult> {
let header = ExprHeader {
version: PROTOCOL_VERSION,
@@ -28,8 +34,12 @@ impl Admin {
header: Some(header),
expr: Some(admin_expr::Expr::Create(expr)),
};
// `remove(0)` is safe because of `do_request`'s invariants.
Ok(self.do_request(vec![expr]).await?.remove(0))
self.do_request(expr).await
}
pub async fn do_request(&self, expr: AdminExpr) -> Result<AdminResult> {
// `remove(0)` is safe because of `do_requests`'s invariants.
Ok(self.do_requests(vec![expr]).await?.remove(0))
}
pub async fn alter(&self, expr: AlterExpr) -> Result<AdminResult> {
@@ -40,11 +50,11 @@ impl Admin {
header: Some(header),
expr: Some(admin_expr::Expr::Alter(expr)),
};
Ok(self.do_request(vec![expr]).await?.remove(0))
Ok(self.do_requests(vec![expr]).await?.remove(0))
}
/// Invariants: the lengths of input vec (`Vec<AdminExpr>`) and output vec (`Vec<AdminResult>`) are equal.
async fn do_request(&self, exprs: Vec<AdminExpr>) -> Result<Vec<AdminResult>> {
async fn do_requests(&self, exprs: Vec<AdminExpr>) -> Result<Vec<AdminResult>> {
let expr_count = exprs.len();
let req = AdminRequest {
name: self.name.clone(),
@@ -65,3 +75,32 @@ impl Admin {
Ok(results)
}
}
pub fn admin_result_to_output(admin_result: AdminResult) -> Result<Output> {
let header = admin_result.header.context(error::MissingHeaderSnafu)?;
if !StatusCode::is_success(header.code) {
return error::DatanodeSnafu {
code: header.code,
msg: header.err_msg,
}
.fail();
}
let result = admin_result.result.context(error::MissingResultSnafu {
name: "result".to_string(),
expected: 1_usize,
actual: 0_usize,
})?;
let output = match result {
admin_result::Result::Mutate(mutate) => {
if mutate.failure != 0 {
return error::MutateFailureSnafu {
failure: mutate.failure,
}
.fail();
}
Output::AffectedRows(mutate.success as usize)
}
};
Ok(output)
}

View File

@@ -5,18 +5,43 @@ use tonic::transport::Channel;
use crate::error;
use crate::Result;
#[derive(Clone, Debug)]
#[derive(Clone, Debug, Default)]
pub struct Client {
client: GreptimeClient<Channel>,
client: Option<GreptimeClient<Channel>>,
}
impl Client {
pub async fn start(&mut self, url: impl Into<String>) -> Result<()> {
match self.client.as_ref() {
None => {
let url = url.into();
let client = GreptimeClient::connect(url.clone())
.await
.context(error::ConnectFailedSnafu { url })?;
self.client = Some(client);
Ok(())
}
Some(_) => error::IllegalGrpcClientStateSnafu {
err_msg: "already started",
}
.fail(),
}
}
pub fn with_client(client: GreptimeClient<Channel>) -> Self {
Self {
client: Some(client),
}
}
pub async fn connect(url: impl Into<String>) -> Result<Self> {
let url = url.into();
let client = GreptimeClient::connect(url.clone())
.await
.context(error::ConnectFailedSnafu { url })?;
Ok(Self { client })
Ok(Self {
client: Some(client),
})
}
pub async fn admin(&self, req: AdminRequest) -> Result<AdminResponse> {
@@ -48,12 +73,18 @@ impl Client {
}
pub async fn batch(&self, req: BatchRequest) -> Result<BatchResponse> {
let res = self
.client
.clone()
.batch(req)
.await
.context(error::TonicStatusSnafu)?;
Ok(res.into_inner())
if let Some(client) = self.client.as_ref() {
let res = client
.clone()
.batch(req)
.await
.context(error::TonicStatusSnafu)?;
Ok(res.into_inner())
} else {
error::IllegalGrpcClientStateSnafu {
err_msg: "not started",
}
.fail()
}
}
}

View File

@@ -1,15 +1,24 @@
use std::sync::Arc;
use api::helper::ColumnDataTypeWrapper;
use api::v1::codec::SelectResult as GrpcSelectResult;
use api::v1::{
object_expr, object_result, select_expr, DatabaseRequest, ExprHeader, InsertExpr,
MutateResult as GrpcMutateResult, ObjectExpr, ObjectResult as GrpcObjectResult, PhysicalPlan,
SelectExpr,
column::Values, object_expr, object_result, select_expr, Column, ColumnDataType,
DatabaseRequest, ExprHeader, InsertExpr, MutateResult as GrpcMutateResult, ObjectExpr,
ObjectResult as GrpcObjectResult, PhysicalPlan, SelectExpr,
};
use common_base::BitVec;
use common_error::status_code::StatusCode;
use common_grpc::AsExcutionPlan;
use common_grpc::DefaultAsPlanImpl;
use common_recordbatch::{RecordBatch, RecordBatches};
use common_time::date::Date;
use common_time::datetime::DateTime;
use common_time::timestamp::Timestamp;
use datafusion::physical_plan::ExecutionPlan;
use datatypes::prelude::*;
use datatypes::schema::{ColumnSchema, Schema};
use query::Output;
use snafu::{ensure, OptionExt, ResultExt};
use crate::error;
@@ -19,8 +28,6 @@ use crate::{
pub const PROTOCOL_VERSION: u32 = 1;
pub type Bytes = Vec<u8>;
#[derive(Clone, Debug)]
pub struct Database {
name: String,
@@ -35,26 +42,23 @@ impl Database {
}
}
pub async fn start(&mut self, url: impl Into<String>) -> Result<()> {
self.client.start(url).await
}
pub fn name(&self) -> &str {
&self.name
}
pub async fn insert(&self, table: impl Into<String>, values: Vec<Bytes>) -> Result<()> {
pub async fn insert(&self, insert: InsertExpr) -> Result<ObjectResult> {
let header = ExprHeader {
version: PROTOCOL_VERSION,
};
let insert = InsertExpr {
table_name: table.into(),
values,
};
let expr = ObjectExpr {
header: Some(header),
expr: Some(object_expr::Expr::Insert(insert)),
};
self.object(expr).await?;
Ok(())
self.object(expr).await?.try_into()
}
pub async fn select(&self, expr: Select) -> Result<ObjectResult> {
@@ -100,7 +104,7 @@ impl Database {
// TODO(jiachun) update/delete
async fn object(&self, expr: ObjectExpr) -> Result<GrpcObjectResult> {
pub async fn object(&self, expr: ObjectExpr) -> Result<GrpcObjectResult> {
let res = self.objects(vec![expr]).await?.pop().unwrap();
Ok(res)
}
@@ -165,3 +169,234 @@ impl TryFrom<api::v1::ObjectResult> for ObjectResult {
pub enum Select {
Sql(String),
}
impl TryFrom<ObjectResult> for Output {
type Error = error::Error;
fn try_from(value: ObjectResult) -> Result<Self> {
let output = match value {
ObjectResult::Select(select) => {
let vectors = select
.columns
.iter()
.map(|column| column_to_vector(column, select.row_count))
.collect::<Result<Vec<VectorRef>>>()?;
let column_schemas = select
.columns
.iter()
.zip(vectors.iter())
.map(|(column, vector)| {
let datatype = vector.data_type();
// nullable or not, does not affect the output
ColumnSchema::new(&column.column_name, datatype, true)
})
.collect::<Vec<ColumnSchema>>();
let schema = Arc::new(Schema::new(column_schemas));
let recordbatches = RecordBatch::new(schema, vectors)
.and_then(|batch| RecordBatches::try_new(batch.schema.clone(), vec![batch]))
.context(error::CreateRecordBatchesSnafu)?;
Output::RecordBatches(recordbatches)
}
ObjectResult::Mutate(mutate) => {
if mutate.failure != 0 {
return error::MutateFailureSnafu {
failure: mutate.failure,
}
.fail();
}
Output::AffectedRows(mutate.success as usize)
}
};
Ok(output)
}
}
fn column_to_vector(column: &Column, rows: u32) -> Result<VectorRef> {
let wrapper =
ColumnDataTypeWrapper::try_new(column.datatype).context(error::ColumnDataTypeSnafu)?;
let column_datatype = wrapper.datatype();
let rows = rows as usize;
let mut vector = VectorBuilder::with_capacity(wrapper.into(), rows);
if let Some(values) = &column.values {
let values = collect_column_values(column_datatype, values);
let mut values_iter = values.into_iter();
let null_mask = BitVec::from_slice(&column.null_mask);
let mut nulls_iter = null_mask.iter().by_vals().fuse();
for i in 0..rows {
if let Some(true) = nulls_iter.next() {
vector.push_null();
} else {
let value_ref = values_iter.next().context(error::InvalidColumnProtoSnafu {
err_msg: format!(
"value not found at position {} of column {}",
i, &column.column_name
),
})?;
vector
.try_push_ref(value_ref)
.context(error::CreateVectorSnafu)?;
}
}
} else {
(0..rows).for_each(|_| vector.push_null());
}
Ok(vector.finish())
}
fn collect_column_values(column_datatype: ColumnDataType, values: &Values) -> Vec<ValueRef> {
macro_rules! collect_values {
($value: expr, $mapper: expr) => {
$value.iter().map($mapper).collect::<Vec<ValueRef>>()
};
}
match column_datatype {
ColumnDataType::Boolean => collect_values!(values.bool_values, |v| ValueRef::from(*v)),
ColumnDataType::Int8 => collect_values!(values.i8_values, |v| ValueRef::from(*v as i8)),
ColumnDataType::Int16 => {
collect_values!(values.i16_values, |v| ValueRef::from(*v as i16))
}
ColumnDataType::Int32 => {
collect_values!(values.i32_values, |v| ValueRef::from(*v))
}
ColumnDataType::Int64 => {
collect_values!(values.i64_values, |v| ValueRef::from(*v as i64))
}
ColumnDataType::Uint8 => {
collect_values!(values.u8_values, |v| ValueRef::from(*v as u8))
}
ColumnDataType::Uint16 => {
collect_values!(values.u16_values, |v| ValueRef::from(*v as u16))
}
ColumnDataType::Uint32 => {
collect_values!(values.u32_values, |v| ValueRef::from(*v))
}
ColumnDataType::Uint64 => {
collect_values!(values.u64_values, |v| ValueRef::from(*v as u64))
}
ColumnDataType::Float32 => collect_values!(values.f32_values, |v| ValueRef::from(*v)),
ColumnDataType::Float64 => collect_values!(values.f64_values, |v| ValueRef::from(*v)),
ColumnDataType::Binary => {
collect_values!(values.binary_values, |v| ValueRef::from(v.as_slice()))
}
ColumnDataType::String => {
collect_values!(values.string_values, |v| ValueRef::from(v.as_str()))
}
ColumnDataType::Date => {
collect_values!(values.date_values, |v| ValueRef::Date(Date::new(*v)))
}
ColumnDataType::Datetime => {
collect_values!(values.datetime_values, |v| ValueRef::DateTime(
DateTime::new(*v)
))
}
ColumnDataType::Timestamp => {
collect_values!(values.ts_millis_values, |v| ValueRef::Timestamp(
Timestamp::from_millis(*v)
))
}
}
}
#[cfg(test)]
mod tests {
use datanode::server::grpc::select::{null_mask, values};
use datatypes::vectors::{
BinaryVector, BooleanVector, DateTimeVector, DateVector, Float32Vector, Float64Vector,
Int16Vector, Int32Vector, Int64Vector, Int8Vector, StringVector, UInt16Vector,
UInt32Vector, UInt64Vector, UInt8Vector,
};
use super::*;
#[test]
fn test_column_to_vector() {
let mut column = create_test_column(Arc::new(BooleanVector::from(vec![true])));
column.datatype = -100;
let result = column_to_vector(&column, 1);
assert!(result.is_err());
assert_eq!(
result.unwrap_err().to_string(),
"Column datatype error, source: Unknown proto column datatype: -100"
);
macro_rules! test_with_vector {
($vector: expr) => {
let vector = Arc::new($vector);
let column = create_test_column(vector.clone());
let result = column_to_vector(&column, vector.len() as u32).unwrap();
assert_eq!(result, vector as VectorRef);
};
}
test_with_vector!(BooleanVector::from(vec![Some(true), None, Some(false)]));
test_with_vector!(Int8Vector::from(vec![Some(i8::MIN), None, Some(i8::MAX)]));
test_with_vector!(Int16Vector::from(vec![
Some(i16::MIN),
None,
Some(i16::MAX)
]));
test_with_vector!(Int32Vector::from(vec![
Some(i32::MIN),
None,
Some(i32::MAX)
]));
test_with_vector!(Int64Vector::from(vec![
Some(i64::MIN),
None,
Some(i64::MAX)
]));
test_with_vector!(UInt8Vector::from(vec![Some(u8::MIN), None, Some(u8::MAX)]));
test_with_vector!(UInt16Vector::from(vec![
Some(u16::MIN),
None,
Some(u16::MAX)
]));
test_with_vector!(UInt32Vector::from(vec![
Some(u32::MIN),
None,
Some(u32::MAX)
]));
test_with_vector!(UInt64Vector::from(vec![
Some(u64::MIN),
None,
Some(u64::MAX)
]));
test_with_vector!(Float32Vector::from(vec![
Some(f32::MIN),
None,
Some(f32::MAX)
]));
test_with_vector!(Float64Vector::from(vec![
Some(f64::MIN),
None,
Some(f64::MAX)
]));
test_with_vector!(BinaryVector::from(vec![
Some(b"".to_vec()),
None,
Some(b"hello".to_vec())
]));
test_with_vector!(StringVector::from(vec![Some(""), None, Some("foo"),]));
test_with_vector!(DateVector::from(vec![Some(1), None, Some(3)]));
test_with_vector!(DateTimeVector::from(vec![Some(4), None, Some(6)]));
}
fn create_test_column(vector: VectorRef) -> Column {
let wrapper: ColumnDataTypeWrapper = vector.data_type().try_into().unwrap();
let array = vector.to_arrow_array();
Column {
column_name: "test".to_string(),
semantic_type: 1,
values: Some(values(&[array.clone()]).unwrap()),
null_mask: null_mask(&vec![array], vector.len()),
datatype: wrapper.datatype() as i32,
}
}
}

View File

@@ -1,3 +1,4 @@
use std::any::Any;
use std::sync::Arc;
use api::serde::DecodeError;
@@ -42,6 +43,67 @@ pub enum Error {
#[snafu(backtrace)]
source: common_grpc::Error,
},
#[snafu(display("Mutate result has failure {}", failure))]
MutateFailure { failure: u32, backtrace: Backtrace },
#[snafu(display("Invalid column proto: {}", err_msg))]
InvalidColumnProto {
err_msg: String,
backtrace: Backtrace,
},
#[snafu(display("Column datatype error, source: {}", source))]
ColumnDataType {
#[snafu(backtrace)]
source: api::error::Error,
},
#[snafu(display("Failed to create vector, source: {}", source))]
CreateVector {
#[snafu(backtrace)]
source: datatypes::error::Error,
},
#[snafu(display("Failed to create RecordBatches, source: {}", source))]
CreateRecordBatches {
#[snafu(backtrace)]
source: common_recordbatch::error::Error,
},
#[snafu(display("Illegal GRPC client state: {}", err_msg))]
IllegalGrpcClientState {
err_msg: String,
backtrace: Backtrace,
},
}
pub type Result<T> = std::result::Result<T, Error>;
impl ErrorExt for Error {
fn status_code(&self) -> StatusCode {
match self {
Error::ConnectFailed { .. }
| Error::MissingResult { .. }
| Error::MissingHeader { .. }
| Error::TonicStatus { .. }
| Error::DecodeSelect { .. }
| Error::Datanode { .. }
| Error::EncodePhysical { .. }
| Error::MutateFailure { .. }
| Error::InvalidColumnProto { .. }
| Error::ColumnDataType { .. } => StatusCode::Internal,
Error::CreateVector { source } => source.status_code(),
Error::CreateRecordBatches { source } => source.status_code(),
Error::IllegalGrpcClientState { .. } => StatusCode::Unexpected,
}
}
fn backtrace_opt(&self) -> Option<&Backtrace> {
ErrorCompat::backtrace(self)
}
fn as_any(&self) -> &dyn Any {
self
}
}