Vineyard Rust SDK

[!注意] 需要Rust夜间版。Vineyard Rust SDK仍在开发中。API可能在将来发生变化。

连接到Vineyard

• 从环境变量 VINEYARD_IPC_SOCKET 解析UNIX域套接字

  use vineyard::client::*;
  
  let mut client = vineyard::default().unwrap();
  

• 或，使用显式参数

  use vineyard::client::*;
  
  let mut client = vineyard::connect("/var/run/vineyard.sock").unwrap();
  

与Vineyard交互

• 创建blob

  let mut blob_writer = client.create_blob(N)?;
  

• 获取对象

  let mut meta_writer = client.get::<DataFrame>(object_id)?;
  

与Python交互： numpy.ndarray

• Python

  import numpy as np
  import vineyard
  
  client = vineyard.connect()
  
  np_array = np.random.rand(10, 20).astype(np.int32)
  object_id = int(client.put(np_array))
  

• Rust

  let mut client = IPCClient::default()?;
  let tensor = client.get::<Int32Tensor>(object_id)?;
  assert_that!(tensor.shape().to_vec()).is_equal_to(vec![10, 20]);
  

与Python交互： pandas.DataFrame

• Python

  import pandas as pd
  import vineyard
  
  client = vineyard.connect()
  
  df = pd.DataFrame({'a': ["1", "2", "3", "4"], 'b': ["5", "6", "7", "8"]})
  object_id = int(client.put(df))
  

• Rust

  let mut client = IPCClient::default()?;
  let dataframe = client.get::<DataFrame>(object_id)?;
  assert_that!(dataframe.num_columns()).is_equal_to(2);
  assert_that!(dataframe.names().to_vec()).is_equal_to(vec!["a".into(), "b".into()]);
  for index in 0..dataframe.num_columns() {
      let column = dataframe.column(index);
      assert_that!(column.len()).is_equal_to(4);
  }
  

与Python交互： pyarrow.RecordBatch

• Python

  import pandas as pd
  import pyarrow as pa
  import vineyard
  
  client = vineyard.connect()
  
  arrays = [
      pa.array([1, 2, 3, 4]),
      pa.array(["foo", "bar", "baz", "qux"]),
      pa.array([3.0, 5.0, 7.0, 9.0]),
  ]
  batch = pa.RecordBatch.from_arrays(arrays, ["f0", "f1", "f2"])
  object_id = int(client.put(batch))
  

• Rust

  let batch = client.get::<RecordBatch>(object_id)?;
  assert_that!(batch.num_columns()).is_equal_to(3);
  assert_that!(batch.num_rows()).is_equal_to(4);
  let schema = batch.schema();
  let names = ["f0", "f1", "f2"];
  let recordbatch = batch.as_ref().as_ref();
  

与Python交互： pyarrow.Table

• Python

  batches = [batch] * 5
  table = pa.Table.from_batches(batches)
  object_id = int(client.put(table))
  

• Rust

  let mut client = IPCClient::default()?;
  let table = client.get::<Table>(object_id)?;
  assert_that!(table.num_batches()).is_equal_to(5);
  for batch in table.batches().iter() {
      // ...
  }
  

与Python交互： polars.DataFrame

• Python

  import polars
  
  dataframe = polars.DataFrame(table)
  object_id = int(client.put(dataframe))
  

• Rust

  use vineyard_polars::ds::dataframe::DataFrame;
  
  let mut client = IPCClient::default()?;
  let dataframe = client.get::<DataFrame>(object_id)?;
  let dataframe = dataframe.as_ref().as_ref();
  assert_that!(dataframe.width()).is_equal_to(3);
  for column in dataframe.get_columns() {
      // ...
  }
  

与Python交互： polars.DataFrame

• Python

  batches = [batch] * 5
  table = pa.Table.from_batches(batches)
  object_id = int(client.put(table))
  

• Rust

  use vineyard_datafusion::ds::dataframe::DataFrame;
  
  let mut client = IPCClient::default()?;
  let dataframe = client.get::<DataFrame>(object_id)?;
  
  let ctx = SessionContext::new();
  let table = ctx.read_table(dataframe.table_provider()).unwrap();
  
  assert_that!(block_on(table.count()).unwrap()).is_equal_to(1000);