Help Center/
MapReduce Service/
Component Operation Guide (LTS) (Paris Region)/
Using Hudi/
Basic Operations/
Read/
Reading COW Table Views
Updated on 2022-12-14 GMT+08:00
Reading COW Table Views
- Reading the real-time view (using Hive and SparkSQL as an example): Directly read the Hudi table stored in Hive.
select count(*) from test;
- Reading the real-time view (using the Spark DataSource API as an example): This is similar to reading a common DataSource table.
QUERY_TYPE_OPT_KEY must be set to QUERY_TYPE_SNAPSHOT_OPT_VAL.
spark.read.format("hudi") .option(QUERY_TYPE_OPT_KEY, QUERY_TYPE_SNAPSHOT_OPT_VAL) // Set the query type to the real-time view. .load("/tmp/default/cow_bugx/*/*/*/*") // Set the path of the Hudi table to be read. The current table has three levels of partitions. .createTempView("mycall") spark.sql("select * from mycall").show(100)
- Reading the incremental view (using Hive and SparkSQL as an example):
set hoodie.test.consume.mode=INCREMENTAL; // Specify the incremental reading mode. set hoodie.test.consume.max.commits=3; // Specify the maximum number of commits to be consumed. set hoodie.test.consume.start.timestamp=20201227153030; // Specify the initial incremental pull commit. select count(*) from default.test where `_hoodie_commit_time`>'20201227153030'; // This filtering condition must be added, and the value is the initial incremental pull commit.
- Reading the incremental view (using the Spark DataSource API as an example):
QUERY_TYPE_OPT_KEY must be set to QUERY_TYPE_INCREMENTAL_OPT_VAL.
spark.read.format("hudi") .option(QUERY_TYPE_OPT_KEY, QUERY_TYPE_INCREMENTAL_OPT_VAL) // Set the query type to the incremental mode. .option(BEGIN_INSTANTTIME_OPT_KEY, "20210308212004") // Specify the initial incremental pull commit. .option(END_INSTANTTIME_OPT_KEY, "20210308212318") //: Specify the end commit of the incremental pull. .load("/tmp/default/cow_bugx/*/*/*/*") // Set the path of the Hudi table to be read. The current table has three levels of partitions. .createTempView("mycall") // Register as a Spark temporary table. spark.sql("select * from mycall where `_hoodie_commit_time`>'20210308211131'")// Start the query. The statement is the same as the Hive incremental query statement. .show(100, false)
- Reading the read-optimized view: The read-optimized view of COW tables is equivalent to the real-time view.
Parent topic: Read
Feedback
Was this page helpful?
Provide feedbackThank you very much for your feedback. We will continue working to improve the documentation.See the reply and handling status in My Cloud VOC.
The system is busy. Please try again later.
For any further questions, feel free to contact us through the chatbot.
Chatbot