The following table shows the basic process of inserting data into a table. It is divided into two parts: Data file writing and tabular meta-information submission. 

When the data file is written, the Flink data workers perform the following:

1. Read the data from the data source, row by row
2. Parse the data in the row according to the currently defined schema
3. Calculate the partition information
4. Write the row into the data file of the corresponding partition

When a checkpoint is reached, data workers switch the write to the new data file, complete the data file being written, and package all completed files into a list of files. The file list contains the path of the data file and its statistics, which are passed to a commit worker to make changes to the metadata of the table. When changes are made, the current table version must be read. All data-file information is merged with the table version number v006 to generate new table metadata and update the version number to v007 as shown in the following figure. This process is linear-consistent, and this document discusses the technical details of this mechanism later.

Figure 3.   Update new data

Other table modification operations are consistent with the process of inserting data. In this process, changes are made to the data file before the changes are submitted to the form.