Very good proposal.
However I don't see how the isolation is guaranteed between concurrent
transactions in this proposal.
Let's say there is one row in the table
k1 = 1, v1 = 1
There are two concurrent SQL executions, both of which execute the
following statement
"update my_table set v1 = v1 + 1 where k1 = 1"
If both statements are executed successfully, the result should be
k1 = 1, v1 = 3
I would like to ask what means you intend to ensure isolation between
concurrent transactions.
Thanks,
Zhao Chun
ling miao <lingm...@apache.org> 于2021年1月20日周三 下午3:51写道：
> Hi Zhengguo,
>
> This is a very valuable feature.
>
> Before Doris supports syntax of update, although Doris can also update data
> indirectly through the uniq model, it is very cumbersome to use.
> The user is required to retain the full amount of new data.
>
> If necessary, I have a better understanding of the query layer, and I can
> do development to support the update function.
>
> Ling Miao
>
> 寒江雪 <yangz...@gmail.com> 于2021年1月20日周三 下午3:14写道：
>
> > ## Background
> >
> > The current Doris table does not support the update operation, but there
> > are many scenarios where the data needs to be updated. Due to the batch
> > delete function we implemented before, it paved the way for the update.
> >
> > ## Syntax
> >
> > We only support single table update
> >
> > ```
> > UPDATE table_name
> > SET assignment_list
> > [WHERE where_condition]
> > [ORDER BY...]
> > [LIMIT row_count]
> >
> > value:
> > {expr | DEFAULT}
> >
> > assignment:
> > col_name = value
> >
> > assignment_list:
> > assignment [, assignment].
> > ```
> >
> > ## Design
> >
> > There are two kinds of updates involved here, one that only contains the
> > value column, and the other that contains the key column.
> >
> > * Include and update of value column
> >
> > This method is relatively simple. You only need to plan the update
> > statement into a query plan like `select * from table insert into table`
> > during query planning, and you need to add an update node to modify the
> > data to the updated value, or plan Into a statement similar to `select a,
> > b,'xx','xx1' from table insert into table` and hand it directly to the
> > tableSink node
> >
> > * Contains the update of the key column
> >
> > This kind of relatively complicated, we need to divide the query into
> two
> > parts, and need to use the MERGE semantics in the batch delete, first
> plan
> > it into a query plan similar to `select * from table insert into table`,
> > and then generate it first for the modification of the key column One
> piece
> > of data that needs to be deleted, and then replace the value of the key
> > column to generate new data that needs to be added, mainly generating two
> > pieces of imported data from one piece of original data. Need update node
> > to have the ability to repeat data
> >
> > ## Subtasks
> >
> > 1. Add index to unique table value column to speed up data scanning
> > 2. FE supports update statements and generates corresponding query plans.
> > Be implements update node, which can be divided into two parts here
> > 1. Support updates that only contain the value column
> > 2. Support the update of the key column
> >
>
