[
https://issues.apache.org/jira/browse/NIFI-3704?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15975828#comment-15975828
]
ASF GitHub Bot commented on NIFI-3704:
--------------------------------------
Github user ijokarumawak commented on the issue:
https://github.com/apache/nifi/pull/1677
@mattyb149 Thanks for the updated commit. I confirmed that my comments are
incorporated.
It works as expected in most cases, however, I found a case which needs to
be addressed.
Let's say I have a table with a primary key like this:
```sql
create table tutorials_tbl(
tutorial_id INT NOT NULL,
tutorial_title VARCHAR(100) NOT NULL,
tutorial_author VARCHAR(40) NOT NULL,
submission_date DATE,
PRIMARY KEY ( tutorial_id )
);
```
And insert some rows, then update a row, especially update its primary key:
```sql
update tutorials_tbl set tutorial_id = 110 where tutorial_id = 11;
```
Above update query generates following JSON via CaptureChangeMySQL:
```json
{ "type" : "update",
"timestamp" : 1492648209000,
"binlog_filename" : "mysql-server-bin.000004",
"binlog_position" : 97152,
"database" : "nifi_test",
"table_name" : "tutorials_tbl",
"table_id" : 222,
"columns" : [ {
"id" : 1, "name" : "tutorial_id", "column_type" : 4,
"last_value" : 11, "value" : 110
}, {
"id" : 2, "name" : "tutorial_title", "column_type" : 12,
"last_value" : "11th", "value" : "11th"
}, {
"id" : 3, "name" : "tutorial_author", "column_type" : 12,
"last_value" : "koji", "value" : "koji"
}, {
"id" : 4, "name" : "submission_date", "column_type" : 91,
"last_value" : null, "value" : null
} ]}
```
`Transform to Flat JSON` (JoltTransform) flattens the event JSON as below.
At this point, the record image before update is dropped:
```json
[ { "tutorial_id" : 110, "tutorial_title" : "11th", "tutorial_author" :
"koji", "submission_date" : null } ]
```
Finally, PutDatabaseRecord generates an update sql statement with `where
tutorial_id = 110`. But it doesn't update anything, because it should have used
`where tutorial_id = 11` with before update row image.
We might be able to handle this by generating two delete and insert records
in a NiFi flow, or do something smart at PutDatabaseRecord.
How do you think?
> Add PutDatabaseRecord processor
> -------------------------------
>
> Key: NIFI-3704
> URL: https://issues.apache.org/jira/browse/NIFI-3704
> Project: Apache NiFi
> Issue Type: New Feature
> Components: Extensions
> Reporter: Matt Burgess
> Assignee: Matt Burgess
> Fix For: 1.2.0
>
>
> With the inclusion of NIFI-1280, which added Controller Services for
> RecordReaders and RecordWriters, we could now support a processor that reads
> records in, generates SQL statements for those records (with a specified verb
> such as INSERT, UPDATE, DELETE, etc.), and can execute all the records in one
> flow file as a batch. This would allow the processor to use a single
> PreparedStatement and, for a flow file containing multiple records, would be
> able to execute them all at once. This is in contrast to PutSQL which handles
> batches across flow files (if fragmented transactions are enabled) or with a
> discrete set (by taking at most a specified number of flow files at a time).
> This processor (called PutDatabaseRecord) would effectively act like the
> combination of ConvertJSONToSQL and PutSQL, with the added features of being
> able to take records in an arbitrary format (given that there is a
> RecordReader implementation for that format) such as Avro, JSON, CSV, etc.
> and execute all the statements for the flow file at once.
> Another improvement upon what can be done in ConvertJSONToSQL would be to
> support BEGIN, COMMIT, and SQL verbs. This could be accomplished by adding an
> AllowableValue to the dropdown, letting the user select "Use statement.type
> Attribute". If this was selected, then the verb would be expected to be in
> the value of the "statement.type" attribute of the incoming flow file. Note
> that this may supercede or deprecate the need for NIFI-3676, unless this
> capability is also desired for that processor.
> For BEGIN and COMMIT verbs, the contents of the record(s) are not needed, as
> the type itself should be enough to generate the appropriate SQL commands.
> For the "SQL" Statement type, the processor could either expect the flow file
> to contain a SQL statement (so the RecordReader would not be used), or it
> could expect a field called "sql" that contains the SQL statement as its
> value.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
