[
https://issues.apache.org/jira/browse/NIFI-8249?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17289336#comment-17289336
]
David commented on NIFI-8249:
-----------------------------
Our result sets are fairly large: we are extracting 500M records from a source
table. When we encountered this scenario in production, our GenerateTableFetch
partition size was 100k, max rows per flow file = 100, fetch size = 100. We
fine tuned these parameters based on extensive testing in our performance
environment.
> ExecuteSQLRecord "administratively yields" and rollls back session when fetch
> size < query partition size and an SQLException is thrown on subsequent next()
> ------------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: NIFI-8249
> URL: https://issues.apache.org/jira/browse/NIFI-8249
> Project: Apache NiFi
> Issue Type: Bug
> Components: Core Framework
> Affects Versions: 1.11.4
> Reporter: David
> Assignee: Matt Burgess
> Priority: Major
> Attachments: ExecuteSQLRecord.png, GenerateTableFetch.png, Screen
> Shot 2021-02-23 at 3.00.47 PM.png, Screen Shot 2021-02-23 at 3.04.55 PM.png
>
>
> When ExecuteSQLRecord executes a query with a partition size > the fetch size
> and one of the fetches throws an SQLException, the exception appears to get
> "swallowed" and a more general "IOException: Could not obtain next record
> from ResultSet: routing to failure" in addition to a "transfer relationship
> not specified;Processor Administratively Yielded for 1 sec" errors are thrown.
> The problematic query/flowfile gets put back into the incoming queue and will
> continue to fail indefinitely until it is manually removed from the queue.
> If the fetch size >= to the query partition size, the exact SQL error message
> is thrown and the flowfile gets routed to the failure queue.
>
> Steps to reliably reproduce:
> # Create an DB table with an ID column (for partitioning queries against)
> and a VARCHAR column
> # Insert a row where the VARCHAR column data is a numeric value (ie '123')
> # Insert a row where the VARCHAR column data is a string value (ie 'abc')
> # Create a GenerateTableFetch processor where the "columns to return"
> includes TO_NUMBER(<varchar column>)
> # Create a ExecuteSQLRecord processor that accepts the "success" connection
> from GenerateTableFetch
> # Set ExecuteSQLRecord fetch size to 1
> # Run flow
> What should happen:
> * ExecuteSQLRecord should throw an java.sql.SQLSyntaxErrorException:
> ORA-01722: invalid number
> * Flowfile should get transferred to failure queue
> What happens:
> * Nifi throws one error and prints another, then puts the flowfile back into
> the incoming queue
> ** ExecuteSQLRecord throws "Unable to execute SQL select query <query> ...
> due to java.io.IOException: java.io.IOException: Could not obtain next record
> from ResultSet"
> ** Nifi prints errror "transfer relationship not specified; Processor
> Administratively Yielded for 1 sec"
> ** Flowfile is kept in current queue and penalized and re-processed again
> after penalty duration expires
> Thus, the problematic query will get executed indefinitely.
>
> Note: I believe the issue could be due to ResultSet .next() being called
> first in the constructor to ResultSetRecordSetWithCallback
> Subsequent .next() is called from resultSetWriter.write(recordSet)
> [https://github.com/apache/nifi/blob/aa741cc5967f62c3c38c2a47e712b7faa6fe19ff/nifi-nar-bundles/nifi-standard-bundle/nifi-standard-processors/src/main/java/org/apache/nifi/processors/standard/sql/RecordSqlWriter.java#L73|https://github.com/apache/nifi/blob/aa741cc5967f62c3c38c2a47e712b7faa6fe19ff/nifi-nar-bundles/nifi-standard-bundle/nifi-standard-processors/src/main/java/org/apache/nifi/processors/standard/sql/RecordSqlWriter.java#L73]
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
