Help Center/ Data Warehouse Service/ Developer Guide/ Developer Guide (9.1.0.x)/ DWS Performance Tuning/ SQL Tuning/ Automatic Retry upon SQL Statement Execution Errors
Updated on 2026-03-04 GMT+08:00
View PDF
Share

Automatic Retry upon SQL Statement Execution Errors

With automatic retry (referred to as CN retry), DWS retries an SQL statement when the execution of a statement fails. If an SQL statement sent from the gsql client, JDBC driver, or ODBC driver fails to be executed, the CN can automatically identify the error reported during execution and re-deliver the task to retry.

Constraints on CN Retry

Table 1 Constraints on CN retry

Constraint

Description

Function scope
  • CN retry increases execution success rate but does not guarantee success.
  • CN retry is enabled by default. In this case, the system records logs about temporary tables. If it is disabled, the system will not record the logs. Therefore, do not repeatedly enable and disable CN retry when temporary tables are used. Otherwise, data inconsistency may occur after a CN retry following a primary/standby switchover.
  • CN retry is enabled by default. In this case, the unlogged keyword is ignored in the statement for creating unlogged tables and thereby ordinary tables will be created by using this statement. If CN retry is disabled, the system records logs about unlogged tables. Therefore, do not repeatedly enable and disable CN retry when unlogged tables are used. Otherwise, data inconsistency may occur after a CN retry following a primary/standby switchover.
  • When GDS is used to export data, CN retry is supported. The existing mechanism checks for duplicate files and deletes duplicate files during data export. Therefore, you are advised not to repeatedly export data for the same foreign table unless you are sure that files with the same name in the data directory need to be deleted.

Error type

Only the error types in Table 2 are supported.

Statement type

Single-statement CN retry, stored procedures, functions, and anonymous blocks are supported. Statements in transaction blocks are not supported.

Stored procedure statement
  • If an error occurs during the execution of a stored procedure containing EXCEPTION (including statement block execution and statement execution in EXCEPTION), the stored procedure can be retried. If an internal error occurs, the stored procedure will retry first, but if the error is captured by EXCEPTION, the stored procedure cannot be retried.
  • Packages that use global variables are not supported.
  • DBMS_JOB is not supported.
  • UTL_FILE is not supported.
  • If the stored procedure has printed information (such as dbms_output.put_line or raise info), the printed information will be output repeatedly when retry occurs, and "Notice: Retry triggered, some message may be duplicated. " will be output before the repeated information.

Cluster status
  • Only DNs or GTMs are faulty.
  • The cluster can be recovered before the number of CN retries reaches the allowed maximum (controlled by max_query_retry_times). Otherwise, CN retry may fail.
  • CN retry is not supported during scale-out.

Data import
  • The COPY FROM STDIN statement is not supported.
  • The gsql \copy from metacommand is not supported.
  • JDBC CopyManager copyIn is not supported.

Error Types Supported by CN Retry

Table 2 lists the error types supported by CN retry and the corresponding error codes. You can use the GUC parameter retry_ecode_list to set the list of error types supported by CN retry. You are not advised to modify this parameter. To modify it, contact technical support.

Table 2 Error types supported by CN retry

Error Type

Error Code

Remarks

CONNECTION_RESET_BY_PEER

YY001

TCP communication errors: Connection reset by peer (communication between the CN and DNs)

STREAM_CONNECTION_RESET_BY_PEER

YY002

TCP communication errors: Stream connection reset by peer (communication between DNs)

LOCK_WAIT_TIMEOUT

YY003

Lock wait timeout

CONNECTION_TIMED_OUT

YY004

TCP communication errors: Connection timed out

SET_QUERY_ERROR

YY005

Failed to deliver the SET command: Set query

OUT_OF_LOGICAL_MEMORY

YY006

Failed to apply for memory: Out of logical memory

SCTP_MEMORY_ALLOC

YY007

SCTP communication errors: Memory allocate error

SCTP_NO_DATA_IN_BUFFER

YY008

SCTP communication errors: SCTP no data in buffer

SCTP_RELEASE_MEMORY_CLOSE

YY009

SCTP communication errors: Release memory close

SCTP_TCP_DISCONNECT

YY010

SCTP communication errors: TCP disconnect

SCTP_DISCONNECT

YY011

SCTP communication errors: SCTP disconnect

SCTP_REMOTE_CLOSE

YY012

SCTP communication errors: Stream closed by remote

SCTP_WAIT_POLL_UNKNOWN

YY013

Waiting for an unknown poll: SCTP wait poll unknown

SNAPSHOT_INVALID

YY014

Snapshot invalid

ERRCODE_CONNECTION_RECEIVE_WRONG

YY015

Connection receive wrong

OUT_OF_MEMORY

53200

Out of memory

CONNECTION_FAILURE

08006

GTM errors: Connection failure

CONNECTION_EXCEPTION

08000

Failed to communicate with DNs due to connection errors: Connection exception

ADMIN_SHUTDOWN

57P01

System shutdown by administrators: Admin shutdown

STREAM_REMOTE_CLOSE_SOCKET

XX003

Remote socket disabled: Stream remote close socket

ERRCODE_STREAM_DUPLICATE_QUERY_ID

XX009

Duplicate query id

ERRCODE_STREAM_CONCURRENT_UPDATE

YY016

Stream concurrent update

ERRCODE_LLVM_BAD_ALLOC_ERROR

CG003

Memory allocation error: Allocate error

ERRCODE_LLVM_FATAL_ERROR

CG004

Fatal error

HashJoin temporary file reading error (ERRCODE_HASHJOIN_TEMP_FILE_ERROR).

F0011

File error

Buffer file reading error (ERRCODE_BUFFER_FILE_ERROR)

F0012

File reading error

Partition number error (ERRCODE_PARTITION_NUM_CHANGED).

45003

During scanning on a list partition table, it is found that the number of partitions is different from that in the optimization phase. This problem usually occurs when the queries and ADD/DROP partitions are concurrently executed. (This error is supported only by clusters of version 8.1.3 or later.)

Unmatched schema name (ERRCODE_UNMATCH_OBJECT_SCHEMA)

42P30

Unmatched schema name

Enabling CN Retry

Set the following GUC parameters to enable CN Retry:
  • Mandatory GUC parameters (need to be set on both CNs and DNs)

    max_query_retry_times: specifies the maximum number of automatic retry times when an SQL statement error occurs. Currently, a statement can start retrying if the following errors occur: Connection reset by peer, Lock wait timeout, and Connection timed out. The default value is 6.

    If CN retry is enabled, temporary table data is logged. For data consistency, do not switch the enabled/disabled status for CN retry when the temporary tables are being used by sessions.

  • Optional GUC parameters (contact technical support to modify them)
    1. cn_send_buffer_size: specifies the size of the data buffer used for data transmission on the CN. The default value is 8 KB.
    2. max_cn_temp_file_size: specifies the maximum size of temporary files used on the CN when SQL statements are automatically retried. The default value is 5 GB.
Parent Topic: SQL Tuning