User guide

Review the MTU value for the Ethernet adapter in the output. If the value is not 1500, run the following
command to set it:
netsh interface ipv4 set subinterface "Ethernet" mtu=1500 store=persistent
After you set this value, restart your computer for the changes to take effect.
Set MTU on a Linux Operating System
If your client runs in a Linux operating system, you can review and set the MTU value by using the ip
command. Run the following command to determine the current MTU value:
$ ip link show eth0
Review the value following mtu in the output. If the value is not 1500, run the following command to set
it:
$ ip link set dev eth0 mtu 1500
Set MTU on Mac OS X
To set the MTU on a Mac operating system, follow the instructions in Mac OS X 10.4 or later: How to
change the MTU for troubleshooting purposes.
For a similar issue with a different solution, see Queries Appear to Hang in the Client and the
Cluster (p. 177).
Queries Appear to Hang in the Client and the Cluster
Example issue:
Users experience issues with their queries completing.The queries run on the user's SQL client tool and
reach the database, but they don't complete and sometimes can't be terminated.
Possible solution:
Make sure that users configure their SQL client tools to automatically commit SQL commands, and that
they use a transaction block (defined by BEGIN and END statements) to enclose a set of commands.
This configuration ensures that each block commits as one transaction that you can roll back if necessary.
The example issue can occur when a user submits a transaction without committing the block as one
transaction, and an error occurs during processing. If the client tool attempts to roll back to a previous
point inside the transaction (called a savepoint), the attempt fails because Amazon Redshift does not
support rolling back to a savepoint. As a result, the transaction cannot proceed. If the
PG_TERMINATE_BACKEND function fails to terminate the connection from which the transaction was
submitted, the locks on the relevant user tables cannot be released. In some cases, this failure can cause
the cluster to become unavailable and require a reboot.
When this issue happens, users see error such as ERROR: SQL command "ROLLBACK TO
savepoint_name;" not supported and ERROR: current transaction is aborted, commands
ignored until end of transaction block.
Many client tools, such as SQL Workbench/J, enable the use of savepoints by default. Because savepoints
are incompatible with Amazon Redshift, users must configure their client tools to bypass this behavior.
You can run the following query to determine which users are using SQL client tools that create savepoints.
API Version 2012-12-01
177
Amazon Redshift Management Guide
Troubleshooting Connection Issues in Amazon Redshift