Restart light

When a failed member cannot be restarted on its original host, or home host, Db2® cluster services restarts that member on one of the other available hosts in the Db2 pureScale® instance. This process, known as restart light, allows member crash recovery to be performed without significantly affecting the rest of the instance.

A member that has been restarted in restart light mode on another host is known as a guest member, whereas a member that is running on its home host is known as a resident member. A guest member uses fewer resources than a resident member, and it cannot accept instance attachments or a database connections from an external application (SQL1032N).

The sole purpose of starting a member in restart light mode is to perform member crash recovery. The guest member is restarted using a pre-allocated, reduced memory model to minimize the affect on the resident member whose host is used for the restart light. After the guest member running in restart light mode completes member crash recovery on all required databases, it waits to be failed back to its home host. It cannot process new transactions until it is failed back to its home host.

Note: A member in restart light mode appears to be up and running (but in WAITING_FOR_FAILBACK state) if its state is queried. When the failed home host comes back online, the member in restart light mode will automatically be failed back to its own home host by Db2 cluster services and will become a fully active member that can process new transactions again.

Restart light is an automated process that works as follows: For software failures where the home host of the member is still active, Db2 cluster services attempts to restart a failed member on its home host. If the first restart attempt of the failed member does not succeed on its home host, that member is restarted in light mode on a different host as a guest member.

Restart light is immediately initiated for a member if one of the following situations occurs:
  • a network failure where the home host loses its connection with the rest of the Db2 pureScale instance
  • the home host becomes inactive unexpectedly because of a power failure or a hardware, LPAR, VM, or OS reset
  • the member terminates abnormally when its home host is being stopped for maintenance

A restart light is a very rapid process because there is a set of Db2 idle processes on each host that preallocates resources for the restarting of guest members. Db2 will activate these processes instead of creating new processes to perform member restart in light mode. As new processes are not created during restart light and because the guest member does not compete with the resident member for resources, member recovery processing is sped up.