Processes involved in reconnecting a node

Note Check the solidDB Release Notes for any limitations that are associated with using a grid in the current release.

The statement ALTER GRID .. SET NODE .. ONLINE reconnects a previously disconnected or unresponsive node to a solidDB grid.

The statement is executed in one of two ways:

▪ manually by an administrator (if the node was manually disconnected), see Reconnecting a node to the grid,

▪ automatically (if the grid determines that a previously unresponsive node is now responding to heartbeat messages).

When the ALTER GRID .. SET NODE .. ONLINE statement is executed, the following steps are performed:

1 The grid leader checks the validity of the disconnected grid node:

▪ The node must still be configured as a grid node, see Processes involved in adding a node.

▪ The grid node must be in MEMBER_OFFLINE or MEMBER_FAILED state and the corresponding information must be in the SYS_GRID_NODES system table.

If the node does not have the correct configuration, the SET NODE .. ONLINE statement fails.

2 The grid leader changes the membership state of the grid node in the grid configuration to MEMBER_RECONNECTING.

3 The grid leader instructs the node to switch state to MEMBER_RECONNECTING. Replication is not enabled at this stage, so a specific instruction is required to deliver the command to the node that is being reconnected.

4 As a consequence of switching state, the reconnecting grid node replicates the required data and starts all the routines that a grid node performs.

5 The grid leader prepares the transaction, writes changes to the transaction log, and waits for acknowledgments that the transaction is prepared on other nodes (including the grid node that is reconnecting).

6 When a majority of grid nodes have prepared the transaction, the grid leader commits the transaction, and writes the commit to the transaction log.

7 All grid nodes commit the transaction and the reconnecting grid node state becomes MEMBER_ONLINE.

Note If the transaction execution fails, the transaction is rolled back and the membership state of the node reverts to the original state (MEMBER_OFFLINE or MEMBER_FAILED).

After the node is reconnected, the Grid Availability Manager (GAM) balances the workload of the grid, see New or reconnected node.