Root Cause and Fix for Galera MySQL 5.7.17 SST Failure Caused by innodb_undo_table_spaces with RSYNC

When adding a node that has already been initialized (‑‑initialize) to a Galera MySQL 5.7.17 cluster, the node completes the State Snapshot Transfer (SST) but then InnoDB rolls back uncommitted transactions using the undo log and triggers an error.

The error indicates that InnoDB accessed a data page outside the undo‑log tablespace, which happens because the SST performed with RSYNC did not correctly synchronize the separate undo‑log tablespace files.

Investigation steps:

1. The suspect was that the SST (performed via RSYNC) failed to copy the undo‑log files.

2. The problematic node’s data directory was deleted, MySQL was restarted, and the node was re‑added. The new data directory contained no undo‑log files, confirming the hypothesis.

Solution:

The RSYNC‑based SST program /usr/bin/wsrep_sst_rsync applies a file‑filter rule before invoking rsync. The original filter included the InnoDB system tablespace (ibdata) but omitted files that start with undo , which store the independent undo‑log tablespaces when innodb_undo_table_spaces is set to a value greater than 0.

By modifying the filter to also include undo* files, the SST correctly copies the undo‑log tablespaces, and the node can be added to the cluster without errors.

The issue has been reported to the Galera MySQL developers and has already been fixed in upstream releases.