Apologies for the cross-post; we submitted this question to the Codership forum a week ago and haven't received a response. https://groups.google.com/forum/#!topic/codership-team/OSnQ1FLloBI Hello, Hoping to get some perspective on an issue we saw yesterday. While scaling up from a single node cluster, we inserted into a table on node 0 and selected the row count for that table on all subsequent nodes (scripts below). For each node that was brought online, we observed that connections to node 0 were suspended during SST, and that there was an additional IST. At this point the nodes reported that they were synced. However, sometimes replication did not occur correctly - the additional nodes reported fewer rows than node 0. Each new node was added by itself, and all other nodes restarted during each scale-up. Nodes that became out-of-sync continued to have an incorrect row count even after restart, and sometimes got further out of sync (i.e. the discrepancy between the number of rows in node 0 and the other nodes increased) We were conducting this experiment to see whether* innodb_disallow_writes *was functioning correctly, but aren't sure whether this issue is related. Has anyone seen similar behavior while scaling up? Insert: #!/bin/bash mysql -u$db_user -p$db_password -e"create database if not exists $db_name;" mysql -u$db_user -p$db_password -e"use $db_name; create table if not exists $table_name (val int);" for i in `seq $start_val $end_val`; do echo "inserting $i" mysql -u$db_user -p$db_password -e"use $db_name; insert into $table_name VALUES ($i); select count(*) from $table_name;" sleep 0.1 done Select: #!/bin/bash for i in `seq $start_val $end_val`; do echo "statement $i" mysql -u$db_user -p$db_password -e"use replication_test; select count(*) from vals;" sleep 0.1 done Thank you, Shannon Cloud Foundry Services, Pivotal