We have a 5 node cluster. Whenever we do a bulk transaction, the cluster would crash with “Communication link failure: 1047 WSREP has not yet prepared node for application use”. I understand that this is an issue with clusters, but is there any way around it? The bulk load tables are all ‘temporary’ tables i.e. we load it full of data, clean it up e.g. remove duplicates, and then load it into the normal tables. The cluster always crashes on the DELETES.
We tried to set these ‘temporary’ tables to MyISAM. The problem here is that the query analyzer gets confused and the transactions run really slowly. If I analyze the table before and after the transaction, then it runs in testing. When we add the analyze to the actual code, we get “Communication link failure: 1047 WSREP has not yet prepared node for application use” again, but this time on the ANALYZE.
Is there a way to have a table in the WRITE node that does not get replicated to the other nodes?
Is there a way to tell the cluster not to crash, but to terminate the offending transaction? We do not ‘care’ about the transaction since it can just be rerun in case of failure.
Any other suggestions out there?
We are running on nodes that have 16GB of memory and 4 CPUs.