One more thing, I also noticed that for the co-ordinates in one case were so that none worked. ie.
for uuid:X it gave duplicate key error and for uuid:X+1 it gave key not found error. So, it looks like the writesets weren't atomic either.
a)
Is there a fsync also in galera code (I see one in gcache but don't see where it is called from)?
b) I also see trx_sys_update_wsrep_checkpoint being called from innobase_wsrep_set_checkpoint and trx_sys_read_wsrep_checkpoint but don't see an associated log flush with them. Is there a fsync called along side them somewhere else?
One more thing, I also noticed that for the co-ordinates in one case were so that none worked. ie.
for uuid:X it gave duplicate key error and for uuid:X+1 it gave key not found error. So, it looks like the writesets weren't atomic either.
a)
Is there a fsync also in galera code (I see one in gcache but don't see where it is called from)?
b) I also see trx_sys_ update_ wsrep_checkpoin t being called from innobase_ wsrep_set_ checkpoint and trx_sys_ read_wsrep_ checkpoint but don't see an associated log flush with them. Is there a fsync called along side them somewhere else?