• in sql dbs (sqlite), you write to primary server and replicated to secondary guys
• read requests distributed to secondary replicas, write goes to master
• read more common than write
• in Logos, writing goes to sequencers. sequencer will write binlog to Bedrock.
• replicators “indexer nodes” will replicate state locally by applying logs posted to Bedrock, people can send read reqs to them
• all validation must be done within sql pessimistic
  • baby step: trust the sequencer
• use python with sqlite
• extract transactions from callback
• wait for not full finality but replicate with some delay
  • or maybe deal with forks? create db copy for each new block & apply changes, then prune any blocks not included after finality
  • we can rely on channel immediate ordering
• for demo, use open source tools that use sqlite & plug it into logos