transactions - how do non-ACID RethinkDB or MongoDB maintain secondary indexes for non-equal queries -

this more of 'inner workings' undestanding question:

how nosql databases not support *a*cid (meaning cannot update/insert , rollback data more 1 object in single transaction) -- update secondary indexes ?

my understanding -- in order keep secondary index in sync (other wise become stale reads) -- has happen withing same transaction.

furthermore, if possible index reside on different host data -- distributed lock needs present and/or two-phase commit such update work atomically.

but if these databases not support multi-object transactions (which means not two-phase commit on data across multiple host) , method use guarantee secondary indices reside in b-trees structures separate data not stale ?

this great question.

rethinkdb stores secondary indexes on same host primary index/data table. in case of joins, rethinkdb brings query data, secondary indexes, primary indexes, , data reside on same node. result, there no need distributed locking protocols such 2 phase commit.

rethinkdb support limited set of transactional functionality -- single document transactions. changes single document recorded atomically. relevant secondary index changes recorded part of transaction, either entire change recorded, or nothing recorded @ all.

it easy extend limited transactional functionality support multiple documents in single shard, hard across shards (for distributed locking reasons brought up), decided not implement transactions multiple documents yet.

hope helps.