[robconery]

OP here - many NoSQL/document DBs will trade off write acks for eventual consistency. I really liked their approach to pushing toward durability by default - that in particular was the thing that impressed me, which I should have been more clear about.

[picardo]

Does it write to a disk or to a log? If it writes to disk, it may still not be completely fail-safe unless it also writes to a log before writing to disk. Postgres for example has a write ahead log and it writes to disk before acknowledging the write.

[coffeemug]

RethinkDB has a log-structured storage engine. There is no separate log like in Postgres, the log is implicitly integrated into the storage engine. You don't have to write data twice (like you would with a traditional log), but you're still guaranteed safety in case of power failure. The design is roughly based on this paper: http://www.cs.berkeley.edu/~brewer/cs262/LFS.pdf.

[bmurphy1976]

They use MVCC: http://rethinkdb.com/docs/architecture/

[eloff]

So does postgres. That alone isn't enough because you can get situations like torn pages where part of the database page is old data and part is new data and nothing makes sense anymore. A log fixes that by first writing to the log so that if the database page gets messed up you have a secondary source you can use to restore it from.

[trhway]

>Does it write to a disk or to a log? If it writes to disk, it may still not be completely fail-safe unless it also writes to a log before writing to disk.

why we still discussing it at a tech forum in 21st century in Silicon Value? Shouldn't it (together with isolation, ACID, CAP, etc...) be a base knowledge taught in elementary school? Like you can't expect Daddy to buy you a firetruck that Mommy promised if Mommy hasn't been able to talk to Daddy yet... though until Mommy talks to Daddy you probably can convince Daddy to buy you a railroad...

[MCRed]

Yeah, that should be table stakes.