Hacker News Reader

> I am still a bit confused how this locking system works for read/write and also a bit skeptical regarding their claim that 'in most cases writes can be performed essentially lock-free'.

Hi, Slava at RethinkDB here.

RethinkDB uses MVCC to do looking. Essentially, when we lock down a block for a write, we make a copy of the block. If another query comes along that wants to read, it reads from the copy. When the write completes, the old copies are destroyed.

There are lots of details I'm glossing over -- optimizations to avoid copying too much, copying entire subbranches of the btree to have a consistent view of the shard, etc. All this stuff isn't unique to RethinkDB -- it's pretty standard database internals stuff, and we haven't done anything new in that department. It's just an implementation of standard database architectures (as far as the caching/query engine/storage engine are concerned).

FYI, with MongoDB, just because you can't and shouldn't index everything, doesn't mean you can't have any indexes... if your most common fields bring your queries down, they're still pretty helpful.

I actually really like where RethinkDB is headed, and within the year most of my issues should be resolved.

Another couple databases to consider, depending on your needs would be ElasticSearch and Cassandra... it reallly depends on your use case.

Could you list your issues with Rethink? It would really help for product prioritization.

As mentioned in another thread, automagic failover the one still pending, and geospatial indexes/searches (now in the product).

It's wild how many options there are that tailor themselves to all kinds of data out there.

I'm not involved with RethinkDB but I lurk on their github issues and I'm pretty confident that automagic failover is dependent on them getting (their own implementation of) Raft integrated with everything. Looks like its getting close as a whole slew of issues were opened just the other day relating to Raft work.

Are unique indexes planned for any of the upcoming releases?

The primary index for the table is always unique.

We do not plan to implement secondary unique indexes. The philosophy behind RethinkDB is that if a feature cannot be efficiently scaled across multiple nodes we don't add it, and unfortunately unique secondary indexes are one such feature.

AFAIK most NoSQL databases don't implement it, and a few that do take two approaches -- forbid it on multiple shards, or just take a massive performance hit during sharding.

We chose to keep the feature out of the database. This way the application can be architected to account for it, so it remains fast as the database scales up.

This is good, for what it's worth. Having different feature sets for sharded vs unsharded is just utterly confusing, and something MongoDB got really wrong.

Oh yes, agreed! and we do have some indexes, but unfortunately it isn't enough :(

And we are actually considering ElasticSearch to deal with this querying performance issue.

We had a similar issue with queries that didn't match an index... later pages would time out, etc... we limited our results for that class of queries... The next generation will use cassandra with some custom searching/caching... that will work a bit differently.

No database can be fast without indexes. If the queries you get are defined by your users, you can still create indexes for the common ones.

The way business logic in the system is currently designed , there are no 'common ones', except for some ids (already indexed) used in other process rather than this dynamic filter.

Based on that we also evaluated the approach of using something else like druid [1] [2] that is built for reading performance, but I am still studying possibilities and have no idea about the impact and problems a change like that would impose.

[1] http://druid.io/

[2] https://metamarkets.com/2014/building-a-data-pipeline-that-h...

Another thing to note is that Mongo is simply very slow. Without indexes, something like Postgres is much more likely to be fast.