Hacker News new | ask | show | jobs
by fiatjaf 3842 days ago
Why not CouchDB + PouchDB? It's offline first and replicates over HTTP.

For more fine-grained permissions and other features I'm also developing a small Go database compatible with PouchDB at https://github.com/fiatjaf/summadb

I hope it will be able to sync through websockets soon.

4 comments

I find it an ongoing mystery how many of these tools and services pop up that aren't based on Couch/PouchDB, don't reference Couch/PouchDB, aren't compatible with the Couch sync protocol, and don't even seem aware the Couch/PouchDB exist.

I'm 100% waiting to be sold on "here's a new tool that does what CouchDB does, but is better in these specific ways", but if your sales pitch is just "we're working on writing an exact clone of CouchDB because we didn't realise CouchDB existed" it doesn't really sell me.

(Nothing wrong with a dose of Not Invented Here Syndrome; a lot of progress comes from reinventing the wheel. What gets me is the No Idea What's Been Invented Syndrome.)

I am a little confused at why PouchDB isnt listed on the comparisons page, I am fairly certain they sent a comparison listing and asked me to review it at some point but in all fairness they did investigate and give reasons to not use PouchDB, this post is linked http://www.servicedenuages.fr/en/generic-storage-ecosystem

In my extremely biased position (as the maintainer of PouchDB) I would really have liked to still see this based on the CouchDB protocol / PouchDB codebase, it is true that neither Couch or PouchDB have particularly expressive and powerful permission systems (which is what the main downside looks to be) however I have been very aware of that building PouchDB and have made a lot of effort to ensure a powerful and flexible permissions system can be built on top of it (for example https://github.com/thaliproject/node_acl_pouchdb)

I do think that when you are tasked with providing cloud services for existing Mozilla products that there is a lot more risk involved in leveraging an existing established ecosystem like Couch / PouchDB however so its fairly likely that for Mozilla services this was the best choice (the same in reverse is true however also when looking at this as a generic library).

As for a disclaimer, I am a maintainer of PouchDB and a Mozilla developer. I have worked / talked to this team about Kinto, work on some features that use it and they are doing awesome stuff, some of which Pouch may possibly look to do as well.

The comparison with Couch and other solutions actually exists in our documentation: check out http://kinto.readthedocs.org/en/latest/overview.html#compari...
Yeah our doc is incomplete, I did add a bug about this the other day. We should compare products (pouchdb vs kinto.js) and separately deployments (kinto server vs couchdb)
As a Firebase customer and former CouchDB user, I can tell you that the use cases are really not as similar as you apparently think they are.

Until very recently, Couch/Pouch would only sync entire databases, and recommended a "One database per user" approach. This is improving now, but it's still very coarse-grained. Last time I checked, anything that has data that you couldn't shard easily is basically unfit for Couch/Pouch - e.g. any messaging app or a Twitter clone. How would you select which data goes to which user? Or do you just want to put all tweets, globally, on everybody's phone?

All of this is absolutely peanuts on Firebase.

Personally, I'm looking for a database with:

- strong consistency (cross-node transactions!)

- strong access controls (don't want clients to see data they're not allowed to!)

- the ability to seamlessly replicate a view from the server to a web client without me having to think about it (although only read-only)

So far none of the web nosql databases I've looked at support these. Strong consistency seems to be unfashionable, presumably because it's hard to shard. I was really surprised to discover that CouchDB apparently doesn't support access controls, though --- if a client has access to the database endpoint, it can see everything. Aren't access controls part of the core competency of a multi-user database?

The Couchbase Mobile Sync Gateway tackles fine grained access control head on. Basically you write a function that stripes your data into channels, and manages access on a per channel basis.

There are a few things that are challenging to model this way, but for 80% of use cases it's the right simplification. Intro documentation: http://developer.couchbase.com/documentation/mobile/current/...

Sync Gateway looks very plausible... but having to run an extra server just as access control to the main database? That's so ugh.

It also looks like this requires me to annotate each document with ACLs. I was rather hoping to be able to just sync a view, so that changing the database would cause players with changed views to automatically resync. I'm not terribly happy with having to recalculate the views of all players and then update the ACLs of all documents manually on every database mutation; that's a lot of writes.

The database speaks a binary database protocol. The Gateway speaks websocket and http.

Access control is managed via channels, and you configure the gateway with a JS function that determines which channels a document belongs to, so you don't ACL annotations, the ACL is determined dynamically by your code at write time.

There's a tiny bit of write amplification as we persist channel membership in an index, but that's to avoid massive read amplification of irrelevant records when clients sync their subset of channels.

> I was really surprised to discover that CouchDB apparently doesn't support access controls, though --- if a client has access to the database endpoint, it can see everything.

Not sure what you mean; CouchDB has strong access controls, although they are at what in SQL terms we'd call the table level, rather than the row level. But it's not really any different than, eg, MySQL? Or MongoDB, if you want to compare NoSQL to NoSQL.

If you have data in a table that a client shouldn't see, don't give that client access to that table.

You might be interested in https://github.com/couchbase/sync_gateway which is the Couchbase Mobile access control and sync layer (compatible with PouchDB etc). It has fine grained access control based on a channel metaphor. And it's also written in Go, so you might be interested in using parts of it.

Specifically if you wrap your storage layer in a facade that looks like https://github.com/couchbaselabs/walrus you'll be able to use our suite of storage layers, and we'll be able to use yours. That means it'd be trivial to swap out your storage for something scalable like Couchbase Server, or something tiny and embedded like ForestDB or RocksDB.

Thank you, jchris.

As far as I could understand, basically I must implement the Bucket interface[1]? I don't if that will work, since I'm storing things in a somewhat different manner, but it's definitely worth examining.

[1]: https://github.com/couchbase/sg-bucket/blob/master/bucket.go

That's right. If you find you enjoy that code... Couchbase is always looking for great engineers!
Or Couchbase + Sync_Gateway + PouchDB / CouchbaseLite?

This is a very full featured combination, basically a PaaS/BaaS with sync, and user accounts, out of the box.

http://www.couchbase.com/nosql-databases/couchbase-mobile

Note, while that website is very corporate, the software is open source and the company is very supportive of developers.

Speaking of finer-grained permissions in CouchDB/PouchDB. I'm working on implementing that now in my application. Can filtered replication (server-side) solve this problem?

For an application targeting small-businesses, I'm considering the following setup on the server. Has anyone done this or something similar?

One DB per user and a Company DB. Filtered replication from Company DB to User DBs. Full replication from User DBs to Company DB. All reads/writes/deletes from client (or PouchDB sync) to User DB only