[dia80]

Genuine question:

In what use cases does mongo kick mysql's ass?

I've used it a couple of times in hobby projects and enjoyed not maintaining a schema. I read so many of these 'gotcha' style articles and for example one commenter here wants to have a manual "recently dirty" flag to combat the master / slave lag mentioned in the article. I know it's faster (tm) but once you have to take in to account all this low level stuff you have to worry about yourself wouldn't it just be better to rent/buy another rack of mysql servers and not worry about it?

Look forward to learning something...

[thibaut_barrere]

MongoDB kicks ass in the following situations (real projects I did as a freelancer):

- dealing with semi-structured input (forms with some variability) and storing as a document, all while being able to query across the data

- used as a store to provide very flexible ETL jobs (with ability to upsert, filter/query, geonear etc)

For those situations, I would definitely use MongoDB again. As a RDBMS replacement, I wouldn't use it today.

[dude_abides]

To slightly rephrase the OP's question:

  In what use cases does mongo kick postgres's ass?

To the two points you mentioned:

- semi-structured input can be saved as hstore type or as json type;

- and for flexible jobs, you can use pretty much any popular language - PL/R, PL/Python, even PL/C if performance is really critical.

[thibaut_barrere]

I would have replied something similar if that was the question :-) (I use PG a lot these days).

Agreed on the first point (but I'm not sure you get exactly the same type of flexibility in all my use cases - I'll have to make a closer comparison).

For the second point, well not having to handle the schema for ETL jobs is sometimes fairly useful and removes a lot of cruft, that was part of my point (those ETL are code-based, only relying on MongoDB as a flexible store).

[d0ugal]

You can't query JSON easily and hstore is only one level deep. So, no, its not as flexible.

[niels]

You can't query the json type as easily in postgres. I would guess that is an important use case.

[jalada]

I don't know about JSON type but the annoying thing about hstore is everything is a string; there are no types.

[taligent]

Just because you can write an app using assembler doesn't mean it is the best way to do it.

[dude_abides]

Very tempted to ask "In what way does postgres compare to an assembler and mongodb to a high level language?" but I think I'll just assume that you're trolling.

Besides, SQL sounds more high-level than map-reduce to me.

[olalonde]

For those like me who have no ides what ETL stands for: https://en.wikipedia.org/wiki/Extract,_transform,_load

[thibaut_barrere]

Thanks for pointing that out!

Here is a presentation (slides + video) I gave about a Ruby ETL, for instance. It illustrates the typical use cases I run into.

http://lanyrd.com/2012/rulu/swxtt/

[jandy]

To elaborate on the semi-structured input point: Monogo and it's kin are great for EAV systems (http://en.wikipedia.org/wiki/Entity%E2%80%93attribute%E2%80%...), where your entities can have an arbitrary number of fields (often user defined). Trying to build this kind of system in a traditional RDBMs can be quite tricky.

[sgolestane]

Since we are a mysql shop, for this use-cases I serialize and store the form as xml in a CLOB column. For any field that needs to be searched on, I create an additional column. Disadvantage of this is you can't run mysql queries against the data stored in the CLOB column.

[thibaut_barrere]

Adding column per searchable field is precisely what I wanted to avoid :-)

If you need to store multiple forms types, it gets hairy very fast.

MongoDB allows to query inside the data (which is not a blob in that case).

[jeltz]

So does PostgreSQL, but I have no idea about MySQL. You can run xpath queries against XML documents, and you can also index specific xpath queries (there is no general indexing infrastructure for XML in PostgreSQL). For JSON no path lookup functions are included in the core yet, but I assume there are extensions which add them.

[thibaut_barrere]

Thanks for elaborating! This is much needed and spot on :-)

A long while back I built a somewhat complex survey app: I can confirm it's fairly more complicated to handle with a RDBMs, compared to a document store.

[joevandyk]

fwiw, postgresql 9.2 supports a json datatype. you can efficiently access/query json fields using plv8 (http://code.google.com/p/plv8js/).

so you could have:

  create table form_results (                                                                                
    id serial primary key,                                                                            
    data json                                                                                         
  );

http://pgeu-plv8.herokuapp.com/ has more information.

[jvehent]

It's just a JSON syntax validator. You can't index on part of the json, postgres treats it as a string.

You can apply full text search on it, but that doesn't tell you if you're matching on a key or a value.

[ismarc]

In postgres, you can index on a stored procedure, so you can use a simple stored procedure written in JavaScript to look up by key and return values, etc. Example available at http://people.planetpostgresql.org/andrew/index.php?/archive...

[vidarh]

Very interesting presentation, though the title of "heralding the death of nosql" is either intentionally exaggerating, or indicates the author doesn't understand all the reasons why people go to nosql databases. In fact, the presentation demonstrates why: Postgres has tons of really fantastic, awesome features that next to nobody uses because they are hidden behind layers of SQL-type incantations and/or require various extensions.

[gbog]

So you'd rather install a completely new storage engine and learn to use it than check a doc and install an extension to postgres?

[thibaut_barrere]

The thing is that 10gen did a really, really good job at polishing the install process and documenting it to get people started.

No surprise to see the GIS part of MongoDB is built-in instead of an extension of some kind. I know a couple of people who used PG without even knowing there was a GIS extension.

[jeltz]

But on the other hand PostGIS being independent from PostgreSQL has resulted in the best opensource GIS database. And with the recent addition of CREATE EXTENSION the PostgreSQL extension installation process has been heavily streamlined. Before CREATE EXTENSION it was a mess for larger extensions.

[ibotty]

is it a good thing to have everything built in?

[amalag]

I don't quite understand how the transition happened from no-json support to json support in those slides. How did the plv8js do this searching on fields?

[alexro]

IMO there is no rational explanation to this phenomena other than: people are different. Some get bored with stored procs and want same hassle but in another form.

[morsch]

A cynical but insightful description of many kinds of progress and change. ;)

[taligent]

Yes. There is no rational explanation for people using the right tool for the right job.

Let me guess. You would build a skyscraper using a trough and cement in a bucket ?

[jamesaguilar]

Are you trying to give us an idea of what it's like to build an even moderate sized app without transactional consistency and with mapreduce? Because that is what it sounds like to me.

[elchief]

When you have so many writes that sharding isn't enough.

When you make changes so fast you need a liquid schema.

When you want to make your boss learn map-reduce so he can query the data.

When the application can take care of integration and not the database itself.

[leif]

This will be obvious, but I work at Tokutek.

> When you have so many writes that sharding isn't enough.

TokuDB does indexed insertions very fast. [1] We've even plugged ourselves in underneath MongoDB (just for fun) and we beat them too. [2]

> When you make changes so fast you need a liquid schema.

TokuDB supports lots of schema changes with zero downtime. [3]

> When you want to make your boss learn map-reduce so he can query the data.

Can't help you there, but I can make you not need to torture your boss that way.

> When the application can care of integration and not the database.

What if it didn't need to?

We also get fantastic compression [4], retain full transactional semantics, and lots of other fun stuff.

Email us if you're curious!

[1]: http://www.tokutek.com/resources/benchmark-results/benchmark...

[2]: http://www.tokutek.com/2012/08/10x-insertion-performance-inc...

[3]: http://www.tokutek.com/resources/benchmark-results/benchmark...

[4]: http://www.tokutek.com/resources/benchmark-results/benchmark...

[DanWaterworth]

In my experience, it's not just throughput that's important, but also 99th percentile latency. If I understand fractal trees correctly, you sometimes need to rewrite all of your elements on disk. How do you do this without causing lag?

[leif]

That's happily not how fractal trees work, at all. We have a few talks online describing how they work. Zardosht has one here http://vimeo.com/m/26471692 . I thought Bradley had a more detailed one at MIT but I can't find it right now. (EDIT: found it! http://video.mit.edu/watch/lecture-19-how-tokudb-fractal-tre...)

Basically, I think you're thinking of the COLA. What we implement does have a literal tree structure, with nodes and children and the whole thing, so at any point you're just writing out new copies of individual nodes, which are on the order of a megabyte. At no point do we have to rewrite a large portion of the tree, so there aren't any latency issues.

[DanWaterworth]

Thank you very much for the links. Is my understanding correct?

Essentially, a fractal tree is a B-Tree (or perhaps a B+Tree?) with buffers on each branch (per child). Operations get added to these buffers and when one becomes full, the operations get passed to the corresponding child node. Operations are applied when they reach the node that is responsible for the data concerned.

[leif]

That's about it! It's like a B+Tree in that the data resides in the leaves (well, and in the buffers), except that the fanout is a lot lower because we save room in the internal nodes for the buffers.

[stinkypete]

Sounds like a Y-Tree to me: "A Novel Index Supporting High Volume Data Warehouse Insertions," C. Jermaine et. al, VLDB '99.

How are fractal trees different?

[leif]

Is that the same as Willard's Y-fast tree? If so, then no they're not very similar. If not then I'm unfamiliar and I should read that paper.

[lwat]

> When you have so many writes that sharding isn't enough.

Does Mongo still have a global write lock?

[elchief]

no longer (but only a few months)

[mgummelt]

Though I believe it still has a per-collection lock.

[Jare]

It has per-database locking. Per-collection is being worked on but no planned date yet. [1]

[1] http://www.mongodb.org/display/DOCS/How+does+concurrency+wor...

[diminoten]

Mongo isn't so important to this question as ODB vs RDBMS. Here's some light reading:

http://en.wikipedia.org/wiki/Object-relational_impedance_mis...

MongoDB is just ODB, and MySQL is just RDB.

Besides, postgres is the real future!

[dleblanc]

Not sure what Wu-Tang has to do with this, but....

Seriously though, is Mongo an ODB, or a document oriented database? ODBs/OODBs imply a much different use-case and functionality, and I think we ought not conflate the two.

[ww520]

Mongo is not an ODB in the traditional ODB sense. Mongo is just a document (or blob) database. ODB supports relationship, link traversal, inheritance, etc. Basically the whole shebang of the OO notions in a database.

[diminoten]

You're right, I didn't know there was a difference.

[taligent]

PostgreSQL is the PAST and it will forever remain there until it solves its biggest problems.

It is the hardest out of all databases I've used to cluster, replicate, shard and manage. And the database world is moving towards scaling horizontally rather than vertically. I can push a button in say CouchDB to replicate and shard. Try doing that in PostgreSQL.

[jeffdavis]

"And the database world is moving towards scaling horizontally rather than vertically."

I think that's an oversimplification. Two counterpoints:

* Database systems will always need to make heavy use of locality. The speed of light means that synchronizing access over long distances (even medium distances -- light only goes about a foot per nanosecond) will always be a challenge.

* Multi-core means vertical scaling is back in (if by "vertical scaling" you mean "scaling on one box"), and probably for a while.

Postgres is doing an excellent job at both.

I do agree with the less-exaggerated point that postgres really needs to improve its multi-machine scaling. But it's far from a solved problem on any system under discussion.

[andyzweb]

postgres does horrible with multicore unless you have many concurrent pooled connections. fork-and-forget

[jeffdavis]

I think you're referring to postgres's inability to use more than one core per query, which is true (or mostly true... there are quite a few helper processes that take on some of the work).

For many smaller queries, postgres does great on multi-core, and pgbouncer is a good connection pooler.

[true_religion]

People still use SQLlite. Why? Because its suits a specific set of circumstances.

Postgres does the same---its an incredibly powerful database with a large suit of features that make application development both easy and sane. To the vast majority of projects where you won't outgrow a single server, it makes sense to use it.

[np422]

You have obviously never used Oracle RAC .. :)

[jeremyjh]

One issue with MySQL in large databases is that schema changes are extremely expensive, so much so that you'll be making design decisions around it (e.g. how do we implement this feature without executing our two-day alter table statement). Not all RDBMS have this problem to such a degree but none can escape it entirely.

A lot of the gotchas that he notes are related to design trade-offs with different default behavior than an RDBMS typically would have. For example as your system gets large enough in MySQL you may find you have to do asynchronous replication as well, and then you will have similar problems with dirty reads.

[bsg75]

> e.g. how do we implement this feature without executing our two-day alter table statement?

With MySQL being so prevalent, IMO this is the principal reason why the concept that RDBMS' are hard to work with exists. None of the other major platforms have this issue, but if a developer's only exposure is to MySQL, then the idea of schema alters are always fearsome.

This is not to say that ALTER TABLE statements are always quick - if constraint checks or default values are included, there can be long runtimes in PostgreSQL, MSSQL, Oracle, etc. - but significant downtime to add an empty nullable column is just plain stupid for a platform that has been around as long as MySQL.

MySQL has forced a major population segment of developers to toss the advantages of DB side validation and rich query languages for the purpose of _avoiding_MySQL_. Sure the whole object-relational impedance mismatch exists, SQL is hard, yada-yada, but I have never seen these reasons cause as much angst as taking a service offline to add a column to a table.

10gen should have a picture of Monty in their CFO's office.

[mrkurt]

Eventually consistant replication is not unique to MongoDB, most DBs have an async replication option. Using "not Mongo" won't really solve it.

[krichman]

An option is not a default, though, so using Postgres or MySQL absolutely would solve the problem of returning with success before persisting the data.

The contentious area isn't really that Mongo does this, it's because for whatever reason the people trying it don't expect Mongo to do this.

[untog]

I think there are legitimate reasons to use a "NoSQL" solution rather than MySQL. I'm more interested to know in what use cases Mongo kicks it's competitors asses? What are it's competitors, even? I'll admit that the NoSQL world is a slightly blurry mess to me, with different products seemingly optimised for different cases.

[taligent]

MongoDB is a JSON document store. Almost all other NoSQL are key/value type stores.

If you have a use case that centres around storing document style data then MongoDB will be better suited.

[trustfundbaby]

not to be persnickety but its a BSON document store. Theres a difference.

[mattparlane]

My number one reason for choosing MongoDB is replication that just works out of the box and doesn't require either a read lock or shutting down the master to set up a new slave.

[sausagefeet]

As far as I can tell, Riak blows mongo away for this particular criteria.

[mattparlane]

That's my understanding too, but MongoDB hit the sweet-spot for us, Riak couldn't handle the raw queries-per-second we needed without requiring extra hardware ($$).

[jaimebuelta]

But Riak is much limited on the queries you can run.

[jeltz]

Serious question: What database requires read locks or shutting down the master for setting up a new slave? Is it MySQL?

Both sounds like really weird requirements for replication which defeats half the purpose of having it. PostgreSQL has never had any of these two problems. Still working on improving usability but with the addition and improvements of pg_basbackup I would say it is almost there. Hopefully 9.3 will get timeline switching to simplify failover.

[mrinterweb]

Map reduce across sharded servers comes to mind as an advantage. For that matter, horizontal scalability in general is a big advantage that many of the NoSQL data stores have over RDBMS databases.