Return of the Mongo Mailbag

On the mongodb-user mailing list last week, someone asked (basically):

I have 4 servers and I want two shards. How do I set it up?

A lot of people have been asking questions about configuring replica sets and sharding, so here’s how to do it in nitty-gritty detail.

The Architecture

Prerequisites: if you aren’t too familiar with replica sets, see my blog post on them. The rest of this post won’t make much sense unless you know what an arbiter is. Also, you should know the basics of sharding.

Each shard should be a replica set, so we’ll need two replica sets (we’ll call them foo and bar). We want our cluster to be okay if one machine goes down or gets separated from the herd (network partition), so we’ll spread out each set among the available machines. Replica sets are color-coded and machines are imaginatively named server1-4.

Each replica set has two hosts and an arbiter. This way, if a server goes down, no functionality is lost (and there won’t be two masters on a single server).

To set this up, run:

server1

$ mkdir -p ~/dbs/foo ~/dbs/bar
$ ./mongod --dbpath ~/dbs/foo --replSet foo
$ ./mongod --dbpath ~/dbs/bar --port 27019 --replSet bar --oplogSize 1

server2

$ mkdir -p ~/dbs/foo
$ ./mongod --dbpath ~/dbs/foo --replSet foo

server3

$ mkdir -p ~/dbs/foo ~/dbs/bar
$ ./mongod --dbpath ~/dbs/foo --port 27019 --replSet foo --oplogSize 1
$ ./mongod --dbpath ~/dbs/bar --replSet bar

server4

$ mkdir -p ~/dbs/bar
$ ./mongod --dbpath ~/dbs/bar --replSet bar

Note that arbiters have an oplog size of 1. By default, oplog size is ~5% of your hard disk, but arbiters don’t need to hold any data so that’s a huge waste of space.

Putting together the replica sets

Now, we’ll start up our two replica sets. Start the mongo shell and type:

> db = connect("server1:27017/admin")
connecting to: server1:27017
admin
> rs.initiate({"_id" : "foo", "members" : [
... {"_id" : 0, "host" : "server1:27017"},
... {"_id" : 1, "host" : "server2:27017"},
... {"_id" : 2, "host" : "server3:27019", arbiterOnly : true}]})
{
        "info" : "Config now saved locally.  Should come online in about a minute.",
        "ok" : 1
}
> db = connect("server3:27017/admin")
connecting to: server3:27017
admin
> rs.initiate({"_id" : "bar", "members" : [
... {"_id" : 0, "host" : "server3:27017"},
... {"_id" : 1, "host" : "server4:27017"},
... {"_id" : 2, "host" : "server1:27019", arbiterOnly : true}]})
{
        "info" : "Config now saved locally.  Should come online in about a minute.",
        "ok" : 1
}

Okay, now we have two replica set running. Let’s create a cluster.

Setting up Sharding

Since we’re trying to set up a system with no single points of failure, we’ll use three configuration servers. We can have as many mongos processes as we want (one on each appserver is recommended), but we’ll start with one.

server2

$ mkdir ~/dbs/config
$ ./mongod --dbpath ~/dbs/config --port 20000

server3

$ mkdir ~/dbs/config
$ ./mongod --dbpath ~/dbs/config --port 20000

server4

$ mkdir ~/dbs/config
$ ./mongod --dbpath ~/dbs/config --port 20000
$ ./mongos --configdb server2:20000,server3:20000,server4:20000 --port 30000

Now we’ll add our replica sets to the cluster. Connect to the mongos and and run the addshard command:

> mongos = connect("server4:30000/admin")
connecting to: server4:30000
admin
> mongos.runCommand({addshard : "foo/server1:27017,server2:27017"})
{ "shardAdded" : "foo", "ok" : 1 }
> mongos.runCommand({addshard : "bar/server3:27017,server4:27017"})
{ "shardAdded" : "bar", "ok" : 1 }

Edit: you must list all of the non-arbiter hosts in the set for now. This is very lame, because given one host, mongos really should be able to figure out everyone in the set, but for now you have to list them.

Tada! As you can see, you end up with one “foo” shard and one “bar” shard. (I actually added that functionality on Friday, so you’ll have to download a nightly to get the nice names. If you’re using an older version, your shards will have the thrilling names “shard0000” and “shard0001”.)

Now you can connect to “server4:30000” in your application and use it just like a “normal” mongod. If you want to add more mongos processes, just start them up with the same configdb parameter used above.

Mongo Mailbag #2: Updating GridFS Files

Welcome to week two of Mongo Mailbag, where I take a question from the Mongo mailing list and answer it in more detail. If you have a question you’d like to see answered in excruciating detail, feel free to email it to me.

Is it possible (with the PHP driver) to storeBytes into GridFS (for example CSS data), and later change that data?!

I get some strange behavior when passing an existing _id value in the $extra array of MongoGridFS::storeBytes, sometimes Apache (under Windows) crashes when reloading the file, sometimes it doesn’t seem to be updated at all.

So I wonder, is it even possible to update files in GridFS?! ūüôā

-Wouter

If you already understand GridFS, feel free to skip to the last section. For everyone else…

Intro to GridFS

GridFS is the standard way MongoDB drivers handle files; a protocol that allows you to save an arbitrarily large file to the database. It’s not the only way, it’s not the best way (necessarily), it’s just the built-in way that all of the drivers support. This means that you can use GridFS to save a file in Ruby and then retrieve it using Perl and visa versa.

Why would you want to store files in the database? Well, it can be handy for a number of reasons:

  • If you set up replication, you’ll have automatic backups of your files.
  • You can keep millions of files in one (logical) directory… something most filesystems either won’t allow or aren’t good at.
  • You can keep information associated with the file (who’s edited it, download count, description, etc.) right with the file itself.
  • You can easily access info from random sections of large files, another thing traditional file tools aren’t good at.

There are some limitations, too:

  • You can’t have an arbitrary number of files per document… it’s one file, one document.
  • You must use a specific naming scheme for the collections involved: prefix.files and prefix.chunks (by default prefix is “fs”: fs.files and fs.chunks).

If you have complex requirements for your files (e.g., YouTube), you’d probably want to come up with your own protocol for file storage. However, for most applications, GridFS is a good solution.

How it Works

GridFS breaks large files into manageable chunks. It saves the chunks to one collection (fs.chunks) and then metadata about the file to another collection (fs.files). When you query for the file, GridFS queries the chunks collection and returns the file one piece at a time.

Here are some common questions about GridFS:

Q: Why not just save the whole file in a single document?
A: MongoDB has a 4MB cap on document size.
Q: That’s inconvenient, why?
A: It’s an arbitrary limit, mostly to prevent bad schema design.
Q: But in this case it would be so handy!
A: Not really. Imagine you’re storing a 20GB file. Do you really want to return the whole thing at once? That means 20GB or memory will be used whenever you query for that document. Do you even have that much memory? Do you want it taken up by a single request?
Q: Well, no.
A: The nice thing about GridFS is that it streams the data back to the client, so you never need more than 4MB of memory.
Q: Now I know.
A: And knowing is half the battle.
Together: G.I. Joe!

Answer the Damn Question

Back to Wouter’s question: changing the metadata is easy: if we wanted to add, say, a “permissions” field, we could run the following PHP code:

$files = $db->fs->files;
$files->update(array("filename" => "installer.bin"), array('$set' => array("permissions" => "555")));

// or, equivalently, from the MongoGridFS object:

$grid->update(array("filename" => "installer.bin"), array('$set' => array("permissions" => "555")));

Updating the file itself, what Wouter is actually asking about, is significantly more complex. If we want to update the binary data, we’ll need to reach into the chunks collection and update every document associated with the file. Edit: Unless you’re using the C# driver! See Sam Corder’s comment below. It would look something like:

// get the target file's chunks
$chunks = $db->fs->chunks;
$cursor = $chunks->find(array("file_id" => $fileId))->sort(array("n" => 1));

$newLength = 0;

foreach ($cursor as $chunk) {
    // read in a string of bytes from the new version of the file
    $bindata = fread($file, MongoGridFS::$chunkSize);
    $newLength += strlen($bindata);

    // put the new version's contents in this chunk
    $chunk->data = new MongoBinData($bindata);

    // update the chunks collection with this new chunk
    $chunks->save($chunk);
}

// update the file length metadata (necessary for retrieving the file)
$db->fs->files->update(array("_id" => $fileId), array('$set' => array("length" => $newLength));

The code above doesn’t handle a bunch of cases (what if the new file is a different number of chunks than the old one?) and anything beyond this basic scenario gets irritatingly complex. If you’re updating individual chunks you should probably just remove the GridFS file and save it again. It’ll end up taking about the same amount of time and be less error-prone.

Mongo Mailbag: Master/Slave Configuration

Trying something new: each week, I’ll take an interesting question from the MongoDB mailing list and answer it in more depth. ¬†Some of the replies on the list are a bit short, given that the developers are trying to, you know, develop (as well as answer over a thousand questions a month). ¬†So, I’m going to grab some interesting ones and flesh things out a bit more.

Hi all,

Assume I have a Mongo master and 2 mongo slaves.  Using PHP, how do I do it so that writes goes to the master while reads are spread across the slaves (+maybe the master)?

1) 1 connect to all 3 nodes in one go, PHP/Mongo handles all the rest
2) 1 connect to the master for writes. Another connection to connect to all slave nodes and read from them.

Thanks all and sorry for the noobiness!

-Mr. Google

Basics first: what is master/slave?

One database server (the “master”) is in charge and can do anything. ¬†A bunch of other database servers keep copies of all the data that’s been written to the master and can optionally be queried (these are the “slaves”). ¬†Slaves cannot be written to directly, they are just copies of the master database. ¬†Setting up a master and slaves allows you to scale reads nicely because you can just keep adding slaves to increase your read capacity. ¬†Slaves also make great backup machines. If your master explodes, you’ll have a copy of your data safe and sound on the slave.

A handy-dandy comparison chart between master database servers and slave database servers:

Master Slave
# of servers 1 ‚ąě
permissions read/write read
used for queries, inserts, updates, removes queries

So, how do you set up Mongo in a master/slave configuration? ¬†Assuming you’ve downloaded MongoDB from mongodb.org,¬†you can start a master and slave by cutting and pasting the following lines into your shell:

$ mkdir -p ~/dbs/master ~/dbs/slave
$ ./mongod --master --dbpath ~/dbs/master >> ~/dbs/master.log &
$ ./mongod --slave --port 27018 --dbpath ~/dbs/slave --source localhost:27017 >> ~/dbs/slave.log &

(I’m assuming you’re running *NIX. ¬†The commands for Windows are similar, but I don’t want to encourage that sort of thing).

What are these lines doing?

  1. First, we’re making directories to keep the database in (~/dbs/master and ~/dbs/slave).
  2. Now we start the master, specifying that it should put its files in the ~/dbs/master directory and its log in the ~/dbs/master.log file.  So, now we have a master running on localhost:27017.
  3. Next, we start the slave. It needs to listen on a different port than the master since they’re on the same machine, so we’ll choose 27018. It will store its files in ~/db/slave and its logs in ~/dbs/slave.log. ¬†The most important part is letting it know who’s boss: the –source localhost:27017 option lets it know that the master it should be reading from is at localhost:27017.

There are tons of possible master/slave configurations. Some examples:

  • You could have a dozen slave boxes where you want to distribute the reads evenly across them all.
  • You might have one wimpy little slave machine that you don’t want any reads to go to, you just use it for backup.
  • You might have the most powerful server in the world as your master machine and you want it to handle both reads and writes… unless you’re getting more than 1,000 requests per second, in which case you want some of them to spill over to your slaves.

In short, Mongo can’t automatically configure your application to take advantage of your master-slave setup. Sorry. ¬†You’ll have to do this yourself. (Edit: the Python driver actually does handle case 1 for you, see Mike’s comment.)

However, it’s not too complicated, especially for what MG wants to do. ¬†MG is using 3 servers: a master and two slaves, so we need three connections: one to the master and one to each slave. ¬†Assuming he’s got the master at master.example.com and the slaves at slave1.example.com and slave2.example.com, he can create the connections with:

$master = new Mongo("master.example.com:27017");
$slave1 = new Mongo("slave1.example.com:27017");
$slave2 = new Mongo("slave2.example.com:27017");

This next bit is a little nasty and it would be cool if someone made a framework to do it (hint hint). ¬†What we want to do is abstract the master-slave logic into a separate layer, so the application talks to the master slave logic which talks to the driver. ¬†I’m lazy, though, so I’ll just extend the MongoCollection class and add some master-slave logic. ¬†Then, if a person creates a MongoMSCollection from their $master connection, they can add their slaves and use the collection as though it were a normal MongoCollection. ¬†Meanwhile, MongoMSCollection will evenly distribute reads amongst the slaves.

class MongoMSCollection extends MongoCollection {
    public $currentSlave = -1;

    // call this once to initialize the slaves
    public function addSlaves($slaves) {
        // extract the namespace for this collection: db name and collection name
        $db = $this->db->__toString();
        $c = $this->getName();

        // create an array of MongoCollections from the slave connections
        $this->slaves = array();
        foreach ($slaves as $slave) {
            $this->slaves[] = $slave->$db->$c;
        }

        $this->numSlaves = count($this->slaves);
    }

    public function find($query, $fields) {
        // get the next slave in the array
        $this->currentSlave = ($this->currentSlave+1) % $this->numSlaves;

        // use a slave connection to do the query
        return $this->slaves[$this->currentSlave]->find();
    }
}

To use this class, we instantiate it with the master database and then add an array of slaves to it:

$master = new Mongo("master.example.com:27017");
$slaves = array(new Mongo("slave1.example.com:27017"), new Mongo("slave2.example.com:27017"));

$c = new MongoMSCollection($master->foo, "bar");
$c->addSlaves($slaves);

Now we can use $c like a normal MongoCollection.  MongoMSCollection::find will alternate between the two slaves and all of the other operations (inserts, updates, and removes) will be done on the master.  If MG wants to have the master handle reads, too, he can just add it to the $slaves array (which might be better named the $reader array, now):

$slaves = array($master, new Mongo("slave1.example.com:27017"), new Mongo("slave2.example.com:27017"));

Alternatively, he could change the logic in the MongoMSCollection::find method.

Edit: as of version 1.4.0, slaveOkay is not neccessary for reading from slaves. slaveOkay should be used if you are using replica sets, not –master and –slave. Thus, the next section doesn’t really apply anymore to normal master/slave.

The only tricky thing about Mongo’s implementation of master/slave is that, by default, a slave isn’t even readable, it’s just a way of doing backup for the master database. ¬†If you actually want to read off of a slave, you have to set a flag on your query, called “slaveOkay”. ¬†Instead of saying:

$cursor = $slave->foo->bar->find();

we have:

$cursor = $slave->foo->bar->find()->slaveOkay();

Or, because this is a pain in the ass to set for every query (and almost impossible to do for findOnes unless you know the internals) you can set a static variable on MongoCursor that will hold for all of your queries:

MongoCursor::$slaveOkay = true;

And now you will be allowed to query your slave normally, without calling slaveOkay() on each cursor.

References: