Querying

Distributing queries to secondaries

All queries (reads and writes) are only sent to the primary member of a replica set by default. This is however easily configurable by using the Read Preferences which allow you to set some generic read preferences (such as allowing secondary reads of the nearest server), and also provide ways to specifically target a server in a specific country, datacenter, or even hardware, by the use of replica set tag sets.

Read preferences can be configured at every level of the driver:

Each class inherits its read preference setting from the "parent" context.

Example #1 Inheriting read preferences from the database level down to the cursor

<?php
$db
->setReadPreference(MongoClient::RP_SECONDARY_PREFERRED);
$c $db->myCollection;

$cursor $c->find();
?>

In this example, the the query will be executed against a secondary. The collection inherits MongoClient::RP_SECONDARY_PREFERRED from the database and the cursor inherits it from the collection.

How secondaries are chosen

Each instance of MongoClient chooses its own secondary using the available secondary with the lowest ping time. So, if we had a PHP client in Europe and one in Australia and we had one secondary in each of these data centers, we could do:

<?php
$options 
= array("replicaSet" => "setName""readPreference" => MongoClient::RP_SECONDARY_PREFERRED);

// on the Australian client
$m = new MongoClient("mongodb://primary,australianhost.secondary,europeanhost.secondary"$options);
$cursor $m->foo->bar->find();
$cursor->getNext();
echo 
"Reading from: "$cursor->info()["server"], "\n";

// on the European client
$m = new MongoClient("mongodb://primary,australianhost.secondary,europeanhost.secondary"$options);
$cursor $m->foo->bar->find();
$cursor->getNext();
echo 
"Reading from: "$cursor->info()["server"], "\n";
?>

The above example will output something similar to:

Reading from: australianHost
Reading from: europeanHost

Note that we have to do a query before a secondary is chosen: secondaries are chosen lazily by the driver, and for each query separately.

You can see what the driver thinks is the current status of the set members by running MongoClient::getHosts() or MongoClient::getConnections().

If no secondary is readable, the driver will send reads to the primary as we specified MongoClient::RP_SECONDARY_PREFERRED which will fallback to execute a query on a primary if no secondaries are available. A server is considered readable if its state is 2 (SECONDARY) and its health is 1. You can check this with MongoClient::getHosts() and MongoClient::getConnections().

Random notes

Writes are always sent to the primary—and by default all reads are sent to the primary too.

Querying by _id

Every object inserted is automatically assigned a unique _id field, which is often a useful field to use in queries. This works similarly to "get last insert ID" functionality, except that the _id is chosen by the client.

Suppose that we wish to find the document we just inserted. Inserting adds an _id field to the document, so we can query by that:

<?php
$person 
= array("name" => "joe");

$people->insert($person);

// now $joe has an _id field
$joe $people->findOne(array("_id" => $person['_id']));
?>

Unless the user has specified otherwise, the _id field is a MongoId. The most common mistake is attempting to use a string to match a MongoId. Keep in mind that these are two different datatypes, and will not match each other in the same way that the string "array()" is not the same as an empty array. For example:

<?php
$person 
= array("name" => "joe");

$people->insert($person);

// convert the _id to a string
$pid $person['_id'] . "";

// FAILS - $pid is a string, not a MongoId
$joe $people->findOne(array("_id" => $pid));
?>

Arrays

Arrays are special in a couple ways. First, there are two types that MongoDB uses: "normal" arrays and associative arrays. Associative arrays can have any mix of key types and values. "Normal" arrays are defined as arrays with ascending numeric indexes starting at 0 and increasing by one for each element. These are, typically, just your usual PHP array.

For instance, if you want to save a list of awards in a document, you could say:

<?php

$collection
->save(array("awards" => array("gold""silver""bronze")));

?>

Queries can reach into arrays to search for elements. Suppose that we wish to find all documents with an array element of a given value. For example, documents with a "gold" award, such as:

{ "_id" : ObjectId("4b06c282edb87a281e09dad9"), "awards" : ["gold", "silver", "bronze"]}

This can be done with a simple query, ignoring the fact that "awards" is an array:

<?php

  $cursor 
$collection->find(array("awards" => "gold"));

?>

Suppose we are querying for a more complex object, if each element of the array were an object itself, such as:

{
     "_id" : ObjectId("4b06c282edb87a281e09dad9"),
     "awards" :
     [
        {
            "first place" : "gold"
        },
        {
            "second place" : "silver"
        },
        {
            "third place" :  "bronze"
        }
     ]
}

Still ignoring that this is an array, we can use dot notation to query the subobject:

<?php

$cursor 
$collection->find(array("awards.first place" => "gold"));

?>

Notice that it doesn't matter that there is a space in the field name (although it may be best not to use spaces, just to make things more readable).

You can also use an array to query for a number of possible values. For instance, if we were looking for documents "gold" or "copper", we could do:

<?php

$cursor 
$collection->find(array("awards" => array('$in' => array("gold""copper"))));

?>

Changelog

Version Description
1.3.0 Introduced the Read Preferences framework to allow more fine grained control over secondary reads.
1.3.0 Deprecated slaveOkay usage, the alternative is Read Preferences.
1.1.0 Introduced the possiblity of routing reads to secondaries of replica set members using Mongo::setSlaveOkay()