PHP Mongo MapReduce - How to Call Server Loaded JavaScript Functions - php

Question

First, I want to say that this is more of a PHP Mongo Driver issue than MongoDB issue.
I have a problem with calling MapReduce through PHP. I have 3 custom functions inside Mongo for MapReduce: mapItems, reduceItems, finalizeItems.
When I test my functions from inside Mongo Shell, everything works great:
db.loadServerScripts();
db.runCommand({
mapreduce: 'items',
map: mapItems,
reduce: reduceItems,
finalize: finalizeItems,
out: {inline: 1},
scope: {members: {a1: 0, a2: 0}
});
Now when I try to do the same in PHP Mongo Driver, nothing works.
$db->command([
'mapreduce' => 'items',
'map' => 'mapItems',
'reduce' => 'reduceItems',
'finalize' => 'finalizeItems',
'out' => ['inline' => 1],
'scope' => ['members' => ['a1' => 0, 'a2' => 0]]
]);
I thought maybe I need to use MongoCode for the functions, but that still didn't work:
$db->command([
'mapreduce' => 'items',
'map' => new \MongoCode('mapItems'),
'reduce' => new \MongoCode('reduceItems'),
'finalize' => new \MongoCode('finalizeItems'),
'out' => ['inline' => 1],
'scope' => ['members' => ['a1' => 0, 'a2' => 0]]
]);
Then I thought maybe the PHP driver is being stupid, so I copy pasted the functions into PHP and tried them:
$db->command([
'mapreduce' => 'items',
'map' => new \MongoCode($mapFn),
'reduce' => new \MongoCode($reduceFn),
'finalize' => new \MongoCode($finalizeFn),
'out' => ['inline' => 1],
'scope' => ['members' => ['a1' => 0, 'a2' => 0]]
]);
And this worked.
How do I need to pass the internal functions to Mongo through PHP?

Answer 1

What you have done so far
This is a classic case of things not working how you seem to think they work. Let's break this down logically for everyone to understand.
So what you clearly have done to date is create some JavaScript function definitions and stored them in system.js. Now anything that has been stored in such a way can be called by server side JavaScript operations, which is fine. An example:
db.system.js.save({
"_id": squareThis",
"value": function(x) { return x*x; }
})
db.test.insert({ "a": 3 ))
db.test.find(function() { return squareThis(this.a) == 9 })
So that works as expected where the "server" can use the function. Not if you did this ( as you mention earlier ) :
db.loadServerScripts();
Now any functions in system.js are now available for the "shell client" to use as well. So you can now do this in the JavaScript REPL:
> squareThis(3) == 9
true
What you are doing wrong
Now you want to code with PHP. So you go to call your JavaScript methods you created on the "server" from your PHP code and you try this:
'map' => new \MongoCode('mapItems'),
So what is MongoCode? Well it basically defines a BSON type for a code reference ( expected to be JavaScript ) and sends that to the server.
The clear problem here is that "mapItems" is not JavaScript code. You "shell client" recognized that, but only after calling db.loadServerScripts(), which of course it can do, because that client understands JavaScript.
The same is true for the earlier attempt. Your PHP mapReduce command expects to have MongoCode objects that actually contain valid JavaScript. These are going to be sent to the server where it can acutally call functions defined there, but you still need to provide the mapReduce command with "real code" that does something.
Fixing It
Fortunately there is a way around this, in that you can basically "define a function to call a function" essentially "wrapping" the method calls so the "server" loaded code can be called.
The sytax varies depending on if the function is supposed to return anything so:
$db->command([
'mapreduce' => 'items',
'map' => new \MongoCode('function() { mapItems(this) }'),
'reduce' => new \MongoCode('function(key,values) { return reduceItems(key,values) }'),
'finalize' => new \MongoCode('function(key,value) { return finalizeItems(key,value)}'),
'out' => ['inline' => 1],
'scope' => ['members' => ['a1' => 0, 'a2' => 0]]
]);
So now your PHP mapReduce command sends up it's "packaged command" with valid JavaScript functions attached to the appropriate method stages. These "code wrappers" execute on the "server" and can then call the respective "server" methods that are stored in system.js.
The reason for the explicit argument passing i.e
function(key,values){ serverFunction(key,values) }
Is because your "wrappers" need to create the same sort of "signatures" as expected by the functions of "mapReduce" in general. And since they are "calling" the other methods "server side", then you need to "pass-through" the expected parameters.
This also has another implication for the "mapper" function. You are likely referring to all document content through the this context of JavaScript objects. Once you do this then you cannot do that anymore and will need to "explicitly" define a document argument to your "mapper"
db.system.js({
"_id": "mapper",
"value": function(doc) {
// use doc instead of this
}
And then even from the Mongo Shell you will also need to call like this:
db.runCommand({
"mapReduce": "items",
"mapper": function() { mapper(this) }
In order for the logic to work.
So server side functions are not really what you thought they are. You can use them, but you need to follow the rules of how they can be used.
In general it is probably better that you simply just code all of these up in your client rather than jump through all the hoops. It is afterall "JavaScript" execution for which the only language is "JavaScript", so that is why alternate methods of processing are generally preferred as long as they can be applied.
And if your document structure needs JavaScript processing, then you probably have your document structure wrong in the first place. So the better case here is "re-evaluate" your structure and processing needs to get you "cleaner" and "faster" code.