I have a large dataset (about 1.1M documents) that I need to run mapreduce on.
The field to group on is an array named xref. Due to the size of the collection and the fact I'm doing this in a 32-bit environment, I'm trying to reduce the collection to another collection in a new database.
First, here's a data sample:
{ "_id" : ObjectId("4ec6d3aa61910ad451f12e01"),
"bii" : -32.9867,
"class" : 2456,
"decdeg" : -82.4856,
"lii" : 297.4896,
"name" : "HD 22237",
"radeg" : 50.3284,
"vmag" : 8,
"xref" : ["HD 22237", "CPD -82 65", "-82 64","PPM 376283", "SAO 258336",
"CP-82 65","GC 4125" ] }
{ "_id" : ObjectId("4ec6d44661910ad451f78eba"),
"bii" : -32.9901,
"class" : 2450,
"decdeg" : -82.4781,
"decpm" : 0.013,
"lii" : 297.4807,
"name" : "PPM 376283",
"radeg" : 50.3543,
"rapm" : 0.0357,
"vmag" : 8.4,
"xref" : ["HD 22237", "CPD -82 65", "-82 64","PPM 376283", "SAO 258336",
"CP-82 65","GC 4125" ] }
{ "_id" : ObjectId("4ec6d48a61910ad451feae04"),
"bii" : -32.9903,
"class" : 2450,
"decdeg" : -82.4779,
"decpm" : 0.027,
"hd_component" : 0,
"lii" : 297.4806,
"name" : "SAO 258336",
"radeg" : 50.3543,
"rapm" : 0.0355,
"vmag" : 8,
"xref" : ["HD 22237", "CPD -82 65", "-82 64","PPM 376283", "SAO 258336",
"CP-82 65","GC 4125" ] }
Here are the map and reduce functions (right now I'm only lii and bii fields):
function map() {
try {
emit(this.xref, {lii:this.lii, bii:this.bii});
} catch(e) {
}
}
function reduce(key, values) {
var result = {xref:key, lii: 0.0, bii: 0.0};
try {
values.forEach(function(value) {
if (value.lii && value.bii) {
result.lii += value.lii;
result.bii += value.bii;
}
});
result.bii /= values.length;
result.lii /= values.length;
} catch(e) {
}
return result;
}
Unfortunately, running this eventually comes up with an error message:
db.catalog.mapReduce(map, reduce, {out:{replace:"catalog2", db:"astro2"}});
Wed Nov 23 10:12:25 uncaught exception: map reduce failed:{
"assertion" : "_id cannot be an array",
"assertionCode" : 10099,
"errmsg" : "db assertion failure",
"ok" : 0
The xref field IS an array, but all values are equal in that array. Is it trying to use that array as the id field in the new collections?
Yes it is not possible to set _id as an array, because it has a special behavior for indexing.
The key you emit by is used as _id in the output collection.
Potentially this could work only with an "inline" output mode if the result is small, since it wont go to a collection.
But ideally you would translate the array into a string (for example concat the values) and use that as _id, or make it a sub-object instead of an array.
Also note that the result of your reduce function should not include the key.
Just return {lii: .., bii: ..}
Related
My MongoDB collection is made up of 2 main collections :
1) Maps
{
"_id" : ObjectId("542489232436657966204394"),
"fileName" : "importFile1.json",
"territories" : [
{
"$ref" : "territories",
"$id" : ObjectId("5424892224366579662042e9")
},
{
"$ref" : "territories",
"$id" : ObjectId("5424892224366579662042ea")
}
]
},
{
"_id" : ObjectId("542489262436657966204398"),
"fileName" : "importFile2.json",
"territories" : [
{
"$ref" : "territories",
"$id" : ObjectId("542489232436657966204395")
}
],
"uploadDate" : ISODate("2012-08-22T09:06:40.000Z")
}
2) Territories, which are referenced in "Map" objects :
{
"_id" : ObjectId("5424892224366579662042e9"),
"name" : "Afghanistan",
"area" : 653958
},
{
"_id" : ObjectId("5424892224366579662042ea"),
"name" : "Angola",
"area" : 1252651
},
{
"_id" : ObjectId("542489232436657966204395"),
"name" : "Unknown",
"area" : 0
}
My objective is to list every map with their cumulative area and number of territories. I am trying the following query :
db.maps.aggregate(
{'$unwind':'$territories'},
{'$group':{
'_id':'$fileName',
'numberOf': {'$sum': '$territories.name'},
'locatedArea':{'$sum':'$territories.area'}
}
})
However the results show 0 for each of these values :
{
"result" : [
{
"_id" : "importFile2.json",
"numberOf" : 0,
"locatedArea" : 0
},
{
"_id" : "importFile1.json",
"numberOf" : 0,
"locatedArea" : 0
}
],
"ok" : 1
}
I probably did something wrong when trying to access to the member variables of Territory (name and area), but I couldn't find an example of such a case in the Mongo doc. area is stored as an integer, and name as a string.
I probably did something wrong when trying to access to the member variables of Territory (name and area), but I couldn't find an example
of such a case in the Mongo doc. area is stored as an integer, and
name as a string.
Yes indeed, the field "territories" has an array of database references and not the actual documents. DBRefs are objects that contain information with which we can locate the actual documents.
In the above example, you can clearly see this, fire the below mongo query:
db.maps.find({"_id":ObjectId("542489232436657966204394")}).forEach(function(do
c){print(doc.territories[0]);})
it will print the DBRef object rather than the document itself:
o/p: DBRef("territories", ObjectId("5424892224366579662042e9"))
so, '$sum': '$territories.name','$sum': '$territories.area' would show you '0' since there are no fields such as name or area.
So you need to resolve this reference to a document before doing something like $territories.name
To achieve what you want, you can make use of the map() function, since aggregation nor Map-reduce support sub queries, and you already have a self-contained map document, with references to its territories.
Steps to achieve:
a) get each map
b) resolve the `DBRef`.
c) calculate the total area, and the number of territories.
d) make and return the desired structure.
Mongo shell script:
db.maps.find().map(function(doc) {
var territory_refs = doc.territories.map(function(terr_ref) {
refName = terr_ref.$ref;
return terr_ref.$id;
});
var areaSum = 0;
db.refName.find({
"_id" : {
$in : territory_refs
}
}).forEach(function(i) {
areaSum += i.area;
});
return {
"id" : doc.fileName,
"noOfTerritories" : territory_refs.length,
"areaSum" : areaSum
};
})
o/p:
[
{
"id" : "importFile1.json",
"noOfTerritories" : 2,
"areaSum" : 1906609
},
{
"id" : "importFile2.json",
"noOfTerritories" : 1,
"areaSum" : 0
}
]
Map-Reduce functions should not be and cannot be used to resolve DBRefs in the server side.
See what the documentation has to say:
The map function should not access the database for any reason.
The map function should be pure, or have no impact outside of the
function (i.e. side effects.)
The reduce function should not access the database, even to perform
read operations. The reduce function should not affect the outside
system.
Moreover, a reduce function even if used(which can never work anyway) will never be called for your problem, since a group w.r.t "fileName" or "ObjectId" would always have only one document, in your dataset.
MongoDB will not call the reduce function for a key that has only a
single value
I have a collection that stores search query logs. It's two main attributes are user_id and search_query. user_id is null for a logged out user. I am trying to run a mapreduce job to find out the count and terms per user.
var map = function(){
if(this.user_id !== null){
emit(this.user_id, this.search_query);
}
}
var reduce = function(id, queries){
return Array.sum(queries + ",");
}
db.searchhistories.mapReduce(map,
reduce,
{
query: { "time" : {
$gte : ISODate("2013-10-26T14:40:00.000Z"),
$lt : ISODate("2013-10-26T14:45:00.000Z")
}
},
out : "mr2"
}
)
throws the following exception
Wed Nov 27 06:00:07 uncaught exception: map reduce failed:{
"errmsg" : "exception: assertion src/mongo/db/commands/mr.cpp:760",
"code" : 0,
"ok" : 0
}
I looked at mr.cpp L#760 but could not gather any vital information. What could be causing this?
My Collection has values like
> db.searchhistories.find()
{ "_id" : ObjectId("5247a9e03815ef4a2a005d8b"), "results" : 82883, "response_time" : 0.86, "time" : ISODate("2013-09-29T04:17:36.768Z"), "type" : 0, "user_id" : null, "search_query" : "awareness campaign" }
{ "_id" : ObjectId("5247a9e0606c791838005cba"), "results" : 39545, "response_time" : 0.369, "time" : ISODate("2013-09-29T04:17:36.794Z"), "type" : 0, "user_id" : 34225174, "search_query" : "eficaz eficiencia efectividad" }
Looking at the docs I could see that this is not possible in the slave. It will work perfectly fine in the master though. If you still want to use the slave then you have to use the following syntax.
db.searchhistories.mapReduce(map,
reduce,
{
query: { "time" : {
$gte : ISODate("2013-10-26T14:40:00.000Z"),
$lt : ISODate("2013-10-26T14:45:00.000Z")
}
},
out : { inline : 1 }
}
)
** Ensure that the output document size does not exceed 16MB limit while using inline function.
In my database, I have millions of documents. Each of them has a time stamp. Some have the same time stamp. I want to get some points (a few hundreds or potentially more like thousands) to draw a graph. I don't want all the points. I want every n points I pick 1 point. I know there's aggregation framework and I tried that. The problem with that is since my data is huge. When I do aggregation work, The result exceeds document maximum size, 16MB, easily. There's also a function called skip in mongodb but it only skips first n documents. Are there good ways to achieve what I want? Or is there way to make aggregation result bigger? Thanks in advance!
I'm not sure how you can do this with either A/F or M/R - just skipping so that you have (f.e.) each 10th point is not something M/R allows you to do—unless you select each point based on a random value with a 10% change... which is probably not what you want. But that does work:
db.so.output.drop();
db.so.find().count();
map = function() {
// rand does 0-1, so < 0.1 means 10%
if (Math.random() < 0.1) {
emit(this._id, this);
}
}
reduce = function(key, values) {
return values;
}
db.so.mapReduce( map, reduce, { out: 'output' } );
db.output.find();
Which outputs something line:
{
"result" : "output",
"timeMillis" : 4,
"counts" : {
"input" : 23,
"emit" : 3,
"reduce" : 0,
"output" : 3
},
"ok" : 1,
}
> db.output.find();
{ "_id" : ObjectId("51ffc4bc16473d7b84172d85"), "value" : { "_id" : ObjectId("51ffc4bc16473d7b84172d85"), "date" : ISODate("2013-08-05T15:24:45Z") } }
{ "_id" : ObjectId("51ffc75316473d7b84172d8e"), "value" : { "_id" : ObjectId("51ffc75316473d7b84172d8e") } }
{ "_id" : ObjectId("51ffc75316473d7b84172d8f"), "value" : { "_id" : ObjectId("51ffc75316473d7b84172d8f") } }
or:
> db.so.mapReduce( map, reduce, { out: 'output' } );
{
"result" : "output",
"timeMillis" : 19,
"counts" : {
"input" : 23,
"emit" : 2,
"reduce" : 0,
"output" : 2
},
"ok" : 1,
}
> db.output.find();
{ "_id" : ObjectId("51ffc4bc16473d7b84172d83"), "value" : { "_id" : ObjectId("51ffc4bc16473d7b84172d83"), "date" : ISODate("2013-08-05T15:24:25Z") } }
{ "_id" : ObjectId("51ffc4bc16473d7b84172d86"), "value" : { "_id" : ObjectId("51ffc4bc16473d7b84172d86"), "date" : ISODate("2013-08-05T15:25:15Z") } }
Depending on a random factor.
I have a below map:
var mapFunction = function() {
if(this.url.match(/http:\/\/test.com\/category\/.*?\/checkout/)) {
var key=this.em;
var value = {
url : 'checkout',
count : 1,
account_id:this.accId
}emit(key,value); };
if(this.url.match(/http:\/\/test.com\/landing/)) {
var key=this.em;
var value = {
url : 'landing',
count : 1,
account_id:this.accId
}emit(key,value); };
}
Then I have defined reduce something like below:
var reduceFunction = function (keys, values) {
var reducedValue = {count_checkout:0, count_landing:0};
for (var idx = 0; idx < values.length; idx++) {
if(values[idx].url=='checkout'){
reducedValue.count_checkout++;
}
else {
reducedValue.count_landing++;
}
}
return reducedValue;
}
Now, lets say I have only 1 record:
{
"_id" : ObjectId("516a7cff6dad5949ddf3f7b6"),
"ip" : "1.2.3.4",
"accId" : 123,
"em" : "testing#test.com",
"pgLdTs" : ISODate("2013-04-11T18:30:00Z"),
"url" : "http://test.com/category/prr/checkout",
"domain" : "www.test.com",
"pgUdTs" : ISODate("2013-04-14T09:55:11.682Z"),
"title" : "Test",
"ua" : "Mozilla",
"res" : "1024*768",
"rfr" : "www.google.com"
}
Now if I fire my map reduce like below:
db.test_views.mapReduce(mapFunction,reduceFunction,{out:{inline:1}})
The I get below result returned:
{
"_id" : "testing#test.com",
"value" : {
"url" : "checkout",
"count" : 1,
"account_id" : 123
}
}
So, its basically returning me the map. Now, if I go a add another document for this email id. Finally it becomes something like below.
{
"_id" : ObjectId("516a7cff6dad5949ddf3f7b6"),
"ip" : "1.2.3.4",
"accId" : 123,
"em" : "testing#test.com",
"pgLdTs" : ISODate("2013-04-11T18:30:00Z"),
"url" : "http://test.com/category/prr/checkout",
"domain" : "www.test.com",
"pgUdTs" : ISODate("2013-04-14T09:55:11.682Z"),
"title" : "Test",
"ua" : "Mozilla",
"res" : "1024*768",
"rfr" : "www.google.com"
}
{
"_id" : ObjectId("516a7e1b6dad5949ddf3f7b7"),
"ip" : "1.2.3.4",
"accId" : 123,
"em" : "testing#test.com",
"pgLdTs" : ISODate("2013-04-11T18:30:00Z"),
"url" : "http://test.com/category/prr/checkout",
"domain" : "www.test.com",
"pgUdTs" : ISODate("2013-04-14T09:59:55.326Z"),
"title" : "Test",
"ua" : "Mozilla",
"res" : "1024*768",
"rfr" : "www.google.com"
}
Then, I go again and fire the map reduce, it gives me proper results
{
"_id" : "testing#test.com",
"value" : {
"count_checkout" : 2,
"count_landing" : 0
}
}
Can anyone please help me out in understanding why it returns me a map for single document and doesn't do the counting in reduce.
Thanks for help.
-Lalit
Can anyone please help me out in understanding why it returns me a map for single document and doesn't do the counting in reduce.
The Reduce step combines documents with the same key into a single result document. If you only have one key in the data emitted by your Map function, the data is already "reduced" and the reduce() will not be called.
This is the expected behaviour of the MapReduce algorithm.
The reduce function should return the same type of value objects as the map function emits.
Like you've experienced, when there's a single value associated with a key - the reduce function will not be called at all .
From the MongoDB MapReduce Documentation:
Requirements for the reduce Function:
...
the type of the return object must be identical to the type of the value emitted by the map function to ensure that the following operations is true:
reduce(key, [ C, reduce(key, [ A, B ]) ] ) == reduce( key, [ C, A, B ] )
I have a collection in my database representing IP addresses pulled from various sources. A sample of which looks like this:
{ "_id" : ObjectId("4e71060444dce16174378b79"), "ip" : "xxx.xxx.xxx.xxx", "sources" : { "Source1" : NumberLong(52), "Source2" : NumberLong(7) } }
Each object will have one or more sources.
My goal is to show the number of entries reported by each source without necessarily knowing the names of every possible source (because new ones can potentially be added at any time). I have attempted to address this with map reduce by simply emitting a 1 for each key in the sources hash of each object, but something is wrong with my syntax, it seems. If I do the following:
var map_s = function(){
for(var source in this.sources) {
emit(source, 1);
}
}
var red_s = function(key, values){
var total = 0;
values.forEach(function(){
total++;
});
return total;
}
var op = db.addresses.mapReduce(map_s, red_s, {out: 'results'});
db.results.find().forEach(printjson);
I get
{ "_id" : "Source1", "value" : 12 }
{ "_id" : "Source2", "value" : 230 }
{ "_id" : "Source3", "value" : 358 }
{ "_id" : "Source4", "value" : 398 }
{ "_id" : "Source5", "value" : 39 }
{ "_id" : "Source6", "value" : 420 }
{ "_id" : "Source7", "value" : 156 }
Which is far too small for the database size. For instance, I get the following in the shell if I count off of a specific source:
> db.addresses.count({"sources.Source4": {$exists: true}});
1260538
Where is my error?
Yes there is a problem in your reduce method, it must be idempotent.
Remember that reduce() may be called many times on intermediary results.
Instead of
values.forEach(function(){
total++;
});
You need:
values.forEach(function(x){
total += x;
});