Aggregate analytics data by day/next day logic - mongodb

Question

I wrote a server for mobile app analytics in which I have a sharded(!Upd) collection with events as follows:
{
"event": "install",
"userId": "a",
"time": 2014-02-09,
"data" : ...
},
{
"event": "login",
"userId": "a",
"time": 2014-02-12,
"data" : ...
},
{
"event": "install",
"userId": "b",
"time": 2014-4-29,
"data" : ...
},
{
"event": "login",
"userId": "b",
"time": 2014-4-30,
"data" : ...
}
...
I need to select users, who have the event install and not login in next day after the install event (in other words, I want to select Users who install the app, but do not login in next day). So the output for above data should be:
{
"userId": "a",
"data" : ...
}
How do carry out this task with aggregation framework or mapreduce? Or maybe another solution?

Answer 1

This is a bit tricky one :-)
You can do it with aggregation, if time is just a date field (without time data),
then
having collection
{
"_id" : ObjectId("57694365ef9176ec54960a66"),
"event" : "install",
"userId" : "a",
"time" : ISODate("2014-09-02T00:00:00.000Z")
},{
"_id" : ObjectId("57694365ef9176ec54960a67"),
"event" : "login",
"userId" : "a",
"time" : ISODate("2014-12-02T00:00:00.000Z")
},{
"_id" : ObjectId("57694365ef9176ec54960a68"),
"event" : "install",
"userId" : "b",
"time" : ISODate("2014-04-29T00:00:00.000Z")
},{
"_id" : ObjectId("57694365ef9176ec54960a69"),
"event" : "login",
"userId" : "b",
"time" : ISODate("2014-04-30T00:00:00.000Z")
}
we can use aggregate query:
var match = {
$match : {
"event" : "install"
}
};
var projectNextDayDate = {
$project : {
_id : 1,
event : 1,
userId : 1,
time : 1,
nextDay : {
$add : ["$time", 24 * 60 * 60 * 1000]
}
}
}
var lookup = {
$lookup : {
from : "zella",
localField : "nextDay",
foreignField : "time",
as : "mergedDocs"
}
}
var nowMatchUsers = {
$project : {
_id : 1,
event : 1,
userId : 1,
time : 1,
nextDay : 1,
mergedDocs : {
$filter : {
input : "mergedDocs",
as : "m",
cond : {
$eq : ["$$m.userId", "$userId"]
}
}
}
}
}
var findEmptyArrays = {
$match : {
mergedDocs : []
}
}
db.zella.aggregate([match, projectNextDayDate, lookup, findEmptyArrays])
with this output:
{
"_id" : ObjectId("57694365ef9176ec54960a66"),
"event" : "install",
"userId" : "a",
"time" : ISODate("2014-09-02T00:00:00.000Z"),
"nextDay" : ISODate("2014-09-03T00:00:00.000Z"),
"mergedDocs" : []
}
The assumption here is that time is a date 2014-09-02T00:00:00.000
the other way to merge collection is to use user ID, as a $lookup point, but then there will be more logic to filter result set and performance could be dropped.

Answer 2

You could try running the following aggregation pipeline:
db.test.aggregate([
{
"$project": {
"event": 1,
"userId": 1,
"time": 1,
"data": 1,
"dayAfter": {
"$add": [ "$time", 24 * 60 * 60 * 1000 ]
}
}
},
{ "$match": { "event": { "$in": ["install", "login"] } } },
{
"$group": {
"_id": "$userId",
"eventsTimeLine": {
"$push": {
"event": "$event",
"time": "$time",
"dayAfter": "$dayAfter"
}
},
"data": { "$push": "$data" }
}
},
{ "$unwind": "$eventsTimeLine" },
{ "$sort": { "eventsTimeLine.event": 1 } },
{
"$group": {
"_id": "$_id",
"dayAfterInstall": { "$first": "$eventsTimeLine.dayAfter" },
"loginTime": { "$last": "$eventsTimeLine.time" },
"data": { "$first": "$data" }
}
},
{
"$project": {
"isChurn": { "$ne": [ "$loginTime", "$dayAfterInstall" ] },
"userId": "$_id", "data": 1, "_id": 0
}
},
{ "$match" : { "isChurn" : true } }
])

Answer 3

Here another solution with mapreduce and aggregation:
var mapFunction = function() {
if (this.event != 'install' && this.event != 'login'){
return;
}
var value = {data: this.data, count: 1};
if (this.event == 'install'){
var nextDay = new Date(this.date.getTime() + 24 * 60 * 60 * 1000)
emit({userId:this.userId, nextDayAfterInstall:nextDay}, value );
} else
if (this.event == 'login'){
emit({userId:this.userId, nextDayAfterInstall:this.date}, value );
}
};
var reduceFunction = function(event, values) {
var value = { data: null, count: 1 };
for (var index = 0; index < values.length; ++index) {
value.count += values[index].count;
value.data = values[index].data;
}
return value ;
};
db.events.mapReduce(
mapFunction,
reduceFunction,
{ out: "case1_mr_out" }
)
var groupByUserId = {
$group :
{
_id : { userId: "$_id.userId" },
data : { $last: '$value.data' },
count : { $max: '$value.count' }
}
}
var filterWhereOnlyOne = {
$match : {
"count" : 1
}
};
db.case1_mr_out.aggregate([groupByUserId,filterWhereOnlyOne])