MongoDB .local Paris 2020: La puissance du Pipeline d'Agrégation de MongoDB

#MDBlocal
How MongoDB 4.2 Pipeline is
Powering Queries, Updates and Views
Guillaume Meister
Principal Solutions Architect
AGGREGATION POWER++

PREVIOUSLY ...
... 2017 and before

#MDBW17
Analytics with MongoDB Aggregation Framework
@asya999 by Asya Kamsky,
Lead MongoDB Maven
PIPELINE POWER

#MDBLocal
ps ax |grep mongod |head 1
*nix command line pipe
PIPELINE

#MDBLocal
$match $group | $sort|
Input stream {} {} {} {} Result {} {} ...
PIPELINE
MongoDB document pipeline

Stage 1 Stage 2 Stage 3 Stage 4
{} {} {} {}
{} {} {} {}
DATA PIPELINE
{} {} {} {}
{"$stage":{ ... }}
START
Collection
View
Special stage
STAGES

{title: "The Great Gatsby",
language: "English",
subjects: "Long Island"}
subjects: "New York"}
subjects: "1920s"}
subjects: [
"Long Island",
"New York",
"1920s"] },
{"$match":{"language":"English"}}
$match
{ _id:"Long Island",
count: 1 },
$group
{ _id: "New York",
count: 2 },
$unwind
{ _id: "1920s",
count: 1 },
$sort $skip$limit $project
{"$unwind":"$subjects"}
{"$group":{"_id":"$subjects", "count":{"$sum:1}}
{ _id: "Harlem",
count: 1 },
{ _id: "Long Island",
count: 1 },
{ _id: "New York",
count: 2 },
{ _id: "1920s",
count: 1 },
{title: "Open City",
subjects: [
"New York"
"Harlem" ] }
{ title: "The Great Gatsby",
subjects: [
"Long Island",
"New York",
"1920s"] },
{ title: "War and Peace",
language: "Russian",
subjects: [
"Russia",
"War of 1812",
"Napoleon"] },
{ title: "Open City",
subjects: [
"New York",
"Harlem" ] },
subjects: "New York"}
subjects: "Harlem"}
{ _id: "Harlem",
count: 1 },
{"$sort:{"count":-1} {"$limit":3}
{"$project":...}

#MDBLocal
INPUT STAGE RESULTSSTAGE
STREAMING RESOURCE USE
Each document is streamed through in RAM

#MDBLocal
INPUT STAGE RESULTSSTAGE
BLOCKING RESOURCE USE
Everything has to be kept in RAM (or spill)

5 minute review
https://github.com/asya999/mdbw17

PREVIOUSLY ...
... 2017 ... 2018

#MDBLocal
THE FUTURE OF AGGREGATION
Better performance & optimizations
More stages & expressions
More options for output
Compass helper for aggregate
Unify different languages

#MDBLocal

#MDBLocal
THE PRESENT OF AGGREGATION

#MDBLocal
Unify Different Languages

#MDBLocal
{children: [
{name:"Max", dob:"1994-12-01", dep:true},
{name:"Sam", dob:"1997-09-28", dep:true},
{name:"Kim", dob:"2000-02-29", dep:true}
]}
AGGREGATION

#MDBLocal
{children: [
]}
AGGREGATION
db.c.aggregate([
{$addFields:{
numChildren:{$size:"$children"},
numDependents:{$size:{
$filter:{
input:"$children.dep",
cond: "$$this"
}
}}
}},
...
])

#MDBLocal
{children: [
]}
AGGREGATION
FIND
db.c.aggregate([
{$addFields:{
numChildren:{$size:"$children"},
numDependents:{$size:{
$filter:{
input:"$children.dep",
cond: "$$this"
}
}}
}},
...
])

#MDBLocal
{children: [
]}
AGGREGATION
FIND
db.c.find (
{$expr:{
$lt:[
{$size:{$filter:{
input: "$children.dep",
cond: "$$this"
}}},
2
]
}}
)

#MDBLocal
{children: [
]}
AGGREGATION
FIND
UPDATE
db.c.find (
{$expr:{
$lt:[
{$size:{$filter:{
input: "$children.dep",
cond: "$$this"
}}},
2
]
}}
)

#MDBLocal
{children: [
]}
AGGREGATION
FIND
UPDATE
db.c.update(
{$expr:{
$anyElementTrue:{$map:{
input:"$children",
in: {$and:[
{$lt:["$$this.dob","1997-01-22"]},
"$$this.dep"
]}
}}
}},
{$set:{ audit:true }}
)

#MDBLocal
Update
db.coll.update(
<query>,
<update>,
<options>
)

#MDBLocal
Update
db.coll.update(
<query>,
<update>,
<options>
)
<update>

#MDBLocal
Update
{
f1: <value>,
f2: <value>,
...
}
{
$set: { },
$inc: { },
$...
}
<update>

#MDBLocal
Update in 4.2
{ } OR [ ]
<update>

#MDBLocal
Update in 4.2
{ <same> } [ ]
<update>

#MDBLocal
Update in 4.2
{ <same> } [ <aggregation-pipeline> ]
<update>

Updates Using Aggregation
Pipeline

#MDBLocal
{ $addFields: { } }
{ $project: { } }
{ $replaceRoot: { } }
{ $set: { } }
{ $unset: [ ] }
{ $replaceWith: { } }

#MDBLocal
db.coll.update({_id:1},
{$inc:{a:1}},
{upsert:true})
{ _id: 1 }
{ _id: 1, a: 10 }
{ _id: 1, a: 100 }
---
{ _id: 1, a: "10" }
{ _id: 1, a: 1 }
{ _id: 1, a: 11 }
{ _id: 1, a: 101 }
{ _id: 1, a: 1 }
"errmsg" : "Cannot apply to a value of
non-numeric type."

#MDBLocal
{ _id: 1 }
{ _id: 1, a: 10 }
{ _id: 1, a: 100 }
---
{ _id: 1, a: "10" }
{ _id: 1, a: 1 }
{ _id: 1, a: 11 }
{ _id: 1, a: 101 }
{ _id: 1, a: 1 }
{ _id: 1, a: 1 }
[ {$set:{a:{$sum:["$a",1]}}} ],
{upsert:true})

#MDBLocal
{ _id: 1 }
{ _id: 1, a: 10 }
{ _id: 1, a: 100 }
---
{ _id: 1, a: "10" }
{ _id: 1, a: 1 }
{ _id: 1, a: 11 }
{ _id: 1, a: 101 }
{ _id: 1, a: 1 }
"errmsg" : "$add only supports
numeric or date types, not string"
[ {$set:{a:{$add:["$a",1]}}} ],
{upsert:true})

#MDBLocal
{ _id: 1 }
{ _id: 1, a: 10 }
{ _id: 1, a: 100 }
---
{ _id: 1, a: "10" }
{ _id:1, a: 21 }
{ _id: 1, a: 11 }
{ _id: 1, a: 101 }
{ _id:1, a: 21 }
[ {$set:{a:{$ }} ],
{upsert:true})

#MDBLocal
{ _id: 1 }
{ _id: 1, a: 10 }
{ _id: 1, a: 100 }
---
{ _id: 1, a: "10" }
{ _id:1, a: 21 }
{ _id: 1, a: 11 }
{ _id: 1, a: 101 }
{ _id:1, a: 21 }
db.coll.update({_id:1}, [ {$set:{a:{$cond:{
if: ,
then: , else: }}}}], {upsert:true})

#MDBLocal
{ _id: 1 }
{ _id: 1, a: 10 }
{ _id: 1, a: 100 }
---
{ _id: 1, a: "10" }
{ _id:1, a: 21 }
{ _id: 1, a: 11 }
{ _id: 1, a: 101 }
{ _id:1, a: 21 }
if: {$eq:[{$type:"$a"},"missing"]},
then: , else: }}}}], {upsert:true})

#MDBLocal
{ _id: 1 }
{ _id: 1, a: 10 }
{ _id: 1, a: 100 }
---
{ _id: 1, a: "10" }
{ _id:1, a: 21 }
{ _id: 1, a: 11 }
{ _id: 1, a: 101 }
{ _id:1, a: 21 }
then: 21, else: {$sum:["$a",1]} }}}}], {upsert:true})

#MDBLocal
{ _id: 1 }
{ _id: 1, a: 10 }
{ _id: 1, a: 100 }
---
{ _id: 1, a: "10" }
{ _id:1, a: 21 }
{ _id: 1, a: 11 }
{ _id: 1, a: 100 }
{ _id:1, a: 21 }
then: 21, else: {$sum:["$a",1]} }}}}], {upsert:true})

#MDBLocal
{ _id: 1 }
{ _id: 1, a: 10 }
{ _id: 1, a: 100 }
---
{ _id: 1, a: "10" }
{ _id:1, a: 21 }
{ _id: 1, a: 11 }
{ _id: 1, a: 100 }
{ _id:1, a: 21 }
db.coll.update({_id:1}, [ {$set:{a:{$min:[100, {$cond:{
then: 21, else: {$sum:["$a",1]} }}]} }}], {upsert:true})

#MDBLocal
{ _id: 1 }
{ _id: 1, a: 10 }
{ _id: 1, a: 100 }
---
{ _id: 1, a: "10" }
{ _id:1, a: 21 }
{ _id: 1, a: 11 }
{ _id: 1, a: 100 }
{ _id:1, a: 21 }
{ _id:1, a: 1 }
then: 21, else: {$sum:["$a",1]} }}]} }}], {upsert:true})

#MDBLocal
{ _id: 1 }
{ _id: 1, a: 10 }
{ _id: 1, a: 100 }
---
{ _id: 1, a: "10" }
{ _id:1, a: 21 }
{ _id: 1, a: 11 }
{ _id: 1, a: 100 }
{ _id:1, a: 21 }
{ _id:1, a: 1 }
then: 21, else: {$sum:["$a",1]} }}]}, prev_a:"$a" }}],
{upsert:true})

#MDBLocal
{ _id: 1 }
{ _id: 1, a: 10 }
{ _id: 1, a: 100 }
---
{ _id: 1, a: "10" }
{ _id:1, a: 21 }
{ _id: 1, a: 11, prev_a: 10 }
{ _id: 1, a: 100, prev_a: 100 }
{ _id:1, a: 21 }
{ _id:1, a: 1, prev_a: "10" }
then: 21, else: {$sum:["$a",1]} }}]}, prev_a:"$a" }}],
{upsert:true})

#MDBLocal
Set Defaults
{_id: 1, a: 5, b: 12}
{_id: 2, a: 15, c: "abc"}
{_id: 3, b: 99, c: "xyz"}
If a or b are missing, set to 0, if c is missing -> "unset"

#MDBLocal
Set Defaults
{_id: 1, a: 5, b: 12}
{_id: 2, a: 15, c: "abc"}
{_id: 3, b: 99, c: "xyz"}
db.coll.update({}, [
{$replaceWith:{
}}
], {multi:true})

#MDBLocal
Set Defaults
{_id: 1, a: 5, b: 12}
{_id: 2, a: 15, c: "abc"}
{_id: 3, b: 99, c: "xyz"}
{$replaceWith:{$mergeObjects:[
]}}
], {multi:true})

#MDBLocal
Set Defaults
{_id: 1, a: 5, b: 12}
{_id: 2, a: 15, c: "abc"}
{_id: 3, b: 99, c: "xyz"}
{ a:0, b:0, c:"unset" },
"$$ROOT"
]}}
], {multi:true})

#MDBLocal
Set Defaults
{_id: 1, a: 5, b: 12}
{_id: 2, a: 15, c: "abc"}
{_id: 3, b: 99, c: "xyz"}
{ a:0, b:0, c:"unset" },
"$$ROOT"
]}}
], {multi:true})
{_id: 1, a: 5, b: 12, c: "unset"}
{_id: 2, a: 15, b: 0, c: "abc"}
{_id: 3, a: 0, b: 99, c: "xyz"}

#MDBLocal
{ id: 1,
d: ISODate("2019-06-04T00:00:00"),
h: [
{ hour:"11", value: 296 },
{ hour:"12", value: 300 }
]}
id: X, d:Y, hour:Z, value: VAL
db.coll.update({id:X, d:Y},
[ {$set:{h:{$cond:{
if:
then:
else:
}}}}],
{upsert:true})

#MDBLocal
{ id: 1,
d: ISODate("2019-06-04T00:00:00"),
h: [
{ hour:"11", value: 296 },
{ hour:"12", value: 300 }
]}
id: X, d:Y, hour:Z, value: VAL
db.coll.update({id:X, d:Y},
[ {$set:{h:{$cond:{
if: {$in:[Z,{$ifNull:["$h.hour",[]]}]},
then:{$map:{
input:"$h",
in: {$cond:{ if:{$ne:["$$this.hour",Z]}, then:"$$this",
else: {hour: Z, value: {$sum:[ "$$this.value", VAL]}}
}}}},
else:{$concatArrays:[{$ifNull:["$h",[]]},[{hour:Z,value:VAL}]]}
}}}}],
{upsert:true})
if:
then:
else:

#MDBLocal
Recap:
Updates can be specified with aggregation pipeline
All fields from existing document can be accessed
Slightly slower, but a lot more powerful

#MDBLocal

#MDBLocal
More Options for Output

#MDBLocal
Prior to MongoDB 4.2
$out
coll
new_coll
$out

#MDBLocal
$out
coll
new_coll
$out
db.coll.aggregate( [ {pipeline}, ...
{$out: "new_coll"} ]);

#MDBLocal
$out
coll
new_coll
$out
db.coll.aggregate( [ {pipeline}, ...
{$out: "new_coll"} ]);
new_coll
○ must be unsharded
○ overwrites existing

New $merge stage
in aggregation pipeline

#MDBLocal
MongoDB 4.2
$merge
coll
coll2
$merge

#MDBLocal
MongoDB 4.2
$merge
db.coll.aggregate( [
{pipeline}, ...,
{$merge: { ... }
]);
coll
coll2
$merge

#MDBLocal
MongoDB 4.2
$merge
db.coll.aggregate( [
{pipeline}, ...,
{$merge: { ... }
]);
coll2
can exist, can be sharded
same or different 'db'
coll
coll2
$merge

#MDBLocal
coll
coll2
$merge
{ } { } { } { }
{ } { } { } { }
MongoDB 4.2

#MDBLocal
{
$merge: {
into: <target>
}
}
$merge syntax

#MDBLocal
{$merge: "collection2"}
$merge syntax
{
$merge: {
into: <target>
}
}

#MDBLocal
{$merge: {into: {db: "db2", coll: "collection2"}}
$merge syntax
{
$merge: {
into: <target>
}
}

#MDBLocal
{
$merge: {
into: <target>,
on: <fields>
}
}
on: "_id"
on: [ "_id", "shardkey(s)" ]
must be unique
$merge syntax

#MDBLocal
{
$merge: {
into: <target>,
on: <fields>
}
}
$merge syntax

#MDBLocal
Actions
source target

#MDBLocal
Actions
nothing matched:
source target

#MDBLocal
Actions
nothing matched: usually insert
source target

#MDBLocal
Actions
document matched:
source target

#MDBLocal
Actions
document matched: overwrite? update? ???
source target

#MDBLocal
Actions
document matched: update
source target

#MDBLocal
Actions
document matched: update (merge)
source target

#MDBLocal
$merge syntax
{
$merge: {
into: <target>,
whenNotMatched:
whenMatched:
}
}

#MDBLocal
$merge syntax
{
$merge: {
into: <target>,
whenNotMatched:"insert",
whenMatched:
}
}

#MDBLocal
$merge syntax
{
$merge: {
into: <target>,
whenNotMatched:"insert",
whenMatched:"merge"
}
}

#MDBLocal
$merge syntax
{
$merge: {
into: <target>,
whenNotMatched:"insert"|"discard"|"fail",
whenMatched:"merge"
}
}

#MDBLocal
$merge syntax
{
$merge: {
into: <target>,
whenNotMatched:"insert"|"discard"|"fail",
whenMatched:"merge"|"replace"|"keepExisting"|"fail"|[...]
}
}

#MDBLocal
$merge syntax
{
$merge: {
into: <target>,
whenMatched:[...]
}
}

#MDBLocal
$merge syntax
{
$merge: {
into: <target>,
whenMatched:[<custom pipeline>]
}
}

#MDBLocal
$merge example
{
$merge: {
into: <target>,
whenMatched:[
{$addFields:{
}}
]
}
}

#MDBLocal
$merge example
{
$merge: {
into: <target>,
whenMatched:[
{$addFields:{
total:{$sum:["$total","$$new.total"]}
}}
]
}
}

#MDBLocal
$merge example
{
$merge: {
into: <target>,
whenMatched:[
{$set:{
}}
]
}
}

#MDBLocal
$merge example
{
$merge: {
into: <target>,
whenMatched:[
{$set:{
}}
]
}
}
Incoming Target
{
_id: "37",
total: 64,
f1: "x"
}
{
_id: "37",
total: 245,
f1: "yyy"
}
Result:
{
}

#MDBLocal
$merge example
{
$merge: {
into: <target>,
whenMatched:[
{$set:{
}}
]
}
}
Incoming Target
{
_id: "37",
total: 64,
f1: "x"
}
{
_id: "37",
total: 245,
f1: "yyy"
}
Result:
{
_id: "37",
total: 309,
f1: "yyy"
}

#MDBLocal
$merge example 2
{
$merge: {
into: <target>,
whenMatched:[
"$$new",
{total:{$sum:["$$new.total", "$total"]}}
]}}
]
}
}

#MDBLocal
$merge example 2
{
$merge: {
into: <target>,
whenMatched:[
"$$new",
]}}
]
}
}
Incoming Target
{
_id: "37",
total: 64,
f1: "x"
}
{
_id: "37",
total: 245,
f1: "yyy"
}
Result:
{
}

#MDBLocal
$merge example 2
{
$merge: {
into: <target>,
whenMatched:[
"$$new",
]}}
]
}
}
Incoming Target
{
_id: "37",
total: 64,
f1: "x"
}
{
_id: "37",
total: 245,
f1: "yyy"
}
Result:
{
_id: "37",
total: 309,
f1: "x"
}

#MDBLocal
$merge syntax
{
$merge: {
into: <target>,
let: { ... },
whenMatched:[ ...]
}
}

#MDBLocal
$merge syntax
{
$merge: {
into: <target>,
let: {new: "$$ROOT"},
whenMatched:[ ...]
}
}

#MDBLocal
{
$merge: {
into: <target>,
whenMatched:[
{$set:{
}}
]
}
}

#MDBLocal
{
$merge: {
into: <target>,
let: {itotal: "$total"},
whenMatched:[
{$set:{
total:{$sum:["$total","$$itotal"]}
}}
]
}
}
{
$merge: {
into: <target>,
whenMatched:[
{$set:{
}}
]
}
}

EXAMPLES
APPEND from TEMP collection

#MDBLocal
temp
real
data
real
Using $merge to append loaded and
cleansed records loaded into db

#MDBLocal
aggregate 'temp' and append valid records to 'data'
db.temp.aggregate( [
{ ... } /* pipeline to massage and cleanse data in temp */,
{$merge:{
into: "data",
whenMatched: "fail"
}}
]);

#MDBLocal
aggregate 'temp' and append valid records to 'data'
db.temp.aggregate( [
{ ... } /* pipeline to massage and cleanse data in temp */,
{$merge:{
into: "data",
whenMatched: "fail"
}}
]);
Similar to SQL's INSERT INTO T1 SELECT * from T2

#MDBLocal
mflix
users
users
mfriendbook
users
sv
Using $merge to populate/update
user fields from other services

#MDBLocal
mflix
users
users
mfriendbook
users
sv
Using $merge to populate/update
user fields from other services
sv.users
{
_id: "user253",
dob: ISODate(...),
f1: "yyy"
}

#MDBLocal
$merge updates fields from mflix.users collection into
sv.users collection. Our "_id" field is unique username
mflix_pipeline = [
{ "$project" : {
"_id" : "$username",
"mflix" : "$$ROOT"
}},
{ "$merge" : {
"into" : {
"db": "sv",
"collection" : "users"
},
"whenNotMatched" : "discard"
}}
]
(in mflix)
sv.users
{
_id: "user253",
dob: ISODate(...),
f1: "yyy"
}

#MDBLocal
$merge updates fields from mflix.users collection into
mflix_pipeline = [
{ "$project" : {
"mflix" : "$$ROOT"
}},
{ "$merge" : {
"into" : {
"db": "sv",
},
}}
]
(in mflix) db.users.aggregate(mflix_pipeline)
sv.users
{
_id: "user253",
dob: ISODate(...),
f1: "yyy",
mflix: { ... }
}

#MDBLocal
$merge updates fields from mfriendbook.users collection into
mfriendbook_pipeline = [
{ "$project" : {
"mfriendbook" : "$$ROOT"
}},
{ "$merge" : {
"into" : {
"db": "sv",
},
}}
]
(in mfriendbook)
sv.users
{
_id: "user253",
dob: ISODate(...),
f1: "yyy",
mflix: { ... }
}

#MDBLocal
$merge updates fields from mfriendbook.users collection into
mfriendbook_pipeline = [
{ "$project" : {
"mfriendbook" : "$$ROOT"
}},
{ "$merge" : {
"into" : {
"db": "sv",
},
}}
]
(in mfriendbook) db.users.aggregate(mfriendbook_pipeline)
sv.users
{
_id: "user253",
dob: ISODate(...),
f1: "yyy",
mflix: { ... },
mfriendbook: { ... }
}

EXAMPLES
Populate ROLLUPS into summary table

registrations
real
regsummary
real
Using $merge to incrementally
update periodic rollups in summary

#MDBLocal
$merge to create/update periodic
rollups in summary collection (for all days)
db.regsummary.createIndex({event:1, date:1}, {unique: true});

#MDBLocal
db.registrations.aggregate([
{$match: {event_id: "MDBW19"}},
{$group:{
_id:{$dateToString:{date:"$date",format:"%Y-%m-%d"}},
count: {$sum:1}
}},
{$project: {_id:0,event:"MDBW19",date:"$_id",total:"$count"}},
{$merge: {
into: "regsummary",
on: ["event", "date"]
}}
])

#MDBLocal
{$match: {event_id: "MDBW19"}},
{$group:{
_id:{$dateToString:{date:"$date",format:"%Y-%m-%d"}},
count: {$sum:1}
}},
{$project: {_id:0,event:"MDBW19",date:"$_id",total:"$count"}},
{$merge: {
into: "regsummary",
}}
])
{ "event" : "MDBW19", "date" : "2019-05-19", "total" : 33 }

#MDBLocal
$merge to incrementally update periodic rollups in
summary collection (for single day)

#MDBLocal
{$match: {
event_id: "MDBW19",
date:{$gte:ISODate("2019-05-22"),$lt:ISODate("2019-05-23")}
}},
{$count: "total"},
{$addFields: {event:"MDBW19", "date":"2019-05-22"}},
{$merge: {
into: "regsummary",
}}
])

#MDBLocal
The aggregation framework is the main language for data
manipulation in MongoDB (unify languages)
It’s now possible to update documents using the aggregation
framework and existing fields (UPDATE)
Aggregation framework output can be used to merge data with a
target collection ($merge)
Key takeaways

MongoDB .local Paris 2020: La puissance du Pipeline d'Agrégation de MongoDB

MongoDB .local Paris 2020: La puissance du Pipeline d'Agrégation de MongoDB

More Related Content

What's hot (20)

Similar to MongoDB .local Paris 2020: La puissance du Pipeline d'Agrégation de MongoDB (20)

More from MongoDB (20)

Recently uploaded (20)

MongoDB .local Paris 2020: La puissance du Pipeline d'Agrégation de MongoDB