GraphQL as a high-performance�cross-database query language

Predrag Gruevski �

Outline

• Common GraphQL patterns
• What works and what doesn’t

​

• Where and why common patterns fail at scale
• How breaking the GraphQL rules can help

​

• Demo of the GraphQL compiler
• GraphQL macros
• Cross-database queries

​

Outline

• Common GraphQL patterns
• What works and what doesn’t

​

• Why common patterns fail at scale
• How breaking the GraphQL rules can help

​

• Demo of the GraphQL compiler
• GraphQL macros
• Cross-database queries

​

GraphQL as most commonly used

• GraphQL + SQL
• Apollo + Sequelize
• Graphene + SQLAlchemy

• Pros:
• Simple, easy to set up
• Cons:
• Needs lots of custom resolvers
• Limited to simple (shallow) queries due to N+1 queries problem

N+1 queries problem

for data_id in get_ids_from_database():

get_data_from_database(data_id)

​

​

N + 1 = 1 query to look up N identifiers� + N queries to look up their data

​

N+1 queries solution: compile your GraphQL

for data_id in get_ids_from_database():

get_data_from_database(data_id)

​

​

N + 1 = 1 query to look up N identifiers� + N queries to look up their data

​

Avoiding the N+1 problem

• GraphQL as the ORM
• Hasura, Prisma, Postgraphile, Warthog, …

• Pros:
• Deeper GraphQL queries work
• Autogenerated resolvers
• Cons:
• Limited database options (Postgres)
• Schema bloat due to complex autogenerated input GraphQL types

Avoiding the N+1 problem

• GraphQL as the ORM
• Hasura, Prisma, Postgraphile, Warthog, …

• Pros:
• Deeper GraphQL queries work
• Autogenerated resolvers
• Cons:
• Limited database options (Postgres)
• Schema bloat due to complex autogenerated input GraphQL types

https://bit.ly/2kaaOQe

Operations on “createdAt”

Operations on “updatedAt”

Operations on “deletedAt”

Operations on “firstName”

Outline

• Common GraphQL patterns
• What works and what doesn’t

​

• Why common patterns fail at scale
• How breaking GraphQL rules can help

​

• Demo of the GraphQL compiler
• GraphQL macros
• Cross-database queries

​

If your company’s architecture looks like this

then you have truly found paradise, enjoy it!

Paradise Lost: before long, it’ll turn into this

Paradise Regained: can we have this instead?

Paradise Regained: can we have this instead?

common �query interface

Paradise Regained: can we have this instead?

millions of lines of stitched schema

Paradise Regained: can we have this instead?

Product�domain

Infrastructure�domain

Paradise Regained: everyone is happy

• Product / ML engineers focus on business logic

​

​

​

​

• Infrastructure engineers handle underlying implementation

Paradise Regained: everyone is happy

• Product / ML engineers focus on business logic
• Database-agnostic: switching MySQL => Postgres has no visible impact
• Fully declarative: ignorant of data locations, indexes, join order…
• “Free” optimizations: no need to rewrite business logic as infra changes

​

• Infrastructure engineers handle underlying implementation

Paradise Regained: everyone is happy

• Product / ML engineers focus on business logic
• Database-agnostic: switching MySQL => Postgres has no visible impact
• Fully declarative: ignorant of data locations, indexes, join order…
• “Free” optimizations: no need to rewrite business logic as infra changes

​

• Infrastructure engineers handle underlying implementation
• Free to choose the best database(s) for the job
• Free to move and reshape data as necessary
• Able to use advanced database features to speed up everyone’s workloads

Paradise Regained: GraphQL compiler

GraphQL syntax, with more powerful semantics

millions of lines of auto-generated cross-db schema

Rules we have to break

• Nested output format => rows and columns
• Nesting: convenient for React, inconvenient for ML and analytics
• Most databases output rows and columns – even graph databases!
• GraphQL compiler returns rows and columns

​

​

Rules we have to break

• Nested output format => rows and columns
• Nesting: convenient for React, inconvenient for ML and analytics
• Most databases output rows and columns – even graph databases!
• GraphQL compiler returns rows and columns

​

• Naming fields != outputting fields
• Vanilla GraphQL: naming a field is asking for that field to be returned
• GraphQL compiler: fields have to be explicitly marked for output,�allowing them to be operated on by directives without outputting them

​

Outline

• Common GraphQL patterns
• What works and what doesn’t

​

• Why common patterns fail at scale
• How breaking the GraphQL rules can help

​

• Demo of the GraphQL compiler
• GraphQL macros
• Cross-database queries

​

interface GeographicArea {

name: String

uuid: ID

in_GeographicArea_SubArea: [GeographicArea]

out_GeographicArea_SubArea: [GeographicArea]

}

​

type Country implements GeographicArea {

alpha2: String

alpha3: String

name: String

uuid: ID

in_GeographicArea_SubArea: [GeographicArea]

out_GeographicArea_SubArea: [GeographicArea]

in_Airport_BasedIn: [Airport]

in_Airline_RegisteredIn: [Airline]

}

​

type Region implements GeographicArea {

name: String

uuid: ID

in_GeographicArea_SubArea: [GeographicArea]

out_GeographicArea_SubArea: [GeographicArea]

}

interface GeographicArea {

name: String

uuid: ID

in_GeographicArea_SubArea: [GeographicArea]

out_GeographicArea_SubArea: [GeographicArea]

}

​

type Country implements GeographicArea {

alpha2: String

alpha3: String

name: String

uuid: ID

in_GeographicArea_SubArea: [GeographicArea]

out_GeographicArea_SubArea: [GeographicArea]

in_Airport_BasedIn: [Airport]

in_Airline_RegisteredIn: [Airline]

}

​

type Region implements GeographicArea {

name: String

uuid: ID

in_GeographicArea_SubArea: [GeographicArea]

out_GeographicArea_SubArea: [GeographicArea]

}

type Airline {

alpha2_country: String

callsign: String

iata_code: String

icao_code: String

id: Int

name: String

in_FlightRoute_OperatingAirline: [FlightRoute]

out_Airline_RegisteredIn: [Country]

}

​

type Airport {

alpha2_country: String

city_served: String

elevation_ft: Int

iata_code: String

icao_code: String

id: Int

name: String

in_FlightRoute_FromAirport: [FlightRoute]

in_FlightRoute_ToAirport: [FlightRoute]

out_Airport_BasedIn: [Country]

}

​

type FlightRoute {

id: Int

stops: Int

out_FlightRoute_FromAirport: [Airport]

out_FlightRoute_OperatingAirline: [Airline]

out_FlightRoute_ToAirport: [Airport]

}