see:
- https://github.com/Netflix/dgs-examples-java
- https://github.com/Netflix/dgs-examples-kotlin
- https://piotrminkowski.com/2021/04/08/an-advanced-graphql-with-spring-boot-and-netflix-dgs/
- https://github.com/hantsy/spring-graphql-sample
- https://www.graphql-java.com/blog/threads/
- https://netflix.github.io/dgs/advanced/instrumentation/#graphql-query-complexity
http://localhost:8080/graphiql
- transaction management: it creates a new transaction for each query instead of joining the current transaction
- queries are slow (e.g.: p95 386ms at concurrency>10)
- dataloaders execute code as async futures in common fork join pool. you may want to execute blocking calls to db within a dedicated "IO" thread pool instead see: https://www.graphql-java.com/blog/threads/
$ make db-local.up
# verbose
$ ab -v 4 -n 1 -c 1 -T application/json -p my_data.json http://localhost:8080/graphql
# quiet
$ ab -n 1 -c 1 -T application/json -p my_data.json http://localhost:8080/graphql
file: my_data.json
{"query":"query {\n \n shows(titleFilter:\"foo\") {\n __typename\n showId\n title\n releaseYear\n reviews {\n __typename\n reviewId\n starScore\n comment\n submittedDate\n }\n }\n \n}\n\n\n","variables":null}
scenario: load all shows and their reviews
datasource: (reviews: 460 records, shows: 116 records)
api requests: $ ab -n 1000, -c 30
metrics
- gql 2 db queries, 2 db transactions: 1600 ms (p95)
- rest 1 db query, 1 db transaction: 700 ms (p95)
- rest 2 db queries, 1 db transaction: 800 ms
- rest 2 db queries, 2 db transactions: 950 ms
conclusion
the rest api is way faster under load
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent