Hypermedia APIs that make sense
6 likes2,579 views
The document discusses ways to design APIs and clients to make them more effective. It argues that traditional APIs like SPARQL place too much load on servers and are difficult to scale. A better approach is to have simple servers that provide data fragments and use hypermedia to explain the API, allowing clients to be clever in how they assemble the data. This empowers clients by distributing intelligence rather than concentrating it on servers. Hypermedia APIs allow measuring how clients perform the same tasks across different interfaces.
![{
"@id": "http://fragments.dbpedia.org/2015/en#dataset",
"search": {
"template": "/2015/en{?subject,predicate,object}",
"mapping": [
{ "variable": "subject", "property": "rdf:subject" },
{ "variable": "predicate", "property": "rdf:predicate" },
{ "variable": "object", "property": "rdf:object" }
]
}
}
JSON-LD](https://image.slidesharecdn.com/hypermediaapisthatmakesense-151208100417-lva1-app6891/85/Hypermedia-APIs-that-make-sense-59-320.jpg)
Hypermedia APIs that make sense
- 3. MY API IS GOOD BECAUSE I USE SWAGGER. YOUR API SUCKS BECAUSE IT’S NOT REST. MY API IS DOING HYPERMEDIA, LIKE, FOR REAL.
- 4. MY WEBSITE IS GOOD BECAUSE I USE FLASH. YOUR WEBSITE SUCKS BECAUSE THE MENU IS ON TOP. MY WEBSITE REQUIRES ONLY 3 CLICKS TO ACCESS STUFF.
- 5. API DESIGN IS STUCK WHERE WEB INTERACTION DESIGN WAS 15 YEARS AGO.
- 6. WEB INTERACTION DESIGN IN 2015 IS MEASURABLE: HOW LONG DOES IT TAKE CUSTOMERS TO BUY WHAT THEY WANT?
- 7. WEB INTERACTION DESIGN IN 2015 IS MEASURABLE: HOW OFTEN IS OUR CONTENT SHARED ON TWITTER?
- 8. HOW CAN WE MEASURE WEB API DESIGN DECISIONS?
- 9. HOW CAN WE MEASURE WEB API DESIGN DECISIONS? www.docsduds.com
- 10. PUBLISHING LINKED DATA QUANTIFYING WEB APIS MAKING SENSE WITH HYPERMEDIA 1 2 3
- 11. PUBLISHING LINKED DATA QUANTIFYING WEB APIS MAKING SENSE WITH HYPERMEDIA 1 2 3
- 12. LINKED DATA CAPTURES FACTS WITH LINKS
- 14. { "@id": "dbpedia:Nikola_Tesla", "foaf:knows": "dbpedia:Thomas_Edison", "foaf:givenName": "Nikola" JSON-LD ! ! ! ! }
- 15. WIKIPEDIA IS TRANSLATED INTO 500 MILLION LINKED DATA FACTS.
- 16. IT CAN BE QUERIED BY THE WEB’S MOST POWERFUL API.
- 18. ? ? ? SPARQL
- 19. WHICH FRENCH CITIES HAVE A POPULATION OF MORE THAN 200.000? SPARQL
- 20. WHICH MOVIES WERE DIRECTED BY WOMEN WHO WROTE NOVELS? SPARQL
- 21. SPARQL WEB API
- 22. THE WEB API IS THE SAME AS THE LANGUAGE: ASK ANYTHING
- 23. /SPARQL?QUERY= WHICH FRENCH CITIES HAVE A POPULATION OF MORE THAN 200.000? SPARQL
- 24. /SPARQL?QUERY= WHICH MOVIES WERE DIRECTED BY WOMEN WHO WROTE NOVELS? SPARQL
- 25. IT’S LIKE THE GRAPHQL OF LINKED DATA WITHOUT MONEY
- 26. THE SPARQL API IS SO POWERFUL THAT THAT MOST ENDPOINTS ARE DOWN > 1.5 DAY EACH MONTH
- 27. THE SPARQL API IS SO POWERFUL THAT THAT MOST ENDPOINTS ARE DOWN > 1.5 DAY EACH MONTH
- 28. PUBLISHING LINKED DATA QUANTIFYING WEB APIS MAKING SENSE WITH HYPERMEDIA 1 2 3
- 29. SPARQL WEB API LET’S MEASURE THE
- 30. S! C
- 32. EACH CLIENT SENDS UNIQUE REQUESTS, SO CACHING IS INEFFECTIVE REQUESTS CAN BE ARBITRARILY COMPLICATED THE INTERFACE IS EXPENSIVE
- 33. 1 240 CLIENTS SPARQL API QUERIES PER HOUR 100 200
- 34. SPARQL API 1 240 CLIENTS100 200 SERVER CPU LOAD
- 35. SPARQL API 1 240 CLIENTS100 200 CLIENT CPU LOAD
- 36. SPARQL API 1 240 CLIENTS100 200 CACHE REUSE
- 37. WHO WILL PAY FOR SUCH AN EXPENSIVE API?
- 38. PUBLISHING LINKED DATA QUANTIFYING WEB APIS MAKING SENSE WITH HYPERMEDIA 1 2 3
- 39. WHAT IF WE MADE A BETTER API? GUESS WHAT? THAT’S WHAT EVERYBODY DOES!
- 41. 2005 2010 2015 186 2,418 14,368 NUMBER OF CLIENTS
- 42. HOW CAN WE THEN EXECUTE UNIFORM QUERIES? WE DON’T.
- 43. HOW CAN WE THEN QUERY MULTIPLE APIS? WE CAN’T.
- 44. WHAT IF WE MADE A SIMPLER API THAT EXPLAINED ITSELF WITH HYPERMEDIA?
- 47. “SO IF I WANT TO DO CLIENT ”
- 48. “THEN I NEED TO ASK YOU CLIENT ”
- 49. “AND COMBINE IT MYSELF INTO CLIENT ”
- 50. CLIENTS STILL ANSWER COMPLEX SPARQL QUERIES, JUST NOT WITH THE SPARQL API
- 52. SERVER CLIENT SPARQL API TPF API (nothing)
- 54. 1 240 CLIENTS SPARQL API QUERIES PER HOUR TPF API 100 200
- 55. SPARQL API TPF API 1 240 CLIENTS100 200 SERVER CPU LOAD
- 56. SPARQL API TPF API 1 240 CLIENTS100 200 CLIENT CPU LOAD
- 57. SPARQL API TPF API 1 240 CLIENTS100 200 CACHE REUSE
- 58. RESPONSES CONTAIN HYPERMEDIA CONTROLS IN JSON-LD USING THE HYDRA VOCABULARY
- 59. { "@id": "http://fragments.dbpedia.org/2015/en#dataset", "search": { "template": "/2015/en{?subject,predicate,object}", "mapping": [ { "variable": "subject", "property": "rdf:subject" }, { "variable": "predicate", "property": "rdf:predicate" }, { "variable": "object", "property": "rdf:object" } ] } } JSON-LD
- 60. I CAN DO I CAN DO I CAN DO I CAN DO
- 62. 2005 2010 2015 186 2,418 14,368 NUMBER OF CLIENTS 14,369
- 63. TRY A SIMPLE SERVER AND A CLEVER CLIENT data.linkeddatafragments.org client.linkeddatafragments.org
- 64. PUBLISHING LINKED DATA QUANTIFYING WEB APIS MAKING SENSE WITH HYPERMEDIA 1 2 3
- 65. WHEN DOES A CLIENT TRULY HAVE POWER?
- 66. IF IT SIMPLY USES CLEVER APIS?
- 67. IF IT CLEVERLY USES SIMPLE APIS?
- 68. IF YOU HAVE THE MONEY, USE SPARQL OR GRAPHQL INTELLIGENCE IS EXPENSIVE IF NOT, LET EACH CLIENT USE ITS OWN CPU
- 69. I STOPPED THINKING ABOUT WEB APIS A LONG TIME AGO I’M NOW THINKING ABOUT WEB CLIENTS AND WHEN I SAY WEB, I MEAN HYPERMEDIA
- 70. I CAN’T MEASURE HYPERMEDIA, BUT I CAN MEASURE CLIENTS DOING THE SAME TASKS ON DIFFERENT APIS
- 71. HYPERMEDIA LETS A SERVER EXPLAIN ITS API TO CLIENTS, SO THEY CAN USE IT IN MEASURABLY DIFFERENT WAYS.