Motivation
The purpose of this document is to outline the design of refactored HTTP (aka REST) API for AsterixDB.
The reason for HTTP API refactoring is threefold:
Currently we have various REST endpoints (/ddl, /query, /update, /aql, /queryService), which do similar things (parsing parameters out of the request parameters, assembling the response) but each endpoint does it in a slightly different manner. All of that should be refactored to use the common base endpoint, which . At the same time old endpoints could be kept for backwards-compatibility if that it really needed.
Current Web interface does the server-side HTML generation and does not use REST calls whatsoever. It would be better to eat our own dog food here and switch to assembling the result on the client side using JavaScript. Especially since we have a potential GSoC project proposal to build such JS-based interface.
- As we are moving towards having more query languages under AsterixDB umbrella (AQL, SQL++, XQuery\JSONiq as a part of VXQuery) it would be nice to design a generic language-agnostic REST API, which later could be reused by VXQuery since it's also lacking proper API as of now.
Tools and existing APIs
The design is inspired by N1QL REST API (http://developer.couchbase.com/documentation/server/4.1/n1ql/n1ql-rest-api/index.html) since it is an example of wel thoguht API in the similar system. I believe we don't need to be 100% compatible, although it would be nice to be to reuse the same clients.
We might also consider use Swagger (http://swagger.io/) to describe the API. This library will allow users to seamlessly generate client SDKs in their favorite language, which is especially usefully since we don't provide drivers for any clients. Here is the complete set of features which Swagger will allow us to do:
- Describe API in developer-friendly way by creating yaml description.
- Validate correctness of the server-side implementation.
- Generate client-side SDK for various languages.
- Generate documentation for API.
REST Endpoints
Proposed API consists of 3 endpoints: Query, Status and Result. The latter two are needed only in a case of asynchronous data delivery, while the former is the main endpoint and serves as an entry point for all client requests.
Query Endpoint (/query)
The following compares various parameters & HTTP headers using in N1QL API and in current AsterixDB API and proposes parameters to be used in new API.
HTTP Request(POST) parameters:
N1QL API Parameter | Value | Old Asterix API Parameter | Proposed API Parameter | Comment |
statement | string | query | statement | Any valid AQL statement (DDL, update\load statement, flowr query), which should be executed |
format | enum | Accept HTTP header | Accept HTTP header | [Optional] Desired format for the query results. Possible values are ADM, JSON, CSV, XML (for VXQuery). (default: ADM) |
signature | boolean | header | signature | [Optional] Defines whether to include a header for the results schema in the response. In current Asterix used to include CSV header into response. (default: false) |
enum | mode | mode | [Optional] Result delivery mode. Possible values are asynchronous, asynchronous-deferred, synchronous. (default: synchronous) | |
boolean | lossless | lossless | [Optional] Defines whether to use lossless-JSON output for JSON-encoded output or keep clean-JSON instead. (default: false) | |
boolean | wrapper-array | wrapper-array | [Optional] Defines whether to wrap ADM-encoded output into array-braces (default: false) | |
boolean | print-expr-tree | expr-tree | [Optional] Defines whether to include an query expression AST into the result (default: false) | |
boolean | print-rewritten-expr-tree | rewritten-expr-tree | [Optional] Defines whether to include a rewritten query expression AST into the result (default: false) | |
boolean | print-logical-plan | logical-plan | [Optional] Defines whether to include a logical plan into the result (default: false) | |
boolean | print-optimized-logical-plan | optimized-plan | [Optional] Defines whether to include a optimized logical plan into the result (default: false) | |
boolean | print-job | hyracks-job | [Optional] Defines whether to include a Hyracks job into the result (default: false) | |
boolean | execute-query | execute-query | [Optional] Defines whether to execute a query (default: true) |
HTTP Response parameters:
Type of the response - JSON document (even if format in request was XML\CSV it will be serialized as one of JSON attributes)
N1QL API Parameter | Value | Old Asterix API Parameter | Proposed API Parameter | Comment |
results | JSONList | HTTP response body | results | One of three possible values depending on result delivery 1) A list of all results returned by the query (sync result delivery). 2) A list of all URIs to QueryStatus endpoint (async result delivery). 3) No value (DDL\update\load) |
status | string | HTTP response status | status | The status of the request; possible values are: Success, Running (async result), Error (parse\optimizer error), Fatal (execution error). |
error | JSONObject | HTTP response body | error | An object containing details of the error |
error.code | int | error-code | error.code | A code that identifies the error. |
error.msg | string | summary | error.msg | A message describing the error in detail. |
stacktrace | A stack trace of the error. | |||
metrics | JSONObject | metrics | An object containing details of the execution metrics. | |
metrics.executionTime | string | metrics.executionTime | The time taken for the execution of the request, i.e. time from when query execution started until the results were returned. (if execute-query=true) | |
metrics.resultCount | unsigned int | metrics.resultCount | The total number of objects in the results (if result delivery is synchronous and execute-query=true) | |
JSONObject | metrics.expr_tree | Serialized query expression tree (if expr-tree=true) | ||
JSONObject | metrics.rewritten_expr_tree | Serialized rewritten query expression tree (if rewritten-expr-tree=true) | ||
JSONObject | metrics.logical_plan | Serialized query logical plan (if logical-plan=true) | ||
JSONObject | metrics.optimized_plan | Serialized query optimized logical plan (if optimized-plan=true) | ||
JSONObject | metrics.hyracks_job | Serialized Hyracks job (if hyracks-job=true) |
Examples:
DDL request
Querycurl -v http://localhost:19002/query -X POST \ -d "statement=create dataverse test;"
Response< HTTP/1.1 200 OK Content-Type: application/json { "status": "success" "metrics": { "executionTime": "1ms" } }
Query which is not executed, but returns logical plan
Querycurl -v http://localhost:19002/query -X POST -H "Accept: application/json" \ -d "statement=for $x in dataset testDS return $x & lossless=true & logical-plan=true & execute-query=false"
Response< HTTP/1.1 200 OK Content-Type: application/json { "status": "success" "metrics": { "executionTime": "5ms", "logical_plan": [ {"operator": "distribute_result", "args": [{"exp": "var_ref", "var": "$$1"}]}, {"operator": "datascan", "output_vars": ["$$0", "$$1"]} } }
- Query which synchronously returns CSV result, but returns logical plan