About the performance of using JDBC and HTTP to request Druids

This is the case. At present, the company always uses HTTP to request Druids, but the problem is also very obvious. HTTP makes a query cumbersome. Can we make the Druid request as simple as JDBC. Therefore, I learned that Druid was requested by JDBC, and everything was very smooth by integrating mybatis plus. Just when I wanted to finish work, the problem occurred。By means of jemter pressure measurement，I found that the speed of HTTP query for the same SQL is 10 times faster than that of JDBC. The pressure test configuration is 10 threads,After checking the relevant information, I still can't get the answer, so please help me to see what the situation is

HTTP request 1 is jdbc test
HTTP request2 is http test


```java
@GetMapping("testList1")
public String list1() {
StopWatch stopWatch = new StopWatch();
stopWatch.start();
List appExceptionDataMinList = druidMapper.getApmService();
stopWatch.stop();
System.out.println("stopWatch.getTotalTimeSeconds() = " + stopWatch.getTotalTimeSeconds());
for (ApmServiceMetric apmServiceMetric : appExceptionDataMinList) {
System.out.println(apmServiceMetric);
}
return "success";
}`

@Options(statementType = StatementType.STATEMENT)
List getApmService();
```
```html

select
(CURRENT_TIMESTAMP - INTERVAL '5' MINUTE) as "datetime",
TIMESTAMP_TO_MILLIS(CURRENT_TIMESTAMP - INTERVAL '5' MINUTE) as "timestamp",
biz_system_id as "bizSystemId",
application_id as "applicationId",
sum("count") / 5.0 as "APMServiceCPM",
sum("count") as "APMServiceCallCount",
sum("count") - sum("success_count") as "APMServiceFailCount",
(sum("resp_time") * 1.0) / sum("success_count") as "APMServiceAvgResponseTime"
from SVR_ACTION_DATA_MIN
where
"data_type" = 'APP'
and "__time" > CURRENT_TIMESTAMP - INTERVAL '10' MINUTE and "__time" CURRENT_TIMESTAMP - INTERVAL '5' MINUTE
group by
biz_system_id, application_id

```

Use HTTP to request Druid
`public String list3() {
StopWatch stopWatch = new StopWatch();

stopWatch.start("组装参数");
HashMap paramMap = new HashMap<>();
paramMap.put("username", "admin");
paramMap.put("password", "nEtben@2_19");
paramMap.put("timeout", 6000);
paramMap.put("query", "select\n" +
" (CURRENT_TIMESTAMP - INTERVAL '5' MINUTE) as \"datetime\",\n" +
" TIMESTAMP_TO_MILLIS(CURRENT_TIMESTAMP - INTERVAL '5' MINUTE) as \"timestamp\",\n" +
" biz_system_id as \"bizSystemId\",\n" +
" application_id as \"applicationId\",\n" +
" sum(\"count\") / 5.0 as \"APMServiceCPM\",\n" +
" sum(\"count\") as \"APMServiceCallCount\",\n" +
" sum(\"count\") - sum(\"success_count\") as \"APMServiceFailCount\",\n" +
" (sum(\"resp_time\") * 1.0) / sum(\"success_count\") as \"APMServiceAvgResponseTime\"\n" +
" from SVR_ACTION_DATA_MIN\n" +
" where\n" +
" \"data_type\" = 'APP'\n" +
" and \"__time\" > CURRENT_TIMESTAMP - INTERVAL '10' MINUTE and \"__time\" <= CURRENT_TIMESTAMP - INTERVAL '5' MINUTE\n" +
" group by\n" +
" biz_system_id, application_id");

stopWatch.stop();

stopWatch.start("发起请求");
String body = HttpRequest.post("http://10.128.2.90:8082/druid/v2/sql")
.header("Content-Type", "application/json")
.body(JSONUtil.toJsonStr(paramMap))
.execute()
.body();
stopWatch.stop();

stopWatch.start("结果处理");
List apmServiceMetricList = JSONUtil.toList(body, ApmServiceMetric.class);
stopWatch.stop();
System.out.println("stopWatch.prettyPrint() = " + stopWatch.prettyPrint());

return "success";
}`

Use pure JDBC without using ORM framework
`public String test12() {
StopWatch stopWatch = new StopWatch();
stopWatch.start("建立链接,准备PreparedStatement");
String url = "jdbc:avatica:remote:url=http://10.128.2.90:8082/druid/v2/sql/avatica/";

Properties connectionProperties = new Properties();

Connection connection = DriverManager.getConnection(url, connectionProperties);
PreparedStatement statement = connection.prepareStatement(
"SELECT\n" +
" (CURRENT_TIMESTAMP - INTERVAL '5' MINUTE) as \"datetime\",\n" +
" TIMESTAMP_TO_MILLIS(CURRENT_TIMESTAMP - INTERVAL '5' MINUTE) as \"timestamp\",\n" +
" biz_system_id as \"biz_system_id\",\n" +
" application_id as \"application_id\",\n" +
" component_name as \"component_name\",\n" +
" op_name as \"op_name\",\n" +
" component_subtype as \"component_subtype\",\n" +
" sum(\"count\") / 5.0 as \"APMServiceCPM\",\n" +
" sum(\"count\") as \"apmservice_call_count\",\n" +
" sum(\"exception_count\") as \"apmservice_error_count\",\n" +
" (sum(\"resp_time\") * 1.0) / sum(\"success_count\") as \"apmservice_avg_response_time\"\n" +
" FROM \"SVR_COMPONENT_DATA_MIN\"\n" +
" WHERE\n" +
" \"data_type\" = 'OP' and \"component_type\" = 'External'\n" +
" and \"__time\" > CURRENT_TIMESTAMP - INTERVAL '10' MINUTE and \"__time\" <= CURRENT_TIMESTAMP - INTERVAL '5' MINUTE\n" +
" GROUP BY biz_system_id, application_id, component_name, op_name, component_subtype");

stopWatch.stop();

stopWatch.start("执行 sql");
ResultSet resultSet = statement.executeQuery();
stopWatch.stop();
stopWatch.start("关闭链接");

connection.close();
resultSet.close();
System.out.println("resultSet = " + resultSet);
stopWatch.stop();
System.out.println(stopWatch.prettyPrint());
return "success";
}`

result

Start with the two provided Java examples and reproduce both paths under the stated 10-thread JMeter setup, using the JDBC URL and /druid/v2/sql endpoint shown. Compare the connection, query execution, and result-processing timings; done means identifying the source of the JDBC/HTTP difference and recording a supported explanation or fix.

Written by the indexing model from the issue text.