Upgrade Alibaba cloud RDS (relational database service) instance to com mysql. jdbc. exceptions. Troubleshooting of jdbc4.communicationsexception

The phenomenon

Currently, alicloud is used in the marketing database MYSQL database , Need support in the near future DBA To Alibaba cloud RDS upgrade , The upgrade announcement is as follows ：

In the morning 5 spot 10 Start to report errors in the log after upgrading , The error message is as follows :

Problem analysis and conclusion

In general duridCommunicationsException The reason is that the connection pool obtains the invalid connection . from mysql Come up and say :

Mysql Server default “wait_timeout” yes 8 Hours ( That is to say, the default value is 28800 second ), That is, a connection More leisure than 8 Hours ,Mysql Will automatically disconnect the connection, Generally speaking, it is a connection between 8 There was no activity for hours , Will automatically disconnect the connection . And the connection pool thinks that the connection is still valid ( Because the validity of the connection is not verified ), When the app requests to use the connection , Will lead to the above error .

from druid There are several properties on configuration properties and connection failures :

• minEvictableIdleTimeMillis： The connection idle time is greater than this value and the free connection in the pool is greater than minIdle Then close the connection . Default values used in configuration 30 minute .
• maxEvictableIdleTimeMillis： Connection idle time is greater than this value , No matter minIdle Close the connection . Default values used in configuration 7 Hours .
• maxidle： at present druid It's abandoned maxidle attribute , Use maxActive and minIdel Instead of , Reference resources https: //github.com/alibaba/druid/wiki/FAQ, from druid In terms of configuration ,minIdle yes 50,maxActive yes 400( Unreasonable code setting , Adjustable ).

In theory druid By default, the maximum number of connection failures 7 An hour is less than the database 8 Hours , There will be no failed connections at the configuration level . And from the announcement , Have a show :“ In the process of migration, there will be 30 Seconds of connection ”, explain rds After the flash, the original database connection It's not working , and smc In the system druid The connection pool is still valid , So it leads to com.mysql.jdbc.exceptions.jdbc4. CommunicationsException happen .

Solution

Direct thinking in druid The thread gets the connection and checks it ok 了 , Related attributes are test- WhileIdle、testOnBorrow, The meaning is as follows :

• testWhileIdle： If true( Default true), When an application applies to the connection pool for a connection , also testOnBorrow by false when , The connection pool will determine if the connection is idle , If it is , Verify that the connection is available ( Efficient , Recommended ).
• testOnBorrow：( Default false) Check the connection validity when applying for connection ( Inefficient , Affect performance ), And testWhileIdle It's almost the same , It is to test the validity of the connection when obtaining the connection , If both are true, be testOnBorrow first High class , You will not use testWhileIdle.
• validationQuery： The query used to verify the connection before the connection pool returns the connection to the caller SQL, A query statement is required ( Turn on testWhileIdle You have to set validationQuery).

druid Add... To the configuration :

spring.datasource.druid.test-on-borrow=false
spring.datasource.druid.test-while-idle=true
spring.datasource.druid.validation-query=SELECT 1

There's another question , Without configuring the above properties ,testWhileIdlel The default is true,testOnBorrow Default by false,validationQuery The default is empty. , Why is it not mandatory validationQuery Well ? At this time, we should divide 2 There are three stages to explain the problem :

• durid Initialization phase ： Played only error, The initialization process is not interrupted
• Get the connection process ： Check the minEvictableIdleTimeMillis and maxEvictableIdleTimeMillis, Find out validation- Query by null Straight back , So the above configuration only needs to add validationQuery The configuration of can also solve the problem .

Welcome to WeChat official account. ： Fang Chen's blog