Introduction to the isolation level of gbase 8s

         The database provides 5 There are different isolation levels ：Dirty read Dirty reading 、Committed read Submit to read 、Last Committed read Finally submit to read 、Cursor read Cursor read 、Repeatable read Repeatable .

1． Dirty reading

         Use this isolation level , The database server does not check whether records are locked , No locks are assigned .

         In the process of dirty read query , You can query any data , Even those records that have been modified but not yet submitted , therefore Dirty-read The isolation level will produce phantom reads 、 The problem of non repetition and dirty reading . There are only... In the non log database Dirty-read An isolation level .

2． Submit to read

         Committing reads means ensuring that all records read are committed to the database , Can avoid reading dirty data .

         The difference between a committed read and a dirty read is , To ensure that the data is not dirty , You need to check the lock status of the record when reading the record , Try to add... To the record S Lock to determine whether the record is added X lock , So as to determine whether the record is dirty data . When the record is being added X In the lock state , Unable to read the record , If there is no record X lock , Then you can add S lock , But submitting a read does not add S lock , Just try to add S Lock to determine whether there is already X lock . One of the characteristics of submitted reading is ： Try to add S lock , But without S lock .

3． Finally submit to read

         To solve the problem Committed Read Deficiency , When the record is locked , Adopted Committed Read The process needs to wait . Application concurrency is affected , Dirty reads cannot guarantee the expected results . The database supports the last commit read isolation level , Also known as optimistic locking mechanism .

         Grammatically , stay COMMITTED READ That's an increase from LAST COMMITTED Options , It is located in the DIRTY READ And COMMITTED READ A level of isolation between , It can avoid dirty reading problems , Can also be solved COMMITTED READ The problem of concurrency lock conflicts .

        Last committed When the isolation level reads data , No record lock check , Directly read the last submitted version , There is no need to add... To the read record S lock .

         because Last committed The isolation level reads the data of the last submitted version , that , The currently read record has not been modified （ No, X The lock state ）, At this point, you can directly read the record ; If the currently read record is being modified （ Combined with the X The lock state ）, So how to read the last submitted record ？ There is only one record in memory that is being modified , The value before modification exists in the logical log （ Logical log cache or logical log file ）, At this point, you need to read the logical log to get the last submitted version . This is why tables that support the last commit read isolation level must use row level locks , Page level locks are not supported .

4． Cursor read

         For non update cursors , Apply a shared lock to all records read by the cursor , When the next row of records is read , The shared lock is released . For update cursors （For update）, A lift lock will be applied to the read record （U lock ）, Until the cursor closes .

         Retrieving data through a cursor , The shared lock will remain in place until the next FETCH sentence . Therefore, cursor reading can ensure that in the process of cursor program processing , Records are not modified .

5． Repeatable

         With this isolation level , The database adds S lock , Verify that the data can be read . Until the transaction is committed , The lock can be released . Other users can read data , But the same data cannot be modified .

         When a record is read at the repeatable read isolation level , Add... To the record S lock , So you can ensure that before the end of the transaction , Other transactions cannot be modified , In the same transaction, the records read before and after are the same .

         In the process of using repeatable read isolation level query , You need to add... To the read record S lock , It can ensure that the records read are submitted , At the same time, it can ensure that the records read multiple times are consistent , therefore Repeatable read Isolation level can solve the problem of parallel scheduling 3 A question ： Dirty reading 、 Do not repeat reading and phantom reading .