What's the best practice to ensure atomic read of a database table using JDBC?

前端 未结 6 1389
情深已故
情深已故 2021-02-19 06:26

I have an java application reading from a database table for jobs to process, and I may have multiple instances of this application running on different servers as each job is i

相关标签:
6条回答
  • 2021-02-19 07:11

    When using databases of a transactional nature, one popular practice is to perform ROW-LEVEL LOCKING. Row-level locks prevent multiple transactions from modifying the same row. SELECT for UPDATE is an easy way to achieve this effect. Assuming you have a processes table:

    SELECT process_id, status 
    from processes
    for UPDATE of status SKIP LOCKED;
    

    When done processing, issue

    update processes set status = 'updated'
    where process_id = :process_id;             --from before
    

    Issue

    commit;
    

    to release the lock.

    Here's an actual example

    Disclaimer: SELECT FOR UPDATE is a form of pessimistic locking and has its caveats as explained by Burleson. However, it might be a viable solution if the client is not web-based and extremely concurrent.

    0 讨论(0)
  • 2021-02-19 07:16

    One approach that would be completely generic*, though perhaps slightly inefficient, would be to use a server-specific identifier to "claim" a job by first updating its status to that identifier, then retrieve the job based on that value. For example, if you were working with Windows servers on the same network then their server name would uniquely identify them. If your table looked like

    JobID  JobName  Status
    -----  -------  ---------
        1  Job_A    Completed
        2  Job_B
        3  Job_C
    

    where unclaimed jobs have a Status of NULL then your application running on SERVER1 could claim a job by doing setAutoCommit(true) followed by

    UPDATE Jobs SET Status='SERVER1'
    WHERE JobID IN (
        SELECT TOP 1 JobID FROM Jobs 
        WHERE Status IS NULL
        ORDER BY JobID)
    

    If ExecuteUpdate returns 0 then there are no jobs pending. If it returns 1 then you can get the row with

    SELECT JobID, ... FROM Jobs WHERE Status='SERVER1'
    

    and then update its Status to 'Running' with a parameterized query like

    UPDATE Jobs SET Status='Running' WHERE JobID=?
    

    where you supply the JobID you retrieved from the previous SELECT.

    *(i.e., not relying on any specific SQL extensions, explicit locking, or transaction handling)

    0 讨论(0)
  • 2021-02-19 07:18

    Use ResultSet that has CONCUR_READ_ONLY and TYPE_FORWARD_ONLY. If your database jdbc driver supports it, it will only return atomic read of your select time.

    According to this documentation, (Table Summary of Visibility of Internal and External Changes)

    forward-only cursor will only show your read time results. CONCUR_READ_ONLY will prevent your internal updates.

    0 讨论(0)
  • 2021-02-19 07:22

    Problem

    Take jobs ready to process and make their status running atomically.

    Solution

    No need for additional locks. Since an update operation is already atomic by itself in terms of the same query (see the excerpt from the docs below), update the jobs table, setting the status running to those that are ready to be processed and get the result of this update - it will be the jobs you took for processing.

    Examples:

    Postgres

    UPDATE jobs SET status = 'running'
      WHERE status is NULL
    RETURNING id;
    

    In terms of JDBC you can go similar to this:

    String sql = "update ... returning ...";
    boolean hasResult = statement.execute(sql);
    if (hasResult) {
        ResultSet rs = statement.getResult();
    }
    

    SQL Server

    UPDATE jobs SET status = 'running'
      WHERE status is NULL
    OUTPUT UPDATED.id;
    

    Excerpt from the Postgres documentation that shows how 2 transactions behave when doing UPDATE on the same table with the same query:

    UPDATE will only find target rows that were committed as of the command start time. However, such a target row might have already been updated (or deleted or locked) by another concurrent transaction by the time it is found. In this case, the would-be updater will wait for the first updating transaction to commit or roll back (if it is still in progress).

    0 讨论(0)
  • 2021-02-19 07:23

    Lock the table using whatever mechanism is supported by your database server.

    For example, in Postgres it would be:

    LOCK yourtable;
    

    And it's your table for the duration of the transaction.

    Other databases will have something similar.

    0 讨论(0)
  • 2021-02-19 07:25

    if you want to ensure proper work in concurrent environment in your specific example you can use the server name.

    The table will look like:

    JobID  JobName  Server  Status
    -----  -------  ------- ---------
        1  Job_A    host-1  Completed
        2  Job_A    host-2  Working
        3  Job_B    host-3  Working
    

    if you have multiple instances on the same host add the process id too:

    JobID  JobName  Server  ProcessID  Status
    -----  -------  ------- ---------- ---------
        1  Job_A    host-1  1000       Completed
        2  Job_A    host-2  1000       Working
        3  Job_A    host-2  1001       Working
        5  Job_B    host-3  1000       Working
    
    0 讨论(0)
提交回复
热议问题