问题
Lets assume that i need to execute a spring batch job with 2 steps.step 1 is to read data from a postgres table and update values in the same table. step 2 is to read data from another postgres table and update this table. How can i achieve transactions at job level for this scenario?
That is, if the second step fails, then the first step should be rolled back.
回答1:
i'm not sure if there even exists an solution with automatic chained/multi-level transaction handling that works reliable (or does not need a lot resources on database side)
if the second step fails, then the first step should be rolled back
well you could combine both steps into one:
- read from first table A
- use processor to update table A
- use processor to read from table B
- use writer to update table B
the performance will suffer a lot, because the read on table B will be a single read vs the cursor based for table a
i would go with a compensating strategy like this
- (optional) tables in use are temporary tables and not the real "production" tables, makes it easier to work with compensating with decoupling the datastores from the production
- a failed step 1 triggers another step or another job/script
- this step/job/script deletes as necessary (rows or complete table)
回答2:
Have you considered using job-level transactionality?
Job level Transactionality in Spring Batch
It's important to consider the potential volume size you are managing, in order to avoid out-of-the-limit commits or rollbacks.
回答3:
Put a BEGIN statement before Step 1 and a COMMIT statement after Step 2.
来源：https://stackoverflow.com/questions/8817691/reg-transaction-support-for-a-spring-batch-job-at-job-level