Question

I have some problem in system I am developing. I have one python script which first works with a virtulisation software and if that operation succeeds, it writes things to database.
If some exception occurs in the virtulisation software then I can manage all things, but the real problem will occur if inserting in database fails. If insert fails , i will have to revert things in that virtulization software otherwise things will become asynchronous. But problem is, reverting things in that software is not possible all the time.

How to handle things so that i can keep the database in sync with that software? Any middle ware or special application??? Or any logic in programming?

Answer 1

You want two actions in your system (OP: operation in your virt. software; WDB: write to database) to be atomic (either both take place, or none). Kind of a distributed transaction, but your virtualized software does not directly support directly a transactionable behaviour (no rollback). If you could make them part of some distributed transactional system, you’d be done (see eg), but that is often impossible or impractical. Different strategies to attain a pseudo-transactional behaviour depends on the particulars of your scenario. Some examples:

---

1. Open TX (transaction in DB)
2. WDB
3. OP
4. If OP succeeded, commit TX, else rollback TX.

Only feasible if what you write to DB does not depend on OP operation (improbable).

---

1. OP1 (first phase of the operation: you get the results, but do not alter anything)
2. Open TX
3. WDB
4. OP2 (second phase: you modify the virt. sofware)
5. Commit TX or rollback

(Steps 4-5 can be switched) This would be a poor’s man “Two-phase commit” implementation. Only feasible if you can divide your operation in those two phases.

---

1. Open TX
2. Dummy WDB (write a dummy result to DB)
3. Rollback TX
4. OP
5. WDB

This checks that the DB is operational, doing a dummy writing before attempting the real operation and writing. Feasible, but not foolproof.

---

1. OP
2. WDB
3. If fail: save data to a raw file, log error, send mail to IT, turn red lights on.

Sound pathetic… but sometimes it’s the only feasible way.