Recovery

Question 1

What is a roll back?

Answer 1

Any writes to disk undone, happens on error.

Question 2

Possible types of failure

Answer 2

• Logical error: Some form of internal error.
• System error: Deadlock, bad scheduler etc.
• System Crash: Power Failure, HW malfunction etc.
• Disk failure

Question 3

How does atomicity affect recovery?

Answer 3

Modifying the database without ensuring that the transaction will commit may leave the database in an inconsistent state.

Question 4

How do we ensure atomicity despite failures?

Answer 4

We output information describing the modifications to stable storage before modifying the database.

Question 5

What two approaches can be used to store modification to the database in memory?

Answer 5

• Log-based recovery

- Shadow paging

Question 6

How does log based recovery work?

Question 7

What two approaches can be used using logs?

Answer 7

• Deferred database modification

- Immediate database modification

Question 8

What is deferred database modification?

Answer 8

• All modifications recorded in log, but writes deferred until after partial commit
• Write operation doesn’t include old value.

Question 9

How does deferred DB modification recover from a crash?

Answer 9

• Transaction only needs to be done if both Ti start> and are in the log.
• Redo Ti sets value of all data items updated by the transaction to new values.

Question 10

What is immediate database modification?

Answer 10

• Allows updates of uncommitted transactions

- Log record wrote to before DB

Question 11

How does immediate DB modification recover from failure?

Answer 11

• Redo Ti sets the value of all data items to their new values, going forward from log Ti
• Undo restores the value of all data update by Ti to their old values, going backward from the last log record for Ti

Question 12

What happens when recovering from failure using immediate database modification?

Answer 12

Ti needs redone if start and commit in log.

Ti needs undone if start in log with no commit.

Redo done first, undo list populated and done after.

Question 13

How do checkpoints work?

Answer 13

Periodic checkpoints flush records to disk and timestamps. Completed transactions are then removed from the log.

Question 14

How are checkpoints used during recovery?

Answer 14

Only consider log records that hadn’t been completed by time of last checkpoint.

Question 15

Why is log record buffering used?

Answer 15

Output to stable storage in blocks using log force/buffer is full. Reduces I/O cost.

Question 16

What is Shadow Paging?

Answer 16

Maintains two page tables during transaction lifetime:

• Current page table
  Used for data accesses during execution
• Shadow page table, persisted to storage and not modified

Question 17

What does shadow paging do in execution?

Answer 17

Copy page onto unused page, current page table then points to the copy and update copy.

On commit, flush all modified pages to memory and make current page table new shadow page table.

Question 18

How is the current page table made the new shadow page table?

Answer 18

Point pointer at fixed disk location to new shadow page table. Transaction is now committed.

Question 19

Benefit of shadow paging?

Answer 19

No recovery is needed after crash

Question 20

Drawback of shadow paging?

Answer 20

Doesn’t work well with concurrent transactions.

Has to be garbage collected.