Recovery and Logging

Question 1

Caching

Answer 1

• Recently accessed database pages are cached in memory to reduce disk I/O

Question 2

Recovery

Answer 2

• Performed after a crash or non-catastrophic failure
• Need to maintain enough state about transactions to redo or undo them

Question 3

Recovery Steps

Answer 3

• Redo changes made by any committed transaction
• Undo changes made by any transaction that didn’t commit

Question 4

Log

Answer 4

• Update records summarizing writes
• Records for transaction begins
• Records for transaction commits
• Append-only

Question 5

Write-Ahead Logging (WAL)

Answer 5

• Both updated database pages and log records are cached
• All update log records must be forced to disk before the database page

Question 6

Undo-Redo Logging

Answer 6

• Update records must include both the old and new values
• At commit, write the commit log record to the in-memory log buffer and force to disk all dirty log records

Question 7

Undo-Redo Backward Pass

Answer 7

• For each commit record, add the transaction to a commit list
• For each update by a transaction not on the commit list, undo the update
• Skip updates on the commit list and all begin records

Question 8

Undo-Redo Forward Pass

Answer 8

• For each update by a transaction on the commit list, redo the update
• Skip updates by transactions not on the commit list

Question 9

Logical Logging

Answer 9

• When a data element is updated, store the LSN of the updated log record with the data element (datum LSN)
• Stores the old LSN of the data element in the log record

Question 10

Logical Logging Backward Pass

Answer 10

• Undo an update if the transaction did not commit and datum LSN == record LSN
• If undo, set datum LSN = olsn

Question 11

Logical Logging Forward Pass

Answer 11

• Redo an update if the transaction did commit and datum LSN == olsn
• If redo, set datum LSN = record LSN

Question 12

Undo-Only Logging

Answer 12

• Force all dirty log records to disk
• Force database pages changed by the transaction to disk
• Write the commit log record
• Force the commit log record to the disk

Question 13

Redo-Only Logging

Answer 13

• Write the commit log record
• Force all dirty log records to disk
• Move database pages to disk anytime after this

Question 14

Advantages and Disadvantages of Undo-Only Logging

Answer 14

• Smaller logs than undo-redo (+)
• Simple and quick recovery (+)
• Forces log and data to disk at commit (wait for I/O) (-)

Question 15

Advantages and Disadvantages of Redo-Only Logging

Answer 15

• Smaller logs than undo-redo (+)
• More complex recovery than undo-only, but less than undo-redo (+/-)
• Must be able to cache all changes until the transaction commits (-)
• Forces only log records to disk at commit (+)

Question 16

Advantages and Disadvantages of Undo-Redo Logging

Answer 16

• Larger logs (-)
• Complex recovery (-)
• Forces only log records to disk at commit (+)
• Don’t need to retain all data in cache until commit (+)

Question 17

Checkpoint

Answer 17

• Force data and log records to disk to create a consistent on-disk database state
• Avoid long recoveries

Question 18

Static Checkpoint

Answer 18

1. Stop the activity and wait for a consistent state
2. Force dirty log records to disk
3. Force dirty database pages to disk
4. Write a checkpoint record to the log

Question 19

Dynamic Checkpoint

Answer 19

1. Prevent all update operations during the checkpoint
2. Force dirty log records to disk
3. Force dirty database pages to disk
4. Write a checkpoint record to the log