Recovery System

Outline
 Failure Classification
 Storage Structure
 Recovery and Atomicity
 Log-Based Recovery
 Remote Backup Systems
Failure Classification
 Transaction failure :
 Logical errors: transaction cannot complete due to some internal error
condition
 System errors: the database system must terminate an active transaction
due to an error condition (e.g., deadlock)
 System crash: a power failure or other hardware or software failure
causes the system to crash.
 Fail-stop assumption: non-volatile storage contents are assumed to not
be corrupted by system crash
 Database systems have numerous integrity checks to prevent corruption of disk
data

 Disk failure: a head crash or similar disk failure destroys all or part
of disk storage
 Destruction is assumed to be detectable: disk drives use checksums to
detect failures
Recovery and Atomicity
 To ensure atomicity despite failures, we first output information
describing the modifications to stable storage without modifying the
database itself.
 We study log-based recovery mechanisms in detail
 We first present key concepts
 And then present the actual recovery algorithm
 Less used alternative: shadow-copy and shadow-paging (brief details
in book)

shadow-copy
Log-Based Recovery
 A log is a sequence of log records. The records keep information
about update activities on the database.
 The log is kept on stable storage
 When transaction Ti starts, it registers itself by writing a
<Ti start> log record

 Before Ti executes write(X), a log record

<Ti, X, V1, V2>
is written, where V1 is the value of X before the write (the old
value), and V2 is the value to be written to X (the new value).
 When Ti finishes it last statement, the log record <Ti commit> is
written.
 Two approaches using logs
 Immediate database modification
 Deferred database modification.
Immediate Database Modification
 The immediate-modification scheme allows updates of an
uncommitted transaction to be made to the buffer, or the disk
itself, before the transaction commits
 Update log record must be written before database item is written
 We assume that the log record is output directly to stable storage
 (Will see later that how to postpone log record output to some extent)
 Output of updated blocks to disk can take place at any time before
or after transaction commit
 Order in which blocks are output can be different from the order in
which they are written.
 The deferred-modification scheme performs updates to
buffer/disk only at the time of transaction commit
 Simplifies some aspects of recovery
 But has overhead of storing local copy
Transaction Commit
 A transaction is said to have committed when its commit log record is
output to stable storage
 All previous log records of the transaction must have been output already
 Writes performed by a transaction may still be in the buffer when the
transaction commits, and may be output later
Immediate Database Modification Example
Log Write Output

<T0 start>
<T0, A, 1000, 950>
<T0, B, 2000, 2050>
A = 950
B = 2050
<T0 commit>
<T1 start>
<T1, C, 700, 600>
BC output before T1
C = 600 commits
B B , BC
<T1 commit>
BA
BA output after T0
 Note: BX denotes block containing X. commits
Concurrency Control and Recovery
 With concurrent transactions, all transactions share a single disk
buffer and a single log
 A buffer block can have data items updated by one or more transactions
 We assume that if a transaction Ti has modified an item, no other
transaction can modify the same item until Ti has committed or
aborted
 i.e., the updates of uncommitted transactions should not be visible to
other transactions
 Otherwise, how to perform undo if T1 updates A, then T2 updates A and commits,
and finally T1 has to abort?

 Can be ensured by obtaining exclusive locks on updated items and holding

the locks till end of transaction (strict two-phase locking)
 Log records of different transactions may be interspersed in the log.
Undo and Redo Operations
 Undo and Redo of Transactions
 undo(Ti) -- restores the value of all data items updated by Ti to their old
values, going backwards from the last log record for Ti
 Each time a data item X is restored to its old value V a special log record <Ti ,
X, V> is written out
 When undo of a transaction is complete, a log record
<Ti abort> is written out.

 redo(Ti) -- sets the value of all data items updated by Ti to the new
values, going forward from the first log record for Ti
 No logging is done in this case
Recovering from Failure
 When recovering after failure:
 Transaction Ti needs to be undone if the log
 Contains the record <Ti start>,
 But does not contain either the record <Ti commit> or <Ti abort>.

 Transaction Ti needs to be redone if the log

 Contains the records <Ti start>
 And contains the record <Ti commit> or <Ti abort>
 Suppose that transaction Ti was undone earlier and the <Ti abort>
record was written to the log, and then a failure occurs,
 On recovery from failure transaction Ti is redone
 Such a redo redoes all the original actions of transaction Ti including the
steps that restored old values
 Known as repeating history
 Seems wasteful, but simplifies recovery greatly
Checkpoints
 Redoing/undoing all transactions recorded in the log can be very slow
 Processing the entire log is time-consuming if the system has run for a
long time
 We might unnecessarily redo transactions which have already output
their updates to the database.
 Streamline recovery procedure by periodically performing
checkpointing
1. Output all log records currently residing in main memory onto stable
storage.
2. Output all modified buffer blocks to the disk.

3. Write a log record < checkpoint L> onto stable storage where L is a
list of all transactions active at the time of checkpoint.
4. All updates are stopped while doing checkpointing
Checkpoints (Cont.)
 During recovery we need to consider only the most recent
transaction Ti that started before the checkpoint, and transactions
that started after Ti.
 Scan backwards from end of log to find the most recent <checkpoint L>
record
 Only transactions that are in L or started after the checkpoint need to
be redone or undone
 Transactions that committed or aborted before the checkpoint already
have all their updates output to stable storage.
 Some earlier part of the log may be needed for undo operations
 Continue scanning backwards till a record <Ti start> is found for every
transaction Ti in L.
 Parts of log prior to earliest <Ti start> record above are not needed for
recovery, and can be erased whenever desired.
Example of Checkpoints

 T1 can be ignored (updates already output to disk due to checkpoint)

 T2 and T3 redone.

 T4 undone

 T1 can be ignored (updates already output to disk due to checkpoint)

 T2 and T3 redone.

 T4 undone
Shadow Paging
 Shadow paging is an alternative to log-based recovery; this scheme
is useful if transactions execute serially
 Idea: maintain two page tables during the lifetime of a transaction –
the current page table, and the shadow page table
 Store the shadow page table in nonvolatile storage, such that state
of the database prior to transaction execution may be recovered.
 Shadow page table is never modified during execution
 To start with, both the page tables are identical. Only current page
table is used for data item accesses during execution of the
transaction.
 Whenever any page is about to be written for the first time
 A copy of this page is made onto an unused page.
 The current page table is then made to point to the copy
 The update is performed on the copy
Sample Page Table
Example of Shadow Paging
Shadow and current page tables after write to page 4
Shadow Paging (Cont.)
 To commit a transaction :
1. Flush all modified pages in main memory to disk
2. Output current page table to disk
3. Make the current page table the new shadow page table, as
follows:
 keep a pointer to the shadow page table at a fixed (known) location on
disk.
 to make the current page table the new shadow page table, simply
update the pointer to point to current page table on disk
 Once pointer to shadow page table has been written, transaction is
committed.
 No recovery is needed after a crash — new transactions can start
right away, using the shadow page table.
 Pages not pointed to from current/shadow page table should be
freed (garbage collected).
Show Paging (Cont.)
 Advantages of shadow-paging over log-based schemes
 no overhead of writing log records
 recovery is trivial
 Disadvantages:
 Copying the entire page table is very expensive
 Can be reduced by using a page table structured like a B+-tree
 No need to copy entire tree, only need to copy paths in the tree that lead to updated
leaf nodes

 Commit overhead is high even with above extension

 Need to flush every updated page, and page table
 Data gets fragmented (related pages get separated on disk)
 After every transaction completion, the database pages containing old
versions of modified data need to be garbage collected
 Hard to extend algorithm to allow transactions to run concurrently
 Easier to extend log based schemes
Recovering from Failure

 T1 can be ignored (updates already output to disk due to checkpoint)

 T2 and T3 redone.
 T4 undone