Oracle Database 11g: Administration Workshop I 9 - 2

Locks
Before the database allows a session to modify data, the session must first lock the data that is
being modified. A lock gives the session exclusive control over the data so that no other
transaction can modify the locked data until the lock is released.
Transactions can lock individual rows of data, multiple rows, or even entire tables. Oracle
Database supports both manual and automatic locking. Automatically acquired locks always
choose the lowest possible level of locking to minimize potential conflicts with other
transactions.
Note: There are many types of locks used by the Oracle instance to maintain internal
consistency. In this course, we will be concentrating only locking used to protect rows and
tables.

Oracle Database 11g: Administration Workshop I 9 - 3

Locking Mechanism
The locking mechanism is designed to provide the maximum possible degree of data
concurrency within the database. Transactions that modify data acquire row-level locks rather
than block-level or table-level locks. Modifications to objects (such as table moves) obtain
object-level locks rather than whole database or schema locks.
Data queries do not require a lock, and a query succeeds even if someone has locked the data
(always showing the original, prelock value reconstructed from undo information).
When multiple transactions need to lock the same resource, the first transaction to request the
lock obtains it. Other transactions wait until the first transaction completes. The queue
mechanism is automatic and requires no administrator interaction.
All locks are released as transactions are completed (that is, when a COMMIT or ROLLBACK is
issued). In the case of a failed transaction, the same background process that automatically rolls
back any changes from the failed transaction releases all locks held by that transaction.

Oracle Database 11g: Administration Workshop I 9 - 4

Data Concurrency
The lock mechanism defaults to a fine-grained, row-level locking mode. Different transactions
can be updating different rows within the same table without interfering with one another.
Although the default model is to lock at the row level, Oracle Database supports manual locking
at higher levels if needed:
SQL> LOCK TABLE employees IN EXCLUSIVE MODE;
Table(s) Locked.
With the preceding statement, any other transaction that tries to update a row in the locked table
must wait until the transaction that issued the lock request completes. EXCLUSIVE is the
strictest lock mode. The following are the other lock modes:
• ROW SHARE: Permits concurrent access to the locked table but prohibits sessions from
locking the entire table for exclusive access
• ROW EXCLUSIVE: Is the same as ROW SHARE, but also prohibits locking in SHARE
mode. The ROW EXCLUSIVE locks are automatically obtained when updating, inserting,
or deleting data. ROW EXCLUSIVE locks allow multiple readers and one writer.
• SHARE: Permits concurrent queries but prohibits updates to the locked table. A SHARE
lock is required (and automatically requested) to create an index on a table. However,
online index creation requires a ROW SHARE lock that is used when building the index.

Oracle Database 11g: Administration Workshop I 9 - 5

Data Concurrency (continued)
Share locks allow multiple readers and no writers. Share locks are also used transparently
when deleting or updating rows in a parent table that has a child table with foreign key
constraints on the parent.
• SHARE ROW EXCLUSIVE: Is used to query a whole table and to allow others to query
rows in the table, but prohibits others from locking the table in SHARE mode or updating
rows
• EXCLUSIVE: Permits queries on the locked table but prohibits any other activity on it. An
EXCLUSIVE lock is required to drop a table.
Like any request for a lock, manual lock statements wait until all sessions that either already
have locks or have previously requested locks release their locks. The LOCK command accepts a
special argument that controls the waiting behavior NOWAIT.
NOWAIT returns control to you immediately if the specified table is already locked by another
session:
SQL> LOCK TABLE [Link] IN SHARE MODE NOWAIT;
LOCK TABLE [Link] IN SHARE MODE NOWAIT
*
ERROR at line 1:
ORA-00054: resource busy and acquire with NOWAIT specified
It is usually not necessary to manually lock objects. The automatic locking mechanism provides
the data concurrency needed for most applications. Oracle recommends that you avoid using
manual locks, especially when developing applications. Severe performance issues often occur
from unnecessarily high locking levels.

Oracle Database 11g: Administration Workshop I 9 - 6

DML Locks
Each DML transaction obtains two locks:
• An EXCLUSIVE row lock on the row or rows being updated
• A table lock (TM) in ROW EXCLUSIVE (RX) mode on the table being updated. This
prevents another session from locking the whole table (possibly to drop or truncate it)
while the change is being made. This mode is also called a subexclusive table lock (SX).
A ROW EXCLUSIVE lock on the table prevents DDL commands from changing the dictionary
metadata in the middle of an uncommited transaction. This preserves dictionary integrity and
read consistency across the life of a transaction.

Oracle Database 11g: Administration Workshop I 9 - 7

Enqueue Mechanism
Requests for locks are automatically queued. As soon as the transaction holding a lock is
completed, the next session in line receives the lock.
The enqueue mechanism tracks the order in which locks are requested and the requested lock
mode.
Sessions that already hold a lock can request that the lock be converted without having to go to
the end of the queue. For example, suppose a session holds a SHARE lock on a table. The
session can request that the SHARE lock be converted to an EXCLUSIVE lock. If no other
transaction already has an EXCLUSIVE or SHARE lock on the table, the session holding the
SHARE lock is granted an EXCLUSIVE lock without having to wait in the queue again.
Note: There are two categories of waiters for enqueues: those waiting without shared ownership
and those with shared ownership that do not choose to escalate the lock level. Waiters in the
second category are known as converters, which are always given priority over normal waiters
even if they have been waiting less time.

Oracle Database 11g: Administration Workshop I 9 - 8

Lock Conflicts
Lock conflicts occur often but are usually resolved through time and the enqueue mechanism. In
certain rare cases, a lock conflict may require administrator intervention. In the case in the slide,
transaction 2 obtains a lock on a single row at [Link] and neglects to commit, leaving the lock
in place. Transaction 1 attempts to update the entire table, requiring a lock on all rows, at
[Link]. Transaction 1 is blocked by transaction 2 until transaction 2 commits at [Link].
A user attempting to perform transaction 1 would almost certainly contact the administrator for
help in this case, and the DBA would have to detect and resolve the conflict.

Oracle Database 11g: Administration Workshop I 9 - 9

Possible Causes of Lock Conflicts
The most common cause of lock conflicts is an uncommitted change, but there are a few other
possible causes:
• Long-running transactions: Many applications use batch processing to perform bulk
updates. These batch jobs are usually scheduled for times of low or no user activity, but in
some cases, they may not have finished or may take too long to run during the low activity
period. Lock conflicts are common when transaction and batch processing are being
performed simultaneously.
• Unnecessarily high locking levels: Not all databases support row-level locking (Oracle
added support for row-level locks in 1988 with release 6). Some databases still lock at the
page or table level. Developers writing applications that are intended to run on many
different databases often write their applications with artificially high locking levels so that
Oracle Database behaves similarly to these less capable database systems. Developers who
are new to Oracle also sometimes unnecessarily code in higher locking levels than are
required by Oracle Database.

Oracle Database 11g: Administration Workshop I 9 - 10

Detecting Lock Conflicts
Use the Blocking Sessions page in Enterprise Manager to locate lock conflicts. Conflicting lock
requests are shown in hierarchical layout, showing the session holding the lock at the top and,
below that, any sessions that are enqueued for the lock.
For each session involved in the conflict, you are given the username, session ID, and number of
seconds for which the session has been waiting. Drill down to the session ID to see actual SQL
statements that are currently being executed or requested by the session.
The Automatic Database Diagnostic Monitor (ADDM) also automatically detects lock conflicts
and can advise you on inefficient locking trends.

Oracle Database 11g: Administration Workshop I 9 - 11

Resolving Lock Conflicts
To resolve a lock conflict, the session holding the lock must release it. The best way to have the
session release the lock is to contact the user and ask that the transaction be completed.
In an emergency, it is possible for the administrator to terminate the session holding the lock by
clicking the Kill Session button. Remember that when a session is killed, all work within the
current transaction is lost (rolled back). A user whose session is killed must log in again and redo
all work since the killed session’s last commit.
Users whose sessions have been killed receive the following error the next time they try to issue
a SQL statement:
ORA-03135: connection lost contact
Note: The PMON session sniper can automatically kill sessions due to idle timeout, and this can
be done by using profiles or Resource Manager.

Oracle Database 11g: Administration Workshop I 9 - 12

Resolving Lock Conflicts Using SQL
Session manipulation, like most other tasks in Enterprise Manager, can also be done by issuing
SQL statements. The V$SESSION table contains details of all connected sessions. The value in
BLOCKING_SESSION is the session ID of the session that is blocking. If you query for SID
and SERIAL# (where SID matches a blocking session ID), you have the information needed to
perform the kill session operation.
Note: Database Resource Manager can be used to automatically log out sessions that block
others and are idle.

Oracle Database 11g: Administration Workshop I 9 - 13

Deadlocks
A deadlock is a special example of a lock conflict. Deadlocks arise when two or more sessions
wait for data that has been locked by the other. Because each is waiting for the other, neither can
complete their transaction to resolve the conflict.
Oracle Database automatically detects deadlocks and terminates the statement with an error. The
proper response to that error is either commit or rollback, which releases any other locks in that
session so that the other session can continue its transaction.
In the example in the slide, transaction 1 must either commit or roll back in response to the
deadlock detected error. If it commits, it must resubmit the second update to complete its
transaction. If it performs a rollback, it must resubmit both statements to complete its
transaction.

Oracle Database 11g: Administration Workshop I 9 - 14

Answer: 1

Oracle Database 11g: Administration Workshop I 9 - 15

Answer: 2

Oracle Database 11g: Administration Workshop I 9 - 16

Oracle Database 11g: Administration Workshop I 9 - 17
Oracle Database 11g: Administration Workshop I 9 - 18