Skip to content

G
Gpdb

Greenplum / Gpdb

通知 7
Star 1
Fork 0
G
Gpdb

体验新版 GitCode，发现更多精彩内容 >>

提交 606db06f 编写于 作者: T Tom Lane
浏览文件
操作

Improve manual's discussion of locking and MVCC.

上级 3dd13ffd
展开全部 隐藏空白更改
内联 并排
Showing with 497 addition and 299 deletion
doc/src/sgml/mvcc.sgml
浏览文件 @ 606db06f
此差异已折叠。
doc/src/sgml/ref/lock.sgml
浏览文件 @ 606db06f
<!--
$Header: /cvsroot/pgsql/doc/src/sgml/ref/lock.sgml,v 1.32 2002/04/23 02:07:16 tgl Exp $
$Header: /cvsroot/pgsql/doc/src/sgml/ref/lock.sgml,v 1.33 2002/05/30 20:45:18 tgl Exp $
PostgreSQL documentation
-->
......@@ -53,12 +53,6 @@ where <replaceable class="PARAMETER">lockmode</replaceable> is one of:
<varlistentry>
<term>ACCESS SHARE MODE</term>
<listitem>
<note>
<para>
This lock mode is acquired automatically over tables being queried.
</para>
</note>
<para>
This is the least restrictive lock mode. It conflicts only with
ACCESS EXCLUSIVE mode. It is used to protect a table from being
......@@ -66,108 +60,114 @@ where <replaceable class="PARAMETER">lockmode</replaceable> is one of:
<command>DROP TABLE</command> and <command>VACUUM FULL</command>
commands.
</para>
<note>
<para>
The <command>SELECT</command> command acquires a
lock of this mode on referenced tables. In general, any query
that only reads a table and does not modify it will acquire
this lock mode.
</para>
</note>
</listitem>
</varlistentry>
<varlistentry>
<term>ROW SHARE MODE</term>
<listitem>
<para>
Conflicts with EXCLUSIVE and ACCESS EXCLUSIVE lock modes.
</para>
<note>
<para>
Automatically acquired by <command>SELECT ... FOR UPDATE</command>.
The <command>SELECT FOR UPDATE</command> command acquires a
lock of this mode on the target table(s) (in addition to
<literal>ACCESS SHARE</literal> locks on any other tables
that are referenced but not selected <option>FOR UPDATE</option>).
</para>
</note>
<para>
Conflicts with EXCLUSIVE and ACCESS EXCLUSIVE lock modes.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>ROW EXCLUSIVE MODE</term>
<listitem>
<para>
Conflicts with SHARE, SHARE ROW EXCLUSIVE, EXCLUSIVE and
ACCESS EXCLUSIVE modes.
</para>
<note>
<para>
Automatically acquired by <command>UPDATE</command>,
The commands <command>UPDATE</command>,
<command>DELETE</command>, and <command>INSERT</command>
statements.
acquire this lock mode on the target table (in addition to
<literal>ACCESS SHARE</literal> locks on any other referenced
tables). In general, this lock mode will be acquired by any
query that modifies the data in a table.
</para>
</note>
<para>
Conflicts with SHARE, SHARE ROW EXCLUSIVE, EXCLUSIVE and
ACCESS EXCLUSIVE modes.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>SHARE UPDATE EXCLUSIVE MODE</term>
<listitem>
<note>
<para>
Automatically acquired by <command>VACUUM</command> (without
<option>FULL</option>).
</para>
</note>
<para>
Conflicts with SHARE UPDATE EXCLUSIVE, SHARE, SHARE ROW EXCLUSIVE,
EXCLUSIVE and
ACCESS EXCLUSIVE modes. This mode protects a table against
concurrent schema changes and VACUUMs.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>SHARE MODE</term>
<listitem>
<note>
<para>
Automatically acquired by <command>CREATE INDEX</command>.
Share-locks the entire table.
Acquired by <command>VACUUM</command> (without
<option>FULL</option>).
</para>
</note>
</listitem>
</varlistentry>
<varlistentry>
<term>SHARE MODE</term>
<listitem>
<para>
Conflicts with ROW EXCLUSIVE, SHARE UPDATE EXCLUSIVE,
SHARE ROW EXCLUSIVE, EXCLUSIVE and
ACCESS EXCLUSIVE modes. This mode protects a table against
concurrent data updates.
concurrent data changes.
</para>
<note>
<para>
Acquired by <command>CREATE INDEX</command>.
</para>
</note>
</listitem>
</varlistentry>
<varlistentry>
<term>SHARE ROW EXCLUSIVE MODE</term>
<listitem>
<note>
<para>
This is like EXCLUSIVE MODE, but allows ROW SHARE locks
by others.
Conflicts with ROW EXCLUSIVE, SHARE UPDATE EXCLUSIVE, SHARE,
SHARE ROW EXCLUSIVE, EXCLUSIVE and ACCESS EXCLUSIVE modes.
</para>
</note>
<note>
<para>
Conflicts with ROW EXCLUSIVE, SHARE UPDATE EXCLUSIVE, SHARE,
SHARE ROW EXCLUSIVE, EXCLUSIVE and ACCESS EXCLUSIVE modes.
This lock mode is not automatically acquired by any
<productname>PostgreSQL</productname> command.
</para>
</note>
</listitem>
</varlistentry>
<varlistentry>
<term>EXCLUSIVE MODE</term>
<listitem>
<note>
<para>
This mode is yet more restrictive than SHARE ROW EXCLUSIVE.
It blocks all concurrent ROW SHARE/SELECT...FOR UPDATE queries.
</para>
</note>
<para>
Conflicts with ROW SHARE, ROW EXCLUSIVE, SHARE UPDATE EXCLUSIVE,
SHARE, SHARE ROW EXCLUSIVE,
......@@ -176,33 +176,33 @@ where <replaceable class="PARAMETER">lockmode</replaceable> is one of:
from the table can proceed in parallel with a transaction holding
this lock mode.
</para>
<note>
<para>
This lock mode is not automatically acquired by any
<productname>PostgreSQL</productname> command.
</para>
</note>
</listitem>
</varlistentry>
<varlistentry>
<term>ACCESS EXCLUSIVE MODE</term>
<listitem>
<note>
<para>
Automatically acquired by <command>ALTER TABLE</command>,
<command>DROP TABLE</command>, <command>VACUUM FULL</command>
statements.
This is the most restrictive lock mode which
protects a locked table from any concurrent operations.
</para>
</note>
<para>
Conflicts with all lock modes. This mode guarantees that the
holder is the only transaction accessing the table in any way.
</para>
<note>
<para>
This lock mode is also acquired by an unqualified
<command>LOCK TABLE</command> (i.e., the command without an explicit
lock mode option).
Acquired by <command>ALTER TABLE</command>,
<command>DROP TABLE</command>, and <command>VACUUM FULL</command>
statements.
This is also the default lock mode for <command>LOCK TABLE</command>
statements that do not specify a mode explicitly.
</para>
</note>
<para>
Conflicts with all lock modes.
</para>
</listitem>
</varlistentry>
</variablelist>
......@@ -255,98 +255,64 @@ ERROR <replaceable class="PARAMETER">name</replaceable>: Table does not exist.
</title>
<para>
<command>LOCK TABLE</command> controls concurrent access to a table
for the duration of a transaction.
<productname>PostgreSQL</productname> always uses the least restrictive
lock mode whenever possible. <command>LOCK TABLE</command>
provides for cases when you might need more restrictive locking.
<command>LOCK TABLE</command> obtains a table-level lock, waiting if
necessary for any conflicting locks to be released. Once obtained,
the lock is held for the remainder of the current transaction.
(There is no <command>UNLOCK TABLE</command> command; locks are always
released at transaction end.)
</para>
<para>
<acronym>RDBMS</acronym> locking uses the following terminology:
<variablelist>
<varlistentry>
<term>EXCLUSIVE</term>
<listitem>
<para>
An exclusive lock prevents other locks of the same type from being
granted. (Note: ROW EXCLUSIVE mode does not follow this naming
convention perfectly, since it is shared at the level of the table;
it is exclusive only with respect to specific rows that are being
updated.)
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>SHARE</term>
<listitem>
<para>
A shared lock allows others to also hold the same type of lock,
but prevents the corresponding EXCLUSIVE lock from being granted.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>ACCESS</term>
<listitem>
<para>
Locks table schema.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>ROW</term>
<listitem>
<para>
Locks individual rows.
</para>
</listitem>
</varlistentry>
</variablelist>
When acquiring locks automatically for commands that reference tables,
<productname>PostgreSQL</productname> always uses the least restrictive
lock mode possible. <command>LOCK TABLE</command>
provides for cases when you might need more restrictive locking.
</para>
<para>
For example, suppose an application runs a transaction at READ COMMITTED
isolation level and needs to ensure the existence of data in a table for
the duration of the
isolation level and needs to ensure that data in a table remains stable
for the duration of the
transaction. To achieve this you could obtain SHARE lock mode over the
table before querying. This will prevent concurrent data changes
and ensure further read operations over the table see data in their
actual current state, because SHARE lock mode conflicts with any ROW
table before querying. This will prevent concurrent data changes
and ensure subsequent reads of the table see a stable
view of committed data, because SHARE lock mode conflicts with the ROW
EXCLUSIVE lock acquired by writers, and your
<command>LOCK TABLE <replaceable class="PARAMETER">name</replaceable> IN SHARE MODE</command>
statement will wait until any concurrent write operations commit or
rollback. Thus, once you obtain the lock, there are no uncommitted
writes outstanding.
statement will wait until any concurrent holders of ROW EXCLUSIVE mode
commit or roll back. Thus, once you obtain the lock, there are no
uncommitted writes outstanding; furthermore none can begin until you
release the lock.
<note>
<para>
To read data in their actual current state when running a transaction
at the SERIALIZABLE isolation level, you have to execute the LOCK TABLE
To achieve a similar effect when running a transaction
at the SERIALIZABLE isolation level, you have to execute the
<command>LOCK TABLE</>
statement before executing any DML statement. A serializable
transaction's view of data will be frozen when its first DML statement
begins.
begins. A later <command>LOCK</> will still prevent concurrent writes
--- but it
won't ensure that what the transaction reads corresponds to the latest
committed values.
</para>
</note>
</para>
<para>
In addition to the requirements above, if a transaction is going to
change data in a table, then SHARE ROW EXCLUSIVE lock mode should
be acquired to prevent deadlock conditions when two concurrent
transactions attempt to lock the table in SHARE mode and then
try to change data in this table, both (implicitly) acquiring
ROW EXCLUSIVE lock mode that conflicts with a concurrent SHARE lock.
If a transaction of this sort is going to
change the data in the table, then it should use SHARE ROW EXCLUSIVE lock
mode instead of SHARE mode. This ensures that only one transaction of
this type runs at a time. Without this, a deadlock is possible: two
transactions might both acquire SHARE mode, and then be unable to also
acquire ROW EXCLUSIVE mode to actually perform their updates. (Note that
a transaction's own locks never conflict, so a transaction can acquire
ROW EXCLUSIVE mode when it holds SHARE mode --- but not if anyone else
holds SHARE mode.)
</para>
<para>
To continue with the deadlock (when two transactions wait for one another)
issue raised above, you should follow two general rules to prevent
deadlock conditions:
Two general rules may be followed to prevent deadlock conditions:
</para>
<itemizedlist>
......@@ -366,9 +332,7 @@ ERROR <replaceable class="PARAMETER">name</replaceable>: Table does not exist.
<listitem>
<para>
Transactions should acquire two conflicting lock modes only if
one of them is self-conflicting (i.e., may be held by only one
transaction at a time). If multiple lock modes are involved,
If multiple lock modes are involved for a single object,
then transactions should always acquire the most restrictive mode first.
</para>
......@@ -380,16 +344,18 @@ ERROR <replaceable class="PARAMETER">name</replaceable>: Table does not exist.
</listitem>
</itemizedlist>
<note>
<para>
<productname>PostgreSQL</productname> does detect deadlocks and will
rollback at least one waiting transaction to resolve the deadlock.
</para>
</note>
<para>
<productname>PostgreSQL</productname> does detect deadlocks and will
rollback at least one waiting transaction to resolve the deadlock.
If it is not practical to code an application to follow the above rules
strictly, an alternative solution is to be prepared to retry transactions
when they are aborted by deadlocks.
</para>
<para>
When locking multiple tables, the command LOCK a, b; is equivalent to LOCK
a; LOCK b;. The tables are locked one-by-one in the order specified in the
When locking multiple tables, the command <literal>LOCK a, b;</> is
equivalent to <literal>LOCK a; LOCK b;</>. The tables are locked one-by-one
in the order specified in the
<command>LOCK</command> command.
</para>
......@@ -415,6 +381,62 @@ ERROR <replaceable class="PARAMETER">name</replaceable>: Table does not exist.
lock will be dropped as soon as it is obtained.
</para>
<para>
<acronym>RDBMS</acronym> locking uses the following standard terminology:
<variablelist>
<varlistentry>
<term>EXCLUSIVE</term>
<listitem>
<para>
An exclusive lock prevents other locks of the same type from being
granted.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>SHARE</term>
<listitem>
<para>
A shared lock allows others to also hold the same type of lock,
but prevents the corresponding EXCLUSIVE lock from being granted.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>ACCESS</term>
<listitem>
<para>
Locks table schema.
</para>
</listitem>
</varlistentry>
<varlistentry>
<term>ROW</term>
<listitem>
<para>
Locks individual rows.
</para>
</listitem>
</varlistentry>
</variablelist>
</para>
<para>
<productname>PostgreSQL</productname> does not follow this terminology
exactly. <command>LOCK TABLE</> only deals with table-level locks, and
so the mode names involving ROW are all misnomers. These mode names
should generally be read as indicating the intention of the user to
acquire row-level locks within the locked table. Also,
ROW EXCLUSIVE mode does not follow this naming convention accurately,
since it is a sharable table lock. Keep in mind that all the lock modes
have identical semantics so far as <command>LOCK TABLE</> is concerned,
differing only in the rules about which modes conflict with which.
</para>
</refsect2>
</refsect1>
......@@ -424,7 +446,7 @@ ERROR <replaceable class="PARAMETER">name</replaceable>: Table does not exist.
</title>
<para>
Illustrate a SHARE lock on a primary key table when going to perform
Obtain a SHARE lock on a primary key table when going to perform
inserts into a foreign key table:
<programlisting>
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或