On Tue, Jul 27, 2004 at 01:32:01PM -0400, Tom Lane wrote:
> Alvaro Herrera <[EMAIL PROTECTED]> writes:
> > There are some likely controversial changes; the Xid caches, in the
> > first place.
>
> No kidding ;-)
Ok, here is another try. This patch includes both the Xid cache
rewrite, source documentation (removed the big comment in xact.c and
added a README file), htup.h comments updated, and some user
documentation.
BTW, the overflow bit is certainly a better approach than the negative
cache. I am not sure about locking w.r.t. the XidCache though.
Please take a look at it.
--
Alvaro Herrera (<alvherre[a]dcc.uchile.cl>)
Maybe there's lots of data loss but the records of data loss are also lost.
(Lincoln Yeoh)
diff -Ncr --exclude-from=diff-ignore 00orig/doc/src/sgml/advanced.sgml
04pgproc/doc/src/sgml/advanced.sgml
*** 00orig/doc/src/sgml/advanced.sgml 2004-04-14 16:45:53.000000000 -0400
--- 04pgproc/doc/src/sgml/advanced.sgml 2004-07-27 10:29:09.000000000 -0400
***************
*** 257,262 ****
--- 257,310 ----
you are using.
</para>
</note>
+
+ <para>
+ It's possible to control the statements in a transaction in a more
+ granular fashion through the use of <firstterm>savepoints</>. Savepoints
+ allow you to selectively discard parts of the transaction, while
+ committing the rest. This is done be defining a savepoint with
+ <command>SAVEPOINT</>, to which you can later roll back using
+ <command>ROLLBACK TO</>. All statements between defining the savepoint
+ and rolling back to it will have no effect on the final transaction.
+ </para>
+
+ <para>
+ After rolling back to a savepoint, it continues to be defined, so you can
+ roll back to it several times. Conversely, if you are sure you won't need
+ to roll back to a particular savepoint again, it can be released, so the
+ system can free some resources. Keep in mind that releasing a savepoint
+ will automatically release all savepoints that were defined after it.
+ </para>
+
+ <para>
+ Remembering the bank database, suppose we debit $100.00 from Alice's
+ account, and credit Bob's account, only to find later that we wanted to
+ credit Wally's account. We could do it using savepoints like
+
+ <programlisting>
+ BEGIN;
+ UPDATE accounts SET balance = balance - 100.00
+ WHERE name = 'Alice';
+ SAVEPOINT my_savepoint;
+ UPDATE accounts SET balance = balance + 100.00
+ WHERE name = 'Bob';
+ -- oops ... forget that and use Wally's account
+ ROLLBACK TO my_savepoint;
+ UPDATE accounts SET balance = balance + 100.00
+ WHERE name = 'Wally';
+ COMMIT;
+ </programlisting>
+ </para>
+
+ <para>
+ This example is, of course, oversimplified, but there's a lot of control
+ to be had over a transaction block through the use of savepoints.
+ Moreover, <command>ROLLBACK TO</> is the only way to regain control of a
+ transaction block that was automatically put on aborted state by the
+ system for some reason, short of rolling it back completely and starting
+ again.
+ </para>
+
</sect1>
diff -Ncr --exclude-from=diff-ignore 00orig/doc/src/sgml/ref/allfiles.sgml
04pgproc/doc/src/sgml/ref/allfiles.sgml
*** 00orig/doc/src/sgml/ref/allfiles.sgml 2004-06-26 00:28:45.000000000 -0400
--- 04pgproc/doc/src/sgml/ref/allfiles.sgml 2004-07-27 10:29:09.000000000 -0400
***************
*** 88,96 ****
--- 88,99 ----
<!entity notify system "notify.sgml">
<!entity prepare system "prepare.sgml">
<!entity reindex system "reindex.sgml">
+ <!entity releaseSavepoint system "release.sgml">
<!entity reset system "reset.sgml">
<!entity revoke system "revoke.sgml">
<!entity rollback system "rollback.sgml">
+ <!entity rollbackTo system "rollback_to.sgml">
+ <!entity savepoint system "savepoint.sgml">
<!entity select system "select.sgml">
<!entity selectInto system "select_into.sgml">
<!entity set system "set.sgml">
diff -Ncr --exclude-from=diff-ignore 00orig/doc/src/sgml/ref/begin.sgml
04pgproc/doc/src/sgml/ref/begin.sgml
*** 00orig/doc/src/sgml/ref/begin.sgml 2004-01-11 06:24:17.000000000 -0300
--- 04pgproc/doc/src/sgml/ref/begin.sgml 2004-07-27 10:29:09.000000000 -0400
***************
*** 145,150 ****
--- 145,151 ----
<member><xref linkend="sql-commit" endterm="sql-commit-title"></member>
<member><xref linkend="sql-rollback" endterm="sql-rollback-title"></member>
<member><xref linkend="sql-start-transaction"
endterm="sql-start-transaction-title"></member>
+ <member><xref linkend="sql-savepoint" endterm="sql-savepoint-title"></member>
</simplelist>
</refsect1>
</refentry>
diff -Ncr --exclude-from=diff-ignore 00orig/doc/src/sgml/ref/release.sgml
04pgproc/doc/src/sgml/ref/release.sgml
*** 00orig/doc/src/sgml/ref/release.sgml 1969-12-31 21:00:00.000000000 -0300
--- 04pgproc/doc/src/sgml/ref/release.sgml 2004-07-27 10:29:09.000000000 -0400
***************
*** 0 ****
--- 1,138 ----
+ <!--
+ $PostgreSQL$
+ PostgreSQL documentation
+ -->
+
+ <refentry id="SQL-RELEASE">
+ <refmeta>
+ <refentrytitle id="SQL-RELEASE-TITLE">RELEASE</refentrytitle>
+ <refmiscinfo>SQL - Language Statements</refmiscinfo>
+ </refmeta>
+
+ <refnamediv>
+ <refname>RELEASE</refname>
+ <refpurpose>destroy a previously defined savepoint</refpurpose>
+ </refnamediv>
+
+ <indexterm zone="sql-release">
+ <primary>RELEASE</primary>
+ </indexterm>
+
+ <indexterm zone="sql-release">
+ <primary>savepoints</primary>
+ <secondary>releasing</secondary>
+ </indexterm>
+
+ <refsynopsisdiv>
+ <synopsis>
+ RELEASE <replaceable>savepoint_name</replaceable>
+ </synopsis>
+ </refsynopsisdiv>
+
+ <refsect1>
+ <title>Description</title>
+
+ <para>
+ <command>RELEASE</command> destroys a previously defined savepoint
+ in the current transaction.
+ </para>
+
+ <para>
+ Destroying a savepoint makes it—and all savepoints established after
+ it was established—unavailable as rollback points,
+ but it has no other user visible behavior. It does not undo the
+ effects of command executed after the savepoint was established.
+ To do that, see <xref linkend="sql-rollback-to"
+ endterm="sql-rollback-to-title">.
+ </para>
+
+ <para>
+ <command>RELEASE</command> also destroys all savepoints that were established
+ after the named savepoint was established.
+ </para>
+
+ <refsect1>
+ <title>Parameters</title>
+
+ <variablelist>
+ <varlistentry>
+ <term><replaceable>savepoint_name</replaceable></term>
+ <listitem>
+ <para>
+ The name of the savepoint to destroy.
+ </para>
+ </listitem>
+ </varlistentry>
+ </variablelist>
+ </refsect1>
+
+ <refsect1>
+ <title>Notes</title>
+
+ <para>
+ Specifying a savepoint name that was not previously defined raises
+ an exception.
+ </para>
+
+ <para>
+ It is not possible to release a savepoint when the transaction is in
+ aborted state.
+ </para>
+
+ <para>
+ If multiple savepoints have the same name, only the one that was last
+ defined is released.
+ </para>
+
+ </refsect1>
+
+ <refsect1>
+ <title>Examples</title>
+
+ <para>
+ To establish and later destroy a savepoint:
+ <programlisting>
+ BEGIN;
+ INSERT INTO table VALUES (3);
+ SAVEPOINT my_savepoint;
+ INSERT INTO table VALUES (4);
+ RELEASE my_savepoint;
+ COMMIT;
+ </refsect1>
+
+ <refsect1>
+ <title>Compatibility</title>
+
+ <para>
+ RELEASE is fully conforming to the SQL standard.
+ </para>
+ </refsect1>
+
+ <refsect1>
+ <title>See Also</title>
+
+ <simplelist type="inline">
+ <member><xref linkend="sql-begin" endterm="sql-begin-title"></member>
+ <member><xref linkend="sql-rollback" endterm="sql-rollback-title"></member>
+ <member><xref linkend="sql-savepoint" endterm="sql-savepoint-title"></member>
+ <member><xref linkend="sql-commit" endterm="sql-commit-title"></member>
+ </simplelist>
+ </refsect1>
+ </refentry>
+
+ <!-- Keep this comment at the end of the file
+ Local variables:
+ mode: sgml
+ sgml-omittag:nil
+ sgml-shorttag:t
+ sgml-minimize-attributes:nil
+ sgml-always-quote-attributes:t
+ sgml-indent-step:1
+ sgml-indent-data:t
+ sgml-parent-document:nil
+ sgml-default-dtd-file:"../reference.ced"
+ sgml-exposed-tags:nil
+ sgml-local-catalogs:"/usr/lib/sgml/catalog"
+ sgml-local-ecat-files:nil
+ End:
+ -->
diff -Ncr --exclude-from=diff-ignore 00orig/doc/src/sgml/ref/rollback.sgml
04pgproc/doc/src/sgml/ref/rollback.sgml
*** 00orig/doc/src/sgml/ref/rollback.sgml 2003-11-29 16:51:39.000000000 -0300
--- 04pgproc/doc/src/sgml/ref/rollback.sgml 2004-07-27 10:29:09.000000000 -0400
***************
*** 90,95 ****
--- 90,96 ----
<simplelist type="inline">
<member><xref linkend="sql-begin" endterm="sql-begin-title"></member>
<member><xref linkend="sql-commit" endterm="sql-commit-title"></member>
+ <member><xref linkend="sql-rollback-to" endterm="sql-rollback-to-title"></member>
</simplelist>
</refsect1>
</refentry>
diff -Ncr --exclude-from=diff-ignore 00orig/doc/src/sgml/ref/rollback_to.sgml
04pgproc/doc/src/sgml/ref/rollback_to.sgml
*** 00orig/doc/src/sgml/ref/rollback_to.sgml 1969-12-31 21:00:00.000000000 -0300
--- 04pgproc/doc/src/sgml/ref/rollback_to.sgml 2004-07-27 10:29:09.000000000 -0400
***************
*** 0 ****
--- 1,158 ----
+ <!--
+ $PostgreSQL$
+ PostgreSQL documentation
+ -->
+
+ <refentry id="SQL-ROLLBACK-TO">
+ <refmeta>
+ <refentrytitle id="SQL-ROLLBACK-TO-TITLE">ROLLBACK TO</refentrytitle>
+ <refmiscinfo>SQL - Language Statements</refmiscinfo>
+ </refmeta>
+
+ <refnamediv>
+ <refname>ROLLBACK TO</refname>
+ <refpurpose>roll back to a savepoint</refpurpose>
+ </refnamediv>
+
+ <indexterm zone="sql-rollback-to">
+ <primary>ROLLBACK TO</primary>
+ </indexterm>
+
+ <indexterm zone="sql-rollback-to">
+ <primary>savepoints</primary>
+ <secondary>rolling back</secondary>
+ </indexterm>
+
+ <refsynopsisdiv>
+ <synopsis>
+ ROLLBACK TO <replaceable>savepoint_name</replaceable>
+ </synopsis>
+ </refsynopsisdiv>
+
+ <refsect1>
+ <title>Description</title>
+
+ <para>
+ Roll back all commands that were executed and destroy all savepoints that
+ were created after the savepoint was established. The savepoint is
+ automatically established again.
+ </para>
+
+ <refsect1>
+ <title>Parameters</title>
+
+ <variablelist>
+ <varlistentry>
+ <term><replaceable class="PARAMETER">savepoint_name</></term>
+ <listitem>
+ <para>
+ The savepoint to roll back to.
+ </para>
+ </listitem>
+ </varlistentry>
+ </variablelist>
+ </refsect1>
+
+ <refsect1>
+ <title>Notes</title>
+
+ <para>
+ Use <xref linkend="SQL-RELEASE" endterm="SQL-RELEASE-TITLE"> to
+ destroy a savepoint without discarding the effects of commands executed
+ after it was established.
+ </para>
+
+ <para>
+ Specifying a savepoint name that has not been established causes an
+ exception to be raised.
+ </para>
+
+ <para>
+ Cursors have somewhat non-transactional behavior with respect to
+ savepoints. Any cursor that is opened inside the savepoint is not closed
+ when the savepoint is rolled back. If a cursor is affected by a
+ <command>FETCH</> command inside a savepoint that is later rolled
+ back, the cursor position remains at the position that <command>FETCH</>
+ left it pointing to (that is, <command>FETCH</> is not rolled back).
+ A cursor whose execution causes a transaction to abort is put in a
+ can't-execute state, so while the transaction can be restored using
+ <command>ROLLBACK TO</>, the cursor no longer can be used.
+ </para>
+ </refsect1>
+
+ <refsect1>
+ <title>Examples</title>
+
+ <para>
+ To undo the effects of the commands executed after <literal>my_savepoint</literal>
+ was established, and establish <literal>my_savepoint</> again:
+ <programlisting>
+ ROLLBACK TO my_savepoint;
+ </programlisting>
+ </para>
+
+ <para>
+ Cursor positions are not affected by savepoint rollback:
+ <programlisting>
+ BEGIN;
+
+ DECLARE foo CURSOR FOR SELECT 1 UNION SELECT 2;
+
+ SAVEPOINT foo;
+
+ FETCH 1 FROM foo;
+ ?column?
+ ----------
+ 1
+
+ ROLLBACK TO foo;
+
+ FETCH 1 FROM foo;
+ ?column?
+ ----------
+ 2
+
+ COMMIT;
+ </programlisting>
+ </para>
+
+
+ </refsect1>
+
+ <refsect1>
+ <title>Compatibility</title>
+
+ <para>
+ This command is fully SQL standard conforming.
+ </para>
+ </refsect1>
+
+ <refsect1>
+ <title>See Also</title>
+
+ <simplelist type="inline">
+ <member><xref linkend="sql-begin" endterm="sql-begin-title"></member>
+ <member><xref linkend="sql-commit" endterm="sql-commit-title"></member>
+ <member><xref linkend="sql-savepoint" endterm="sql-savepoint-title"></member>
+ <member><xref linkend="sql-release" endterm="sql-release-title"></member>
+ <member><xref linkend="sql-rollback" endterm="sql-rollback-title"></member>
+ </simplelist>
+ </refsect1>
+ </refentry>
+
+ <!-- Keep this comment at the end of the file
+ Local variables:
+ mode: sgml
+ sgml-omittag:nil
+ sgml-shorttag:t
+ sgml-minimize-attributes:nil
+ sgml-always-quote-attributes:t
+ sgml-indent-step:1
+ sgml-indent-data:t
+ sgml-parent-document:nil
+ sgml-default-dtd-file:"../reference.ced"
+ sgml-exposed-tags:nil
+ sgml-local-catalogs:"/usr/lib/sgml/catalog"
+ sgml-local-ecat-files:nil
+ End:
+ -->
diff -Ncr --exclude-from=diff-ignore 00orig/doc/src/sgml/ref/savepoint.sgml
04pgproc/doc/src/sgml/ref/savepoint.sgml
*** 00orig/doc/src/sgml/ref/savepoint.sgml 1969-12-31 21:00:00.000000000 -0300
--- 04pgproc/doc/src/sgml/ref/savepoint.sgml 2004-07-27 10:29:09.000000000 -0400
***************
*** 0 ****
--- 1,153 ----
+ <!--
+ $PostgreSQL$
+ PostgreSQL documentation
+ -->
+
+ <refentry id="SQL-SAVEPOINT">
+ <refmeta>
+ <refentrytitle id="SQL-SAVEPOINT-TITLE">SAVEPOINT</refentrytitle>
+ <refmiscinfo>SQL - Language Statements</refmiscinfo>
+ </refmeta>
+
+ <refnamediv>
+ <refname>SAVEPOINT</refname>
+ <refpurpose>define a new savepoint within the current transaction</refpurpose>
+ </refnamediv>
+
+ <indexterm zone="sql-savepoint">
+ <primary>SAVEPOINT</primary>
+ </indexterm>
+
+ <indexterm zone="sql-savepoint">
+ <primary>savepoints</primary>
+ <secondary>defining</secondary>
+ </indexterm>
+
+ <refsynopsisdiv>
+ <synopsis>
+ SAVEPOINT <replaceable>savepoint_name</replaceable>
+ </synopsis>
+ </refsynopsisdiv>
+
+ <refsect1>
+ <title>Description</title>
+
+ <para>
+ <command>SAVEPOINT</command> establishes a new savepoint within
+ the current transaction.
+ </para>
+
+ </refsect1>
+
+ <refsect1>
+ <title>Parameters</title>
+
+ <variablelist>
+ <varlistentry>
+ <term><replaceable>savepoint_name</replaceable></term>
+ <listitem>
+ <para>
+ The name to give to the new savepoint.
+ </para>
+ </listitem>
+ </varlistentry>
+ </variablelist>
+ </refsect1>
+
+ <refsect1>
+ <title>Notes</title>
+
+ <para>
+ A savepoint is a special mark inside a transaction that allows all commands
+ that are executed after it was established to be rolled back.
+ Alternatively, a savepoint can be destroyed so that it isn't a possible
+ rollback destination anymore. In this case, all commands that were executed after
+ the savepoint was established are preserved.
+ </para>
+
+ <para>
+ Use <xref linkend="SQL-ROLLBACK-TO" endterm="SQL-ROLLBACK-TO-TITLE"> to
+ rollback to a savepoint. Use <xref linkend="SQL-RELEASE"
+ endterm="SQL-RELEASE-TITLE"> to destroy a savepoint, keeping
+ the effects of commands executed after it was established.
+ </para>
+
+ <para>
+ Savepoints can only be established when inside a transaction block.
+ Issuing <command>SAVEPOINT</> when not inside a transaction block
+ will cause an exception to be raised.
+ </para>
+
+ <para>
+ There can be multiple savepoints defined within a transaction.
+ </para>
+ </refsect1>
+
+ <refsect1>
+ <title>Examples</title>
+
+ <para>
+ To establish a savepoint and undo the effects of all commands executed
+ after it was established, keeping only the first inserted value
+ in the table:
+ <programlisting>
+ BEGIN;
+ INSERT INTO table VALUES (1);
+ SAVEPOINT my_savepoint;
+ INSERT INTO table VALUES (2);
+ ROLLBACK TO my_savepoint;
+ COMMIT;
+ </programlisting>
+ </para>
+
+ <para>
+ To establish and later destroy a savepoint, keeping both values in the table:
+ <programlisting>
+ BEGIN;
+ INSERT INTO table VALUES (3);
+ SAVEPOINT my_savepoint;
+ INSERT INTO table VALUES (4);
+ RELEASE my_savepoint;
+ COMMIT;
+ </refsect1>
+
+ <refsect1>
+ <title>Compatibility</title>
+
+ <para>
+ SQL requires a savepoint to be automatically destroyed when another savepoint
+ with the same name is established. In <productname>PostgreSQL</>, the old
+ savepoint is kept, though only the last one will be used when rolling back or
+ releasing. Other than that, <command>SAVEPOINT</command> is fully SQL conforming.
+ </para>
+ </refsect1>
+
+ <refsect1>
+ <title>See Also</title>
+
+ <simplelist type="inline">
+ <member><xref linkend="sql-begin" endterm="sql-begin-title"></member>
+ <member><xref linkend="sql-rollback" endterm="sql-rollback-title"></member>
+ <member><xref linkend="sql-rollback-to" endterm="sql-rollback-to-title"></member>
+ <member><xref linkend="sql-release" endterm="sql-release-title"></member>
+ <member><xref linkend="sql-commit" endterm="sql-commit-title"></member>
+ </simplelist>
+ </refsect1>
+ </refentry>
+
+ <!-- Keep this comment at the end of the file
+ Local variables:
+ mode: sgml
+ sgml-omittag:nil
+ sgml-shorttag:t
+ sgml-minimize-attributes:nil
+ sgml-always-quote-attributes:t
+ sgml-indent-step:1
+ sgml-indent-data:t
+ sgml-parent-document:nil
+ sgml-default-dtd-file:"../reference.ced"
+ sgml-exposed-tags:nil
+ sgml-local-catalogs:"/usr/lib/sgml/catalog"
+ sgml-local-ecat-files:nil
+ End:
+ -->
diff -Ncr --exclude-from=diff-ignore 00orig/doc/src/sgml/ref/start_transaction.sgml
04pgproc/doc/src/sgml/ref/start_transaction.sgml
*** 00orig/doc/src/sgml/ref/start_transaction.sgml 2004-01-11 02:46:58.000000000
-0300
--- 04pgproc/doc/src/sgml/ref/start_transaction.sgml 2004-07-27 10:29:09.000000000
-0400
***************
*** 66,71 ****
--- 66,72 ----
<member><xref linkend="sql-commit" endterm="sql-commit-title"></member>
<member><xref linkend="sql-rollback" endterm="sql-rollback-title"></member>
<member><xref linkend="sql-set-transaction"
endterm="sql-set-transaction-title"></member>
+ <member><xref linkend="sql-savepoint" endterm="sql-savepoint-title"></member>
</simplelist>
</refsect1>
</refentry>
diff -Ncr --exclude-from=diff-ignore 00orig/doc/src/sgml/reference.sgml
04pgproc/doc/src/sgml/reference.sgml
*** 00orig/doc/src/sgml/reference.sgml 2004-06-26 00:28:44.000000000 -0400
--- 04pgproc/doc/src/sgml/reference.sgml 2004-07-27 10:29:09.000000000 -0400
***************
*** 120,128 ****
--- 120,131 ----
¬ify;
&prepare;
&reindex;
+ &releaseSavepoint;
&reset;
&revoke;
&rollback;
+ &rollbackTo;
+ &savepoint;
&select;
&selectInto;
&set;
diff -Ncr --exclude-from=diff-ignore 00orig/src/backend/access/transam/README
04pgproc/src/backend/access/transam/README
*** 00orig/src/backend/access/transam/README 1969-12-31 21:00:00.000000000 -0300
--- 04pgproc/src/backend/access/transam/README 2004-07-28 19:18:11.856226058 -0400
***************
*** 0 ****
--- 1,224 ----
+ The Transaction System
+ ======================
+
+ xact.c
+ ------
+
+ PostgreSQL's transaction system is a three-layer system, implementing
+ low-level transactions and subtransactions, on top of which rests the
+ mainloop's control code, which in turn implements user-visible transactions
+ and savepoints.
+
+ The middle layer of code is called by postgres.c before and after the
+ processing of each query,
+
+ StartTransactionCommand
+ CommitTransactionCommand
+ AbortCurrentTransaction
+
+ Meanwhile, the user can alter the system's state by issuing the SQL commands
+ BEGIN, ROLLBACK, SAVEPOINT or RELEASE. The traffic cop redirects these calls
+ to the toplevel routines
+
+ BeginTransactionBlock
+ EndTransactionBlock
+ UserAbortTransactionBlock
+ DefineSavepoint
+ RollbackToSavepoint
+ ReleaseSavepoint
+ RollbackAndReleaseSavepoint
+
+ Depending on the current state of the system, these function call low level
+ functions to activate the real transaction system,
+
+ StartTransaction
+ CommitTransaction
+ AbortTransaction
+ CleanupTransaction
+ StartSubTransaction
+ CommitSubTransaction
+ AbortSubTransaction
+ CleanupSubTransaction
+
+ Additionally, within a transaction, CommandCounterIncrement is called to
+ increment the command counter, which allows future commands to "see" the
+ effects of previous commands within the same transaction. Note that this is
+ done automatically by CommitTransactionCommand after each query inside a
+ transaction block, but some utility functions also do it to allow some
+ operations (usually in the system catalogs) to be seen by future operations in
+ the same utility command processing (for example, in DefineRelation it is done
+ after creating the heap so the pg_class row is visible, to be able to lock
+ it).
+
+
+ For example, consider the following sequence of user commands:
+
+ 1) BEGIN
+ 2) SELECT * FROM foo
+ 3) INSERT INTO foo VALUES (...)
+ 4) COMMIT
+
+ In the main processing loop, this results in the following function call
+ sequence:
+
+ / StartTransactionCommand;
+ / ProcessUtility; << BEGIN
+ 1) < BeginTransactionBlock;
+ \ StartTransaction;
+ \ CommitTransactionCommand;
+
+ / StartTransactionCommand;
+ 2) / ProcessQuery; << SELECT * FROM foo
+ \ CommitTransactionCommand;
+ \ CommandCounterIncrement;
+
+ / StartTransactionCommand;
+ 3) / ProcessQuery; << INSERT INTO foo VALUES (...)
+ \ CommitTransactionCommand;
+ \ CommandCounterIncrement;
+
+ / StartTransactionCommand;
+ / ProcessUtility; << COMMIT
+ 4) < EndTransactionBlock;
+ \ CommitTransaction;
+ \ CommitTransactionCommand;
+
+ The point of this example is to demonstrate the need for
+ StartTransactionCommand and CommitTransactionCommand to be state smart -- they
+ should do nothing in between the calls to BeginTransactionBlock and
+ EndTransactionBlock and outside these calls they need to do normal start,
+ commit or abort processing.
+
+ Furthermore, suppose the "SELECT * FROM foo" caused an abort condition. In
+ this case AbortCurrentTransaction is called, and the transaction is put in
+ aborted state. In this state, any user input is ignored except for
+ transaction-termination statements, or ROLLBACK TO <savepoint> commands.
+
+ Transaction aborts can occur in two ways:
+
+ 1) system dies from some internal cause (syntax error, etc)
+ 2) user types ROLLBACK
+
+ The reason we have to distinguish them is illustrated by the following two
+ situations:
+
+ case 1 case 2
+ ------ ------
+ 1) user types BEGIN 1) user types BEGIN
+ 2) user does something 2) user does something
+ 3) user does not like what 3) system aborts for some reason
+ she sees and types ABORT (syntax error, etc)
+
+ In case 1, we want to abort the transaction and return to the default state.
+ In case 2, there may be more commands coming our way which are part of the
+ same transaction block; we have to ignore these commands until we see a COMMIT
+ or ROLLBACK.
+
+ Internal aborts are handled by AbortCurrentTransaction, while user aborts are
+ handled by UserAbortTransactionBlock. Both of them rely on AbortTransaction
+ to do all the real work. The only difference is what state we enter after
+ AbortTransaction does its work:
+
+ * AbortCurrentTransaction leaves us in TBLOCK_ABORT,
+ * UserAbortTransactionBlock leaves us in TBLOCK_ENDABORT
+
+ Low-level transaction abort handling is divided in two phases:
+ * AbortTransaction executes as soon as we realize the transaction has
+ failed. It should release all shared resources (locks etc) so that we do
+ not delay other backends unnecessarily.
+ * CleanupTransaction executes when we finally see a user COMMIT
+ or ROLLBACK command; it cleans things up and gets us out of the transaction
+ internally. In particular, we mustn't destroy TopTransactionContext until
+ this point.
+
+ Also, note that when a transaction is committed, we don't close it right away.
+ Rather it's put in TBLOCK_END state, which means that when
+ CommitTransactionCommand is called after the query has finished processing,
+ the transaction has to be closed. The distinction is subtle but important,
+ because it means that control will leave the xact.c code with the transaction
+ open, and the main loop will be able to keep processing inside the same
+ transaction. So, in a sense, transaction commit is also handled in two
+ phases, the first at EndTransactionBlock and the second at
+ CommitTransactionCommand.
+
+ The rest of the code in xact.c are routines to support the creation and
+ finishing of transactions and subtransactions. For example, AtStart_Memory
+ takes care of initializing the memory subsystem at main transaction start.
+
+
+ Subtransaction handling
+ -----------------------
+
+ Subtransactions are implemented using a stack of TransactionState structures,
+ which has a pointer to its parent transaction. When a new subtransaction is
+ to be opened, PushTransaction is called, which creates a new TransactionState,
+ with its parent pointing to the current transaction. StartSubTransaction is
+ in charge of initializing the new TransactionState to sane values, and
+ properly initializing other subsystems.
+
+ When closing a subtransaction, either CommitSubTransaction has to be called
+ (if the subtransaction is committing), or AbortSubTransaction and
+ CleanupSubTransaction (if it's aborting). In either case, PopTransaction is
+ called so the system returns to the parent transaction.
+
+ One important point regarding subtransaction handling is that several may need
+ to be closed in response to a single user command. That's because savepoints
+ have names, and we allow to commit or rollback a savepoint by name, which not
+ necessarily is the one that was last opened. In the case of subtransaction
+ commit this is not a problem, and we close all the involved subtransactions
+ right away by calling CommitTransactionToLevel, which in turn calls
+ CommitSubTransaction and PopTransaction as many times as needed.
+
+ In the case of subtransaction abort (when the user issues ROLLBACK TO
+ <savepoint>), things are not so easy. We have to keep the subtransactions
+ open and return control to the main loop. So what RollbackToSavepoint does is
+ abort the innermost subtransaction and put it in TBLOCK_SUBENDABORT state, and
+ put the rest in TBLOCK_SUBABORT_PENDING state. Then we return control to the
+ main loop, which will in turn return control to us by calling
+ CommitTransactionCommand. At this point we can close all subtransactions that
+ are marked with the "abort pending" state.
+
+
+
+ pg_clog and pg_subtrans
+ -----------------------
+
+ pg_clog and pg_subtrans are permanent (on-disk) storage of transaction related
+ information. There is a limited number of pages of each kept in memory, so
+ in many cases there is no need to actually read from disk. However, if
+ there's a long running transaction or a backend sitting idle with an open
+ transaction, it's necessary to be able to read and write this information from
+ disk. They also allow information to be permanent across server restarts.
+
+ pg_clog records the commit status for each transaction. A transaction can be
+ in progress, committed, aborted, or "sub-committed". This last state means
+ that it's a subtransaction that's no longer running, but it's parent has not
+ updated its state yet (either it is still running, or the backend crashed
+ without updating its status). A sub-committed transaction's status will be
+ updated again to the final value as soon as the parent commits or aborts, or
+ when the parent is detected to be aborted.
+
+ Savepoints are implemented using subtransactions. A subtransaction is a
+ transaction inside a transaction; it gets its own TransactionId, but its
+ commit or abort status is not only dependent on whether it committed itself,
+ but also whether its parent transaction committed. To implement multiple
+ savepoints in a transaction we allow unlimited transaction nesting depth, so
+ any particular subtransaction's commit state is dependent on the commit status
+ of each and every ancestor transaction.
+
+ The "subtransaction parent" (pg_subtrans) mechanism records, for each
+ transaction, the TransactionId of its parent transaction. This information is
+ stored as soon as the subtransaction is created.
+
+ pg_subtrans is used to know whether the transaction in question is still
+ running --- the main Xid of a transaction is recorded in the PGPROC struct,
+ but since we allow arbitrary nesting of subtransactions, we can't fit all Xids
+ in shared memory, so we have to store them on disk. Note, however, that for
+ each transaction we keep a "cache" of Xids that are known to be part of the
+ transaction tree, so we can skip looking at pg_subtrans unless we know the
+ cache has been overflowed. See storage/ipc/sinval.c for the gory details.
+
+ slru.c is the supporting mechanism for both pg_clog and pg_subtrans. It
+ implements the LRU policy for in-memory buffer pages. The high-level routines
+ for pg_clog are implemented transam.c, while the low-level functions are in
+ clog.c. pg_subtrans is contained completely in subtrans.c.
diff -Ncr --exclude-from=diff-ignore 00orig/src/backend/access/transam/xact.c
04pgproc/src/backend/access/transam/xact.c
*** 00orig/src/backend/access/transam/xact.c 2004-07-28 14:36:40.351828862 -0400
--- 04pgproc/src/backend/access/transam/xact.c 2004-07-28 14:24:59.073029982 -0400
***************
*** 10,141 ****
* IDENTIFICATION
* $PostgreSQL: pgsql-server/src/backend/access/transam/xact.c,v 1.172
2004/07/27 05:10:49 tgl Exp $
*
- * NOTES
- * Transaction aborts can now occur two ways:
- *
- * 1) system dies from some internal cause (syntax error, etc..)
- * 2) user types ABORT
- *
- * These two cases used to be treated identically, but now
- * we need to distinguish them. Why? consider the following
- * two situations:
- *
- * case 1
case 2
- * ------
------
- * 1) user types BEGIN 1) user types BEGIN
- * 2) user does something 2) user does something
- * 3) user does not like what 3) system aborts for some
reason
- * she sees and types ABORT
- *
- * In case 1, we want to abort the transaction and return to the
- * default state. In case 2, there may be more commands coming
- * our way which are part of the same transaction block and we have
- * to ignore these commands until we see a COMMIT transaction or
- * ROLLBACK.
- *
- * Internal aborts are now handled by AbortTransactionBlock(), just as
- * they always have been, and user aborts are now handled by
- * UserAbortTransactionBlock(). Both of them rely on AbortTransaction()
- * to do all the real work. The only difference is what state we
- * enter after AbortTransaction() does its work:
- *
- * * AbortTransactionBlock() leaves us in TBLOCK_ABORT and
- * * UserAbortTransactionBlock() leaves us in TBLOCK_ENDABORT
- *
- * Low-level transaction abort handling is divided into two phases:
- * * AbortTransaction() executes as soon as we realize the transaction
- * has failed. It should release all shared resources (locks etc)
- * so that we do not delay other backends unnecessarily.
- * * CleanupTransaction() executes when we finally see a user COMMIT
- * or ROLLBACK command; it cleans things up and gets us out of
- * the transaction internally. In particular, we mustn't destroy
- * TopTransactionContext until this point.
- *
- * NOTES
- * The essential aspects of the transaction system are:
- *
- * o transaction id generation
- * o transaction log updating
- * o memory cleanup
- * o cache invalidation
- * o lock cleanup
- *
- * Hence, the functional division of the transaction code is
- * based on which of the above things need to be done during
- * a start/commit/abort transaction. For instance, the
- * routine AtCommit_Memory() takes care of all the memory
- * cleanup stuff done at commit time.
- *
- * The code is layered as follows:
- *
- * StartTransaction
- * CommitTransaction
- * AbortTransaction
- * CleanupTransaction
- *
- * are provided to do the lower level work like recording
- * the transaction status in the log and doing memory cleanup.
- * above these routines are another set of functions:
- *
- * StartTransactionCommand
- * CommitTransactionCommand
- * AbortCurrentTransaction
- *
- * These are the routines used in the postgres main processing
- * loop. They are sensitive to the current transaction block state
- * and make calls to the lower level routines appropriately.
- *
- * Support for transaction blocks is provided via the functions:
- *
- * BeginTransactionBlock
- * CommitTransactionBlock
- * AbortTransactionBlock
- *
- * These are invoked only in response to a user "BEGIN WORK", "COMMIT",
- * or "ROLLBACK" command. The tricky part about these functions
- * is that they are called within the postgres main loop, in between
- * the StartTransactionCommand() and CommitTransactionCommand().
- *
- * For example, consider the following sequence of user commands:
- *
- * 1) begin
- * 2) select * from foo
- * 3) insert into foo (bar = baz)
- * 4) commit
- *
- * in the main processing loop, this results in the following
- * transaction sequence:
- *
- * / StartTransactionCommand();
- * 1) / ProcessUtility(); << begin
- * \ BeginTransactionBlock();
- * \ CommitTransactionCommand();
- *
- * / StartTransactionCommand();
- * 2) < ProcessQuery(); << select *
from foo
- * \ CommitTransactionCommand();
- *
- * / StartTransactionCommand();
- * 3) < ProcessQuery(); << insert into
foo (bar = baz)
- * \ CommitTransactionCommand();
- *
- * / StartTransactionCommand();
- * 4) / ProcessUtility(); << commit
- * \ CommitTransactionBlock();
- * \ CommitTransactionCommand();
- *
- * The point of this example is to demonstrate the need for
- * StartTransactionCommand() and CommitTransactionCommand() to
- * be state smart -- they should do nothing in between the calls
- * to BeginTransactionBlock() and EndTransactionBlock() and
- * outside these calls they need to do normal start/commit
- * processing.
- *
- * Furthermore, suppose the "select * from foo" caused an abort
- * condition. We would then want to abort the transaction and
- * ignore all subsequent commands up to the "commit".
- * -cim 3/23/90
- *
*-------------------------------------------------------------------------
*/
--- 10,15 ----
***************
*** 1020,1025 ****
--- 894,901 ----
TransactionIdAbortTree(nchildren, children);
TransactionIdAbort(xid);
+ XidCacheClean();
+
END_CRIT_SECTION();
}
***************
*** 1159,1164 ****
--- 1035,1042 ----
TransactionIdAbortTree(nchildren, children);
TransactionIdAbort(xid);
+ XidCacheRemoveRunningXids(nchildren, children, xid);
+
END_CRIT_SECTION();
}
***************
*** 1390,1395 ****
--- 1268,1278 ----
MyProc->xid = InvalidTransactionId;
MyProc->xmin = InvalidTransactionId;
LWLockRelease(SInvalLock);
+
+ /*
+ * Clean up the Xid cache.
+ */
+ XidCacheClean();
}
/*
***************
*** 1589,1595 ****
* State should still be TRANS_ABORT from AbortTransaction().
*/
if (s->state != TRANS_ABORT)
! elog(FATAL, "CleanupTransaction and not in abort state");
/*
* do abort cleanup processing
--- 1472,1479 ----
* State should still be TRANS_ABORT from AbortTransaction().
*/
if (s->state != TRANS_ABORT)
! elog(FATAL, "CleanupTransaction while in %s state",
! TransStateAsString(s->state));
/*
* do abort cleanup processing
***************
*** 2897,2902 ****
--- 2781,2793 ----
XactLockTableInsert(s->transactionIdData);
/*
+ * Ideally, we would only cache Xids of subtransactions that write tuples
+ * in permanent storage. We have no clean way of knowing that, however
+ * (much less in advance ...)
+ */
+ XidCacheAddRunningXid(s->transactionIdData);
+
+ /*
* Finish setup of other transaction state fields.
*/
s->currentUser = GetUserId();
diff -Ncr --exclude-from=diff-ignore 00orig/src/backend/storage/ipc/sinval.c
04pgproc/src/backend/storage/ipc/sinval.c
*** 00orig/src/backend/storage/ipc/sinval.c 2004-06-30 23:50:07.000000000 -0400
--- 04pgproc/src/backend/storage/ipc/sinval.c 2004-07-28 19:44:01.587585261 -0400
***************
*** 27,32 ****
--- 27,53 ----
#include "utils/tqual.h"
#include "miscadmin.h"
+ #ifdef XIDCACHE_DEBUG
+ static void
+ DisplayXidCache(int code, Datum arg);
+
+ /* counters for XidCache measurement */
+ static int xc_by_recent_xmin = 0;
+ static int xc_by_main_xid = 0;
+ static int xc_by_child_xid = 0;
+ static int xc_slow_answer = 0;
+ #define xc_by_recent_xmin_inc xc_by_recent_xmin++
+ #define xc_by_main_xid_inc xc_by_main_xid++
+ #define xc_by_child_xid_inc xc_by_child_xid++
+ #define xc_slow_answer_inc xc_slow_answer++
+
+ #else /* XIDCACHE_DEBUG */
+
+ #define xc_by_recent_xmin_inc
+ #define xc_by_main_xid_inc
+ #define xc_by_child_xid_inc
+ #define xc_slow_answer_inc
+ #endif /* XIDCACHE_DEBUG */
/*
* Because backends sitting idle will not be reading sinval events, we
***************
*** 80,85 ****
--- 101,110 ----
ereport(FATAL,
(errcode(ERRCODE_TOO_MANY_CONNECTIONS),
errmsg("sorry, too many clients already")));
+
+ #ifdef XIDCACHE_DEBUG
+ on_proc_exit(DisplayXidCache, (Datum) NULL);
+ #endif /* XIDCACHE_DEBUG */
}
/*
***************
*** 444,451 ****
*
* SInvalLock has to be held while we do 1 and 2. If we save all the Xids
* while doing 1, we can release the SInvalLock while we do 3. This buys back
! * some concurrency (we can't retrieve the main Xids from PGPROC again anyway,
! * see GetNewTransactionId)
*/
bool
TransactionIdIsInProgress(TransactionId xid)
--- 469,476 ----
*
* SInvalLock has to be held while we do 1 and 2. If we save all the Xids
* while doing 1, we can release the SInvalLock while we do 3. This buys back
! * some concurrency (we can't retrieve the main Xids from PGPROC again anyway;
! * see GetNewTransactionId).
*/
bool
TransactionIdIsInProgress(TransactionId xid)
***************
*** 453,465 ****
bool result = false;
SISeg *segP = shmInvalBuffer;
ProcState *stateP = segP->procState;
! int i;
! int nxids = 0;
TransactionId *xids;
! xids = (TransactionId *)palloc(sizeof(TransactionId) * segP->maxBackends);
LWLockAcquire(SInvalLock, LW_SHARED);
for (i = 0; i < segP->lastBackend; i++)
{
--- 478,502 ----
bool result = false;
SISeg *segP = shmInvalBuffer;
ProcState *stateP = segP->procState;
! int i,
! j;
TransactionId *xids;
+ bool locked;
+ bool overflowed = false;
+
+ /*
+ * Don't bother checking a very old transaction.
+ */
+ if (TransactionIdPrecedes(xid, RecentGlobalXmin))
+ {
+ xc_by_recent_xmin_inc;
+ return false;
+ }
! xids = (TransactionId *) palloc(sizeof(TransactionId) * segP->maxBackends);
LWLockAcquire(SInvalLock, LW_SHARED);
+ locked = true;
for (i = 0; i < segP->lastBackend; i++)
{
***************
*** 473,545 ****
TransactionId pxid = proc->xid;
/*
! * check the main Xid (step 1 above)
*/
if (TransactionIdEquals(pxid, xid))
{
result = true;
! break;
}
! /*
! * save the main Xid for step 3.
! */
! xids[nxids++] = pxid;
! #ifdef NOT_USED
! FIXME -- waiting to save the Xids in PGPROC ...
/*
! * check the saved Xids array (step 2)
*/
! for (j = 0; j < PGPROC_MAX_SAVED_XIDS; j++)
{
! pxid = proc->savedxids[j];
! if (!TransactionIdIsValid(pxids))
! break;
if (TransactionIdEquals(pxid, xid))
{
result = true;
! break;
}
}
- #endif
-
- if (result)
- break;
}
}
LWLockRelease(SInvalLock);
/*
* Step 3: have to check pg_subtrans. Use the saved Xids.
*
! * XXX Could save the cached Xids too for further improvement.
*/
! if (!result)
{
! /* this is a potentially expensive call. */
! xid = SubTransGetTopmostTransaction(xid);
!
! Assert(TransactionIdIsValid(xid));
! /*
! * We don't care if it aborted, because if it did, we won't find
! * it in the array.
! */
! for (i = 0; i < nxids; i++)
{
! if (TransactionIdEquals(xids[i], xid))
! {
! result = true;
! break;
! }
}
}
pfree(xids);
return result;
--- 510,629 ----
TransactionId pxid = proc->xid;
/*
! * Step 1: check the main Xid
*/
if (TransactionIdEquals(pxid, xid))
{
+ xc_by_main_xid_inc;
result = true;
! goto result_known;
}
! /* save the main Xid for step 3. */
! xids[i] = pxid;
! if (proc->cache.overflow)
! overflowed = true;
/*
! * Step 2: check the cached Xids arrays
*/
! for (j = 0; j < PGPROC_MAX_CACHED_SUBXIDS; j++)
{
! pxid = proc->cache.xids[j];
! if (!TransactionIdIsValid(pxid))
! continue;
if (TransactionIdEquals(pxid, xid))
{
+ xc_by_child_xid_inc;
result = true;
! goto result_known;
}
}
}
}
LWLockRelease(SInvalLock);
+ locked = false;
+
+ /*
+ * If none of the caches overflowed, we know the Xid is
+ * not running without looking at pg_subtrans.
+ */
+ if (!overflowed)
+ goto result_known;
/*
* Step 3: have to check pg_subtrans. Use the saved Xids.
+ */
+ xc_slow_answer_inc;
+
+ /*
+ * At this point, we know it's either a subtransaction or
+ * it's not running. If it's a subtransaction, we have to
+ * check whether it's part of a running subtransaction tree
+ * or it was aborted. So we have to look at pg_clog, but
+ * since we already checked the PGPROC array we don't have to
+ * worry about a race condition.
+ */
+ if (TransactionIdDidAbort(xid))
+ {
+ result = false;
+ goto result_known;
+ }
+
+ /*
+ * It isn't aborted, so check whether the transaction tree it
+ * belongs to is still running (or, more precisely, whether it
+ * was running when this routine started -- note that we just
+ * released SInvalLock.)
+ */
+ xid = SubTransGetTopmostTransaction(xid);
+ Assert(TransactionIdIsValid(xid));
+
+ for (i = 0; i < segP->maxBackends; i++)
+ {
+ if (TransactionIdEquals(xids[i], xid))
+ {
+ result = true;
+ break;
+ }
+ }
+
+ /*
+ * pg_subtrans says it's running in the i-nd backend. Check if it's
+ * still true.
*
! * This is strictly not needed, but 1) an eternity has passed since
! * the main Xid was read (we took a peek at both pg_clog and pg_subtrans
! * in the meantime), and 2) it's a cheap test.
*/
! if (result)
{
! LWLockAcquire(SInvalLock, LW_SHARED);
! locked = true;
! SHMEM_OFFSET pOffset = stateP[i].procStruct;
!
! if (pOffset != INVALID_OFFSET)
{
! PGPROC *proc = (PGPROC *) MAKE_PTR(pOffset);
!
! /* Fetch xid just once - see GetNewTransactionId */
! TransactionId pxid = proc->xid;
!
! if (!TransactionIdEquals(pxid, xid))
! result = false;
}
}
+ result_known: ;
+
+ if (locked)
+ LWLockRelease(SInvalLock);
+
pfree(xids);
return result;
***************
*** 794,799 ****
--- 878,885 ----
snapshot->curcid = GetCurrentCommandId();
+ DisplayXidCache(0, 0);
+
return snapshot;
}
***************
*** 928,930 ****
--- 1014,1128 ----
return count;
}
+
+ /*
+ * XidCacheAddRunningXid
+ *
+ * Add a TransactionId to the list of known-running transactions.
+ * If there is no space in the cache, mark overflow and return.
+ */
+ void
+ XidCacheAddRunningXid(TransactionId xid)
+ {
+ LWLockAcquire(SInvalLock, LW_SHARED);
+
+ if (MyProc->cache.nxids >= PGPROC_MAX_CACHED_SUBXIDS)
+ {
+ MyProc->cache.overflow = true;
+ LWLockRelease(SInvalLock);
+ return;
+ }
+
+ TransactionIdStore(xid, &(MyProc->cache.xids[MyProc->cache.nxids]));
+ MyProc->cache.nxids ++;
+ LWLockRelease(SInvalLock);
+ }
+
+ #define XidCacheRemove(i) \
+ do { \
+ TransactionIdStore(InvalidTransactionId, &(MyProc->cache.xids[i])); \
+ MyProc->cache.nxids --; \
+ } while (0);
+
+ /*
+ * XidCacheRemoveRunningXids
+ *
+ * Remove a bunch of TransactionIds from the list of known-running
+ * transactions.
+ */
+ void
+ XidCacheRemoveRunningXids(int nxids, TransactionId *xids, TransactionId xid)
+ {
+ int i, j;
+
+ Assert(!TransactionIdEquals(xid, InvalidTransactionId));
+
+ LWLockAcquire(SInvalLock, LW_SHARED);
+
+ for (i = 0; i < nxids; i++)
+ {
+ for (j = 0; j < PGPROC_MAX_CACHED_SUBXIDS; j++)
+ {
+ if (TransactionIdEquals(MyProc->cache.xids[j], xids[i]))
+ {
+ XidCacheRemove(j);
+ break;
+ }
+ }
+ }
+
+ for (j = 0; j < PGPROC_MAX_CACHED_SUBXIDS; j++)
+ {
+ if (TransactionIdEquals(MyProc->cache.xids[j], xid))
+ {
+ XidCacheRemove(j);
+ break;
+ }
+ }
+
+ LWLockRelease(SInvalLock);
+ }
+
+ /*
+ * XidCacheClean
+ *
+ * Fast cache cleanup at transaction end.
+ */
+ void
+ XidCacheClean(void)
+ {
+ LWLockAcquire(SInvalLock, LW_SHARED);
+
+ MyProc->cache.overflow = false;
+ MyProc->cache.nxids = 0;
+ MemSet(MyProc->cache.xids, '\0',
+ PGPROC_MAX_CACHED_SUBXIDS * sizeof(TransactionId));
+
+ LWLockRelease(SInvalLock);
+ }
+
+ #ifdef XIDCACHE_DEBUG
+ static void
+ DisplayXidCache(int code, Datum arg)
+ {
+ int i;
+ if (MyProc == NULL)
+ {
+ fprintf(stderr,"XidCache: xmin: %d, mainxid: %d, childxid: %d, slow:
%d\n",
+ xc_by_recent_xmin,
+ xc_by_main_xid,
+ xc_by_child_xid,
+ xc_slow_answer);
+ }
+ else
+ {
+ fprintf(stderr, "(%s) children:\t",
+ MyProc->cache.overflow ? "overf" : "no overf");
+ for (i = 0; i < PGPROC_MAX_CACHED_SUBXIDS; i++)
+ {
+ fprintf(stderr, "%d ", MyProc->cache.xids[i]);
+ }
+ fprintf(stderr, "\n");
+ }
+ }
+ #endif /* XIDCACHE_DEBUG */
diff -Ncr --exclude-from=diff-ignore 00orig/src/backend/tcop/utility.c
04pgproc/src/backend/tcop/utility.c
*** 00orig/src/backend/tcop/utility.c 2004-07-26 21:59:36.000000000 -0400
--- 04pgproc/src/backend/tcop/utility.c 2004-07-27 10:29:30.000000000 -0400
***************
*** 326,333 ****
{
/*
* START TRANSACTION, as defined by SQL99:
! * Identical to BEGIN, except that it takes a
few
! * additional options. Same code for both.
*/
case TRANS_STMT_BEGIN:
case TRANS_STMT_START:
--- 326,332 ----
{
/*
* START TRANSACTION, as defined by SQL99:
! * Identical to BEGIN. Same code for both.
*/
case TRANS_STMT_BEGIN:
case TRANS_STMT_START:
diff -Ncr --exclude-from=diff-ignore 00orig/src/include/access/htup.h
04pgproc/src/include/access/htup.h
*** 00orig/src/include/access/htup.h 2004-07-17 18:10:20.000000000 -0400
--- 04pgproc/src/include/access/htup.h 2004-07-27 10:52:44.000000000 -0400
***************
*** 68,101 ****
* object ID (if HEAP_HASOID is set in t_infomask)
* user data fields
*
! * We store five "virtual" fields Xmin, Cmin, Xmax, Cmax, and Xvac
! * in just three physical fields. Xmin is always really stored, but
! * Cmin and Xmax share a field, as do Cmax and Xvac. This works because
! * we know that there are only a limited number of states that a tuple can
! * be in, and that Cmin and Cmax are only interesting for the lifetime of
! * the inserting and deleting transactions respectively. We have the
! * following possible states of a tuple:
! *
! * XMIN CMIN XMAX CMAX XVAC
! *
! * NEW (never deleted, not moved by vacuum):
! * valid valid invalid invalid invalid
! *
! * DELETED BY CREATING XACT:
! * valid valid = XMIN valid invalid
! *
! * DELETED BY OTHER XACT:
! * valid unneeded valid valid invalid
! *
! * MOVED BY VACUUM FULL:
! * valid unneeded maybe-valid unneeded valid
! *
! * This assumes that VACUUM FULL never tries to move a tuple whose Cmin or
! * Cmax is still interesting (ie, insert-in-progress or delete-in-progress).
! *
! * This table shows that if we use an infomask bit to handle the case
! * XMAX=XMIN specially, we never need to store Cmin and Xmax at the same
! * time. Nor do we need to store Cmax and Xvac at the same time.
*
* Following the fixed header fields, the nulls bitmap is stored (beginning
* at t_bits). The bitmap is *not* stored if t_infomask shows that there
--- 68,84 ----
* object ID (if HEAP_HASOID is set in t_infomask)
* user data fields
*
! * We store five "virtual" fields Xmin, Cmin, Xmax, Cmax, and Xvac in four
! * physical fields. Xmin, Cmin and Xmax are always really stored, but
! * Cmax and Xvac share a field. This works because we know that there are
! * only a limited number of states that a tuple can be in, and that Cmax
! * is only interesting for the lifetime of the deleting transactions
! * respectively. This assumes that VACUUM FULL never tries to move a
! * tuple whose Cmax is still interesting (ie, delete-in-progress).
! *
! * Note that in 7.3 and 7.4 a similar idea was applied to Xmax and Cmin.
! * However, with the advent of subtransactions, a tuple may need both Xmax
! * and Cmin simultaneously, so this is no longer possible.
*
* Following the fixed header fields, the nulls bitmap is stored (beginning
* at t_bits). The bitmap is *not* stored if t_infomask shows that there
***************
*** 424,430 ****
#define XLOG_HEAP_MOVE 0x30
#define XLOG_HEAP_CLEAN 0x40
#define XLOG_HEAP_NEWPAGE 0x50
! /* opcodes 0x60, 0x70 still free */
#define XLOG_HEAP_OPMASK 0x70
/*
* When we insert 1st item on new page in INSERT/UPDATE
--- 407,413 ----
#define XLOG_HEAP_MOVE 0x30
#define XLOG_HEAP_CLEAN 0x40
#define XLOG_HEAP_NEWPAGE 0x50
! /* opcode 0x60 still free */
#define XLOG_HEAP_OPMASK 0x70
/*
* When we insert 1st item on new page in INSERT/UPDATE
diff -Ncr --exclude-from=diff-ignore 00orig/src/include/storage/proc.h
04pgproc/src/include/storage/proc.h
*** 00orig/src/include/storage/proc.h 2004-07-26 21:59:44.000000000 -0400
--- 04pgproc/src/include/storage/proc.h 2004-07-28 14:41:03.827806803 -0400
***************
*** 19,24 ****
--- 19,44 ----
#include "storage/lock.h"
#include "storage/pg_sema.h"
+ /*
+ * XXX This number is made up ...
+ */
+ #define PGPROC_MAX_CACHED_SUBXIDS 32
+
+ /*
+ * Each backend keeps track of (some of) its subtransactions'
+ * TransactionIds in the PGPROC struct.
+ *
+ * We also keep track of whether the cache overflowed. If it
+ * hasn't overflowed, we can assume that a Xid that's not present
+ * in the cache is not a running transaction. Else we have to look
+ * at pg_subtrans.
+ */
+ struct XidCache {
+ /* running Xids cache */
+ int nxids;
+ TransactionId xids[PGPROC_MAX_CACHED_SUBXIDS];
+ bool overflow;
+ };
/*
* Each backend has a PGPROC struct in shared memory. There is also a list of
***************
*** 39,44 ****
--- 59,66 ----
TransactionId xid; /* transaction currently being
executed by
* this proc */
+ struct XidCache cache; /* Xid cache */
+
TransactionId xmin; /* minimal running XID as it was when
we
* were starting our
xact: vacuum must not
* remove tuples
deleted by xid >= xmin ! */
diff -Ncr --exclude-from=diff-ignore 00orig/src/include/storage/sinval.h
04pgproc/src/include/storage/sinval.h
*** 00orig/src/include/storage/sinval.h 2004-06-03 15:59:03.000000000 -0400
--- 04pgproc/src/include/storage/sinval.h 2004-07-27 23:30:29.000000000 -0400
***************
*** 115,118 ****
--- 115,123 ----
extern void EnableCatchupInterrupt(void);
extern bool DisableCatchupInterrupt(void);
+ /* Xid cache updaters */
+ extern void XidCacheAddRunningXid(TransactionId xid);
+ extern void XidCacheRemoveRunningXids(int nxids, TransactionId *xids, TransactionId
xid);
+ extern void XidCacheClean(void);
+
#endif /* SINVAL_H */
---------------------------(end of broadcast)---------------------------
TIP 9: the planner will ignore your desire to choose an index scan if your
joining column's datatypes do not match
