* Add $wgLegacySchemaConversion update-time option to reduce amount of

  copying during the schema upgrade: creates HistoryBlobCurStub reference
  records in text instead of copying all the cur_text fields. Requires
  that the cur table be left in place until/unless such fields are migrated
  into the main text store.

In my test with a nl.wikipedia.org dump, this sped up the cur-to-old copy
operation from a few hours to a few minutes. Also added some hacky timer
echos in the updater for debug purposes. They should be prettied up but
probably kept.

diff --git a/RELEASE-NOTES b/RELEASE-NOTES
index bd8bfc4..2dde358 100644 (file)
--- a/RELEASE-NOTES
+++ b/RELEASE-NOTES
@@ -166,7 +166,11 @@ Various bugfixes, small features, and a few experimental things:
 * Fix for reading incorrectly re-gzipped HistoryBlob entries
 * (bug 1906) Generalize project namespace for Latin localization, update namespaces
 * (bug 2075) Corrected namespace definitions in Tamil localization
-
+* Add $wgLegacySchemaConversion update-time option to reduce amount of
+  copying during the schema upgrade: creates HistoryBlobCurStub reference
+  records in text instead of copying all the cur_text fields. Requires
+  that the cur table be left in place until/unless such fields are migrated
+  into the main text store.
 
 === Caveats ===
 

diff --git a/includes/DefaultSettings.php b/includes/DefaultSettings.php
index 587870c..6b3c96c 100644 (file)
--- a/includes/DefaultSettings.php
+++ b/includes/DefaultSettings.php
@@ -373,6 +373,23 @@ $wgEditEncoding            = '';
 # en masse in the database before continuing as a UTF-8 wiki.
 $wgLegacyEncoding   = false;
 
+/**
+ * If set to true, the MediaWiki 1.4 to 1.5 schema conversion will
+ * create stub reference rows in the text table instead of copying
+ * the full text of all current entries from 'cur' to 'text'.
+ *
+ * This will speed up the conversion step for large sites, but
+ * requires that the cur table be kept around for those revisions
+ * to remain viewable.
+ *
+ * maintenance/migrateCurStubs.php can be used to complete the
+ * migration in the background once the wiki is back online.
+ *
+ * This option affects the updaters *only*. Any present cur stub
+ * revisions will be readable at runtime regardless of this setting.
+ */
+$wgLegacySchemaConversion = false;
+
 $wgMimeType                    = 'text/html';
 $wgJsMimeType                  = 'text/javascript';
 $wgDocType                     = '-//W3C//DTD XHTML 1.0 Transitional//EN';

diff --git a/includes/HistoryBlob.php b/includes/HistoryBlob.php
index 2fe5d1f..5b1f453 100644 (file)
--- a/includes/HistoryBlob.php
+++ b/includes/HistoryBlob.php
@@ -218,4 +218,43 @@ class HistoryBlobStub {
                return $this->mHash;
        }
 }
+
+
+/**
+ * To speed up conversion from 1.4 to 1.5 schema, text rows can refer to the
+ * leftover cur table as the backend. This avoids expensively copying hundreds
+ * of megabytes of data during the conversion downtime.
+ *
+ * Serialized HistoryBlobCurStub objects will be inserted into the text table
+ * on conversion if $wgFastSchemaUpgrades is set to true.
+ *
+ * @package MediaWiki
+ */
+class HistoryBlobCurStub {
+       var $mCurId;
+
+       /** @todo document */
+       function HistoryBlobCurStub( $curid = 0 ) {
+               $this->mCurId = $curid;
+       }
+       
+       /**
+        * Sets the location (cur_id) of the main object to which this object
+        * points
+        */
+       function setLocation( $id ) {
+               $this->mCurId = $id;
+       }
+
+       /** @todo document */
+       function getText() {
+               $dbr =& wfGetDB( DB_SLAVE );
+               $row = $dbr->selectRow( 'cur', array( 'cur_text' ), array( 'cur_id' => $this->mCurId ) );
+               if( !$row ) {
+                       return false;
+               }
+               return $row->cur_text;
+       }
+}
+
 ?>

diff --git a/includes/Parser.php b/includes/Parser.php
index f4a8896..c24cb82 100644 (file)
--- a/includes/Parser.php
+++ b/includes/Parser.php
@@ -3081,6 +3081,7 @@ class Parser
                                # remember to set an alignment, don't render immediately
                                $align = 'none';
                        } elseif ( $wgUseImageResize && ! is_null( $match = $mwWidth->matchVariableStartToEnd($val) ) ) {
+                               wfDebug( "MAG_IMG_WIDTH match: $match\n" );
                                # $match is the image width in pixels
                                if ( preg_match( '/^([0-9]*)x([0-9]*)$/', $match, $m ) ) {
                                        $width = intval( $m[1] );

diff --git a/maintenance/updaters.inc b/maintenance/updaters.inc
index 01c6a68..6a0f24e 100644 (file)
--- a/maintenance/updaters.inc
+++ b/maintenance/updaters.inc
@@ -282,6 +282,7 @@ function do_schema_restructuring() {
                echo "...page table already exists.\n";
        } else {
                echo "...converting from cur/old to page/revision/text DB structure.\n"; flush();
+               echo wfTimestamp();
                echo "......checking for duplicate entries.\n"; flush();
                
                extract( $wgDatabase->tableNames( 'cur', 'old', 'page', 'revision', 'text' ) );
@@ -290,6 +291,7 @@ function do_schema_restructuring() {
                                FROM $cur GROUP BY cur_title, cur_namespace HAVING c>1", $fname );
 
                if ( $wgDatabase->numRows( $rows ) > 0 ) {
+                       echo wfTimestamp();
                        echo "......<b>Found duplicate entries</b>\n";
                        echo ( sprintf( "<b>      %-60s %3s %5s</b>\n", 'Title', 'NS', 'Count' ) );
                        while ( $row = $wgDatabase->fetchObject( $rows ) ) {
@@ -337,10 +339,12 @@ function do_schema_restructuring() {
                        }
                        $sql = "DELETE FROM $cur WHERE cur_id IN ( " . join( ',', $deleteId ) . ')';
                        $rows = $wgDatabase->query( $sql, $fname );
+                       echo wfTimestamp();
                        echo "......<b>Deleted</b> ".$wgDatabase->affectedRows()." records.\n";
                }
                
 
+               echo wfTimestamp();
                echo "......Creating tables.\n";
                $wgDatabase->query("CREATE TABLE $page (
                        page_id int(8) unsigned NOT NULL auto_increment,
@@ -378,18 +382,35 @@ function do_schema_restructuring() {
                        INDEX usertext_timestamp (rev_user_text,rev_timestamp)
                        ) TYPE=InnoDB", $fname );
 
+               echo wfTimestamp();
                echo "......Locking tables.\n";
                $wgDatabase->query( "LOCK TABLES $page WRITE, $revision WRITE, $old WRITE, $cur WRITE", $fname );
 
                $maxold = $wgDatabase->selectField( 'old', 'max(old_id)', '', $fname );
+               echo wfTimestamp();
                echo "......maxold is {$maxold}\n";
 
-               echo "......Moving text from cur.\n";
+               echo wfTimestamp();
+               global $wgLegacySchemaConversion;
+               if( $wgLegacySchemaConversion ) {
+                       // Create HistoryBlobCurStub entries.
+                       // Text will be pulled from the leftover 'cur' table at runtime.
+                       echo "......Moving metadata from cur; using blob references to text in cur table.\n";
+                       $cur_text = "concat('O:18:\"historyblobcurstub\":1:{s:6:\"mCurId\";i:',cur_id,';}')";
+                       $cur_flags = "'object'";
+               } else {
+                       // Copy all cur text in immediately: this may take longer but avoids
+                       // having to keep an extra table around.
+                       echo "......Moving text from cur.\n";
+                       $cur_text = 'cur_text';
+                       $cur_flags = "''";
+               }
                $wgDatabase->query( "INSERT INTO $old (old_namespace, old_title, old_text, old_comment, old_user, old_user_text,
                                old_timestamp, old_minor_edit, old_flags)
-                       SELECT cur_namespace, cur_title, cur_text, cur_comment, cur_user, cur_user_text, cur_timestamp, cur_minor_edit,''
+                       SELECT cur_namespace, cur_title, $cur_text, cur_comment, cur_user, cur_user_text, cur_timestamp, cur_minor_edit, $cur_flags
                        FROM $cur", $fname );
 
+               echo wfTimestamp();
                echo "......Setting up revision table.\n";
                $wgDatabase->query( "INSERT INTO $revision (rev_id, rev_page, rev_comment, rev_user, rev_user_text, rev_timestamp,
                                rev_minor_edit)
@@ -397,6 +418,7 @@ function do_schema_restructuring() {
                                old_timestamp, old_minor_edit
                        FROM $old,$cur WHERE old_namespace=cur_namespace AND old_title=cur_title", $fname );
 
+               echo wfTimestamp();
                echo "......Setting up page table.\n";
                $wgDatabase->query( "INSERT INTO $page (page_id, page_namespace, page_title, page_restrictions, page_counter,
                                page_is_redirect, page_is_new, page_random, page_touched, page_latest, page_len)
@@ -405,11 +427,15 @@ function do_schema_restructuring() {
                        FROM $cur,$revision
                        WHERE cur_id=rev_page AND rev_timestamp=cur_timestamp AND rev_id > {$maxold}", $fname );
 
+               echo wfTimestamp();
                echo "......Unlocking tables.\n";
                $wgDatabase->query( "UNLOCK TABLES", $fname );
 
+               echo wfTimestamp();
                echo "......Renaming old.\n";
                $wgDatabase->query( "ALTER TABLE $old RENAME TO $text", $fname );
+               
+               echo wfTimestamp();
                echo "...done.\n";
        }
 }