Brian Wolff has uploaded a new change for review.
https://gerrit.wikimedia.org/r/67468
Change subject: Make Special:MIMESearch a non-expensive special page.
......................................................................
Make Special:MIMESearch a non-expensive special page.
This tries to use the index on (img_media_type, img_major_mime, img_minor_mime).
This is not the ideal index for this purpose (due to starting with
img_media_type, and not having an img_name at the end). However it
seems workable. It does a couple extra queries (at worst 9, one
for each media type defined. However the common case of a bitmap
image, it doesn't need any extra queries. These extra queries should
be cheap indexed 0 result queries). Additionally, even though the
index doesn't have img_name on the end, mysql seems to return
results in img_name ASC order, which is nice (Even if it didn't,
I still think this change would be worth it).
I've does DESCRIBE on all the queries, and they look good to me.
The only potentially problematic thing I see is that it uses
LIMIT X,Y for the offsets (due to being a query page) which can
be expensive on big offsets. If that's an issue, we could limit
the number of results to be at most $wgQueryCacheLimit or something.
Note: There's some discussion on the bug rather negative to this
special page in general. While I agree with much of the criticism,
I still think this special page can be useful, and if it can be
made efficient without needing to modify the indicies, we might
as well.
Bug: 13438
Change-Id: I88826ae7bff0812374157e596f86a19b248220e1
---
M RELEASE-NOTES-1.22
M includes/QueryPage.php
M includes/specials/SpecialMIMEsearch.php
3 files changed, 62 insertions(+), 5 deletions(-)
git pull ssh://gerrit.wikimedia.org:29418/mediawiki/core
refs/changes/68/67468/1
diff --git a/RELEASE-NOTES-1.22 b/RELEASE-NOTES-1.22
index 07763d9..327684c 100644
--- a/RELEASE-NOTES-1.22
+++ b/RELEASE-NOTES-1.22
@@ -132,6 +132,7 @@
* (bug 48294) API: Fix chunk upload async mode.
* (bug 46749) Broken files tracking category removed from pages if an image
with that name is uploaded.
+* (bug 13438) Special:MIMESearch no longer an expensive special page.
=== API changes in 1.22 ===
* (bug 46626) xmldoublequote parameter was removed. Because of a bug, the
diff --git a/includes/QueryPage.php b/includes/QueryPage.php
index a93639a..e96c7bd 100644
--- a/includes/QueryPage.php
+++ b/includes/QueryPage.php
@@ -588,7 +588,7 @@
# $res might contain the whole 1,000 rows, so we read
up to
# $num [should update this to use a Pager]
- for ( $i = 0; $i < $num && $row = $dbr->fetchObject(
$res ); $i++ ) {
+ for ( $i = 0; $i < $num && $row = $res->fetchObject();
$i++ ) {
$line = $this->formatResult( $skin, $row );
if ( $line ) {
$attr = ( isset( $row->usepatrol ) &&
$row->usepatrol && $row->patrolled == 0 )
diff --git a/includes/specials/SpecialMIMEsearch.php
b/includes/specials/SpecialMIMEsearch.php
index 7ff5d7b..4a4a4e1 100644
--- a/includes/specials/SpecialMIMEsearch.php
+++ b/includes/specials/SpecialMIMEsearch.php
@@ -28,14 +28,14 @@
* @ingroup SpecialPage
*/
class MIMEsearchPage extends QueryPage {
- protected $major, $minor;
+ protected $major, $minor, $mediaType = false;
function __construct( $name = 'MIMEsearch' ) {
parent::__construct( $name );
}
function isExpensive() {
- return true;
+ return false;
}
function isSyndicated() {
@@ -51,12 +51,14 @@
}
public function getQueryInfo() {
- return array(
+ $qi = array(
'tables' => array( 'image' ),
'fields' => array(
'namespace' => NS_FILE,
'title' => 'img_name',
- 'value' => 'img_major_mime',
+ // Still have a value field just in case,
+ // but it isn't actually used for sorting.
+ 'value' => 'img_name',
'img_size',
'img_width',
'img_height',
@@ -68,6 +70,60 @@
'img_minor_mime' => $this->minor
)
);
+ if ( $this->mediaType ) {
+ $qi['conds']['img_media_type'] = $this->mediaType;
+ }
+ return $qi;
+ }
+
+ /**
+ * The index is on (img_media_type, img_major_mime, img_minor_mime)
+ * which unfortunately doesn't have img_name at the end for sorting.
+ * So tell db to sort it however it wishes (Its not super important
+ * that this report gives results in a logical order). As an aditional
+ * note, mysql seems to by default order things by img_name ASC, which
+ * is what we ideally want, so everything works out fine anyhow.
+ */
+ function getOrderFields() {
+ return array();
+ }
+
+ /**
+ * Work around the index structure of the db, where we
+ * have to first search for img_media_type. Luckily there
+ * is only a limitted number of such types, so we can
+ * just do a bunch of indexed queries (most of which would
+ * be quick 0 results).
+ */
+ function reallyDoQuery( $limit, $offset = false ) {
+ $mediaTypes = array(
+ MEDIATYPE_BITMAP,
+ MEDIATYPE_DRAWING,
+ MEDIATYPE_AUDIO,
+ MEDIATYPE_VIDEO,
+ MEDIATYPE_MULTIMEDIA,
+ MEDIATYPE_UNKNOWN,
+ MEDIATYPE_OFFICE,
+ MEDIATYPE_TEXT,
+ MEDIATYPE_EXECUTABLE,
+ MEDIATYPE_ARCHIVE,
+ );
+ $results = array();
+ $curLimit = $limit; // Number of results left to fetch
+
+ foreach( $mediaTypes as $curMediaType ) {
+ $this->mediaType = $curMediaType;
+ $tmpResult = parent::reallyDoQuery( $curLimit, $offset
);
+ foreach( $tmpResult as $row ) {
+ $results[] = $row;
+ $curLimit--;
+ }
+ if ( $curLimit <= 0 ) {
+ break;
+ }
+ }
+ $this->mediaType = false;
+ return new FakeResultWrapper( $results );
}
function execute( $par ) {
--
To view, visit https://gerrit.wikimedia.org/r/67468
To unsubscribe, visit https://gerrit.wikimedia.org/r/settings
Gerrit-MessageType: newchange
Gerrit-Change-Id: I88826ae7bff0812374157e596f86a19b248220e1
Gerrit-PatchSet: 1
Gerrit-Project: mediawiki/core
Gerrit-Branch: master
Gerrit-Owner: Brian Wolff <[email protected]>
_______________________________________________
MediaWiki-commits mailing list
[email protected]
https://lists.wikimedia.org/mailman/listinfo/mediawiki-commits
