[Rt-commit] rt branch, 4.4/fts-refactor-performance, created. rt-4.2.9-94-gdff4059
Alex Vandiver
alexmv at bestpractical.com
Mon Dec 22 18:36:54 EST 2014
The branch, 4.4/fts-refactor-performance has been created
at dff4059ab03803ee0990573ca9397b034ccc998a (commit)
- Log -----------------------------------------------------------------
commit ed183c86f62042d97c29c4755caf7e7658ef9040
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 00:17:32 2014 -0400
extract_text and extract_html are identical; inline them
The original premise may have been that HTML->text conversion would be
fone on the HTML before indexing. While this is still an option for the
future, there is currently no reason to provide to identical methods.
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index 6329c00..9a7b0af 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -183,9 +183,10 @@ foreach my $type ( @types ) {
while ( my $a = $attachments->Next ) {
next if filter( $type, $a );
debug("Found attachment #". $a->id );
- my $txt = extract($type, $a) or next;
+ my $text = $a->Content;
+ next unless defined $text && length($text);
$found++;
- process( $type, $a, $txt );
+ process( $type, $a, \$text );
debug("Processed attachment #". $a->id );
}
goto REDO if $OPT{'all'} and $attachments->Count == ($OPT{'limit'} || 100)
@@ -242,15 +243,6 @@ sub filter {
);
}
-sub extract {
- my $type = shift;
- return goto_specific(
- suffix => $type,
- error => "No way to convert $type attachment into text",
- arguments => \@_,
- );
-}
-
sub process {
return goto_specific(
suffix => $db_type,
@@ -342,13 +334,6 @@ sub attachments_text {
return $res;
}
-sub extract_text {
- my $attachment = shift;
- my $text = $attachment->Content;
- return undef unless defined $text && length($text);
- return \$text;
-}
-
sub attachments_html {
my $res = shift;
$res->Limit( FIELD => 'ContentType', VALUE => 'text/html' );
@@ -365,14 +350,6 @@ sub filter_html {
return 0;
}
-sub extract_html {
- my $attachment = shift;
- my $text = $attachment->Content;
- return undef unless defined $text && length($text);
-# TODO: html -> text
- return \$text;
-}
-
sub goto_specific {
my %args = (@_);
commit 0d40c4336f140c238fef58cb5cf570c587e88229
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 00:19:10 2014 -0400
Inline the differences between text/plain and text/html attachment lists
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index 9a7b0af..7bfa2e3 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -218,12 +218,11 @@ sub attachments {
OPERATOR => '!=',
VALUE => 'deleted'
);
-
- return goto_specific(
- suffix => $type,
- error => "Don't know how to find $type attachments",
- arguments => [$res],
+ $res->Limit(
+ FIELD => 'ContentType',
+ VALUE => ($type eq "html" ? "text/html" : "text/plain"),
);
+ return $res;
}
sub last_indexed {
@@ -328,18 +327,6 @@ sub process_pg {
}
}
-sub attachments_text {
- my $res = shift;
- $res->Limit( FIELD => 'ContentType', VALUE => 'text/plain' );
- return $res;
-}
-
-sub attachments_html {
- my $res = shift;
- $res->Limit( FIELD => 'ContentType', VALUE => 'text/html' );
- return $res;
-}
-
sub filter_html {
my $attachment = shift;
if ( my $parent = $attachment->ParentObj ) {
commit 86e69b61bc4834c669e757d043e88e9c401598ac
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 00:19:55 2014 -0400
Stop skipping indexing of text/html within multipart/alternative
Not all text/html within a multipart/alternative is strictly identical
tothe text part; some mailers send out text/plain parts of alternatives
that simply say "Your client must be able to render HTML."
The additional space required to index these attachments is worth the
consitency of having all potential content indexed. This also is a
speed increase, as it reduces the number of ad-hoc queries during
indexing.
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index 7bfa2e3..b7adaec 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -181,7 +181,6 @@ foreach my $type ( @types ) {
my $found = 0;
while ( my $a = $attachments->Next ) {
- next if filter( $type, $a );
debug("Found attachment #". $a->id );
my $text = $a->Content;
next unless defined $text && length($text);
@@ -234,14 +233,6 @@ sub last_indexed {
);
}
-sub filter {
- my $type = shift;
- return goto_specific(
- suffix => $type,
- arguments => \@_,
- );
-}
-
sub process {
return goto_specific(
suffix => $db_type,
@@ -327,16 +318,6 @@ sub process_pg {
}
}
-sub filter_html {
- my $attachment = shift;
- if ( my $parent = $attachment->ParentObj ) {
-# skip html parts that are alternatives
- return 1 if $parent->id
- && $parent->ContentType eq 'mulitpart/alternative';
- }
- return 0;
-}
-
sub goto_specific {
my %args = (@_);
commit d1076aa24132e825cecf231238481aac5e6a5079
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 00:21:52 2014 -0400
Use the new, shorter, initialization form
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index b7adaec..1df6514 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -67,11 +67,7 @@ BEGIN { # BEGIN RT CMD BOILERPLATE
}
-BEGIN {
- use RT;
- RT::LoadConfig();
- RT::Init();
-};
+use RT -init;
use RT::Interface::CLI ();
my %OPT = (
commit 133c2b887a97f6b9f2e164f7920ccf610f2dcdba
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 00:23:40 2014 -0400
Simplify and condense option parsing
This causes the indexer to be able to accept options that are
potentially not applicable (for instance, --memory with a non-Oracle
backend); however, the comprehensibility benefits are worth this tiny
cost.
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index 1df6514..4fb84c1 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -70,40 +70,18 @@ BEGIN { # BEGIN RT CMD BOILERPLATE
use RT -init;
use RT::Interface::CLI ();
-my %OPT = (
- help => 0,
- debug => 0,
- quiet => 0,
-);
-my @OPT_LIST = qw(help|h! debug! quiet);
+use Getopt::Long qw(GetOptions);
+my %OPT = ( memory => '2M', limit => 0 );
+GetOptions( \%OPT,
+ "help|h!",
+ "debug!",
+ "quiet!",
-my $db_type = RT->Config->Get('DatabaseType');
-if ( $db_type eq 'Pg' ) {
- %OPT = (
- %OPT,
- limit => 0,
- all => 0,
- );
- push @OPT_LIST, 'limit=i', 'all!';
-}
-elsif ( $db_type eq 'mysql' ) {
- %OPT = (
- %OPT,
- limit => 0,
- all => 0,
- );
- push @OPT_LIST, 'limit=i', 'all!';
-}
-elsif ( $db_type eq 'Oracle' ) {
- %OPT = (
- %OPT,
- memory => '2M',
- );
- push @OPT_LIST, qw(memory=s);
-}
+ "all!",
+ "limit=i",
-use Getopt::Long qw(GetOptions);
-GetOptions( \%OPT, @OPT_LIST );
+ "memory=s",
+);
if ( $OPT{'help'} ) {
RT::Interface::CLI->ShowHelp(
@@ -123,6 +101,7 @@ if ( !flock main::DATA, LOCK_EX | LOCK_NB ) {
}
}
+my $db_type = RT->Config->Get('DatabaseType');
my $fts_config = RT->Config->Get('FullTextSearch') || {};
unless ( $fts_config->{'Enable'} ) {
print STDERR <<EOT;
commit 16c12970dcfc7873c11c4bef224e4e859330edeb
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 00:24:01 2014 -0400
Documentation has moved out; update --help accordingly
The documentation (which included =head1 sections for each database
type) was moved to the more centralized docs/full_text_indexing.pod in
fa5dffcb; however, --help still looked for the relevant sections.
Remove the special-case code which is no logner relevant.
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index 4fb84c1..a24017a 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -83,11 +83,7 @@ GetOptions( \%OPT,
"memory=s",
);
-if ( $OPT{'help'} ) {
- RT::Interface::CLI->ShowHelp(
- Sections => 'NAME|DESCRIPTION|'. uc($db_type),
- );
-}
+RT::Interface::CLI->ShowHelp if $OPT{help};
use Fcntl ':flock';
if ( !flock main::DATA, LOCK_EX | LOCK_NB ) {
commit bf3f508ea59cc6360e0f8caa989d4d1279feae96
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 01:44:24 2014 -0400
Remove AUTHOR section; it is unnecessary in core sbin files
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index a24017a..3a4aebf 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -319,11 +319,6 @@ This is a helper script to keep full text indexes in sync with data.
Read F<docs/full_text_indexing.pod> for complete details on how and when
to run it.
-=head1 AUTHOR
-
-Ruslan Zakirov E<lt>ruz at bestpractical.comE<gt>,
-Alex Vandiver E<lt>alexmv at bestpractical.comE<gt>
-
=cut
__DATA__
commit 0750cf9bcccf6c670fd127ac54452688fad3827a
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 00:28:27 2014 -0400
Skipping ACL checks yields a sizable performance increase
The indexer spends quite a bit of time checking that the attachment
content is visible to the current user before indexing it -- all of
which is unnecessary, as it is run as the system user. Explicitly
disable the ACL checking for the indexer, which removes a large number
of queries and yields a correspondingly sizable increase in indexing
speed.
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index 3a4aebf..24c2834 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -138,6 +138,15 @@ EOT
exit 1;
}
+# Skip ACL checks. This saves a large number of unnecessary queries
+# (for tickets, ACLs, and users) which are unnecessary, as we are
+# running as the system user.
+{
+ no warnings 'redefine';
+ *RT::Attachment::_Value = \&DBIx::SearchBuilder::Record::_Value;
+ *RT::Attachments::Next = \&DBIx::SearchBuilder::Next;
+}
+
my @types = qw(text html);
foreach my $type ( @types ) {
REDO:
commit a78cba29fad2776eb294e959a7b0afeda01ad04e
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 00:36:49 2014 -0400
Index attachments in one pass through the database, not two
There is no reason to perform two passes through the database instead of
one; doing one allows for better progress estimates, as well as
potentially increasing locality for the database.
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index 24c2834..b50df40 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -147,14 +147,12 @@ EOT
*RT::Attachments::Next = \&DBIx::SearchBuilder::Next;
}
-my @types = qw(text html);
-foreach my $type ( @types ) {
- REDO:
- my $attachments = attachments($type);
+{
+ my $attachments = attachments();
$attachments->Limit(
FIELD => 'id',
OPERATOR => '>',
- VALUE => last_indexed($type)
+ VALUE => last_indexed()
);
$attachments->OrderBy( FIELD => 'id', ORDER => 'asc' );
$attachments->RowsPerPage( $OPT{'limit'} || 100 );
@@ -165,14 +163,13 @@ foreach my $type ( @types ) {
my $text = $a->Content;
next unless defined $text && length($text);
$found++;
- process( $type, $a, \$text );
+ process( $a, \$text );
debug("Processed attachment #". $a->id );
}
- goto REDO if $OPT{'all'} and $attachments->Count == ($OPT{'limit'} || 100)
+ redo if $OPT{'all'} and $attachments->Count == ($OPT{'limit'} || 100);
}
sub attachments {
- my $type = shift;
my $res = RT::Attachments->new( RT->SystemUser );
my $txn_alias = $res->Join(
ALIAS1 => 'main',
@@ -198,17 +195,17 @@ sub attachments {
VALUE => 'deleted'
);
$res->Limit(
- FIELD => 'ContentType',
- VALUE => ($type eq "html" ? "text/html" : "text/plain"),
+ FIELD => 'ContentType',
+ OPERATOR => 'IN',
+ VALUE => ['text/plain', 'text/html'],
);
return $res;
}
sub last_indexed {
- my ($type) = (@_);
return goto_specific(
suffix => $db_type,
- error => "Don't know how to find last indexed $type attachment for $db_type DB",
+ error => "Don't know how to find last indexed attachment for $db_type DB",
arguments => \@_,
);
}
@@ -223,7 +220,7 @@ sub process {
sub last_indexed_mysql { last_indexed_pg(@_); }
sub process_mysql {
- my ($type, $attachment, $text) = (@_);
+ my ($attachment, $text) = (@_);
my $dbh = $RT::Handle->dbh;
my $table = $fts_config->{'Table'};
@@ -239,8 +236,7 @@ sub process_mysql {
}
sub last_indexed_pg {
- my $type = shift;
- my $attachments = attachments( $type );
+ my $attachments = attachments();
my $alias = 'main';
if ( $fts_config->{'Table'} && $fts_config->{'Table'} ne 'Attachments' ) {
$alias = $attachments->Join(
@@ -264,7 +260,7 @@ sub last_indexed_pg {
}
sub process_pg {
- my ($type, $attachment, $text) = (@_);
+ my ($attachment, $text) = (@_);
my $dbh = $RT::Handle->dbh;
my $table = $fts_config->{'Table'};
commit e4583629fff2fd2eaa7402726ceef66e617af2d4
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 00:45:35 2014 -0400
Index attachments even on deleted tickets
Skipping indexing on tickets which are currently deleted may lead to
content which is later not indexed and not findable. Even if a ticket
becomes un-deleted at a later point, it will never again be indexed, as
last_indexed() will limit to new attachments since the last run. The
overhead of joining through Attachments and into Tickets is also not to
be overlooked -- on MySQL, it causes pessimal performance for large
systems, involving more than one filesort and temporary table.
Index all attachments, no matter the source, on the premise that
limiting to non-deleted tickets is better done at query time, not at
index time. It is both more timely, and better limited.
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index b50df40..f869b56 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -148,7 +148,12 @@ EOT
}
{
- my $attachments = attachments();
+ my $attachments = RT::Attachments->new( RT->SystemUser );
+ $attachments->Limit(
+ FIELD => 'ContentType',
+ OPERATOR => 'IN',
+ VALUE => ['text/plain', 'text/html'],
+ );
$attachments->Limit(
FIELD => 'id',
OPERATOR => '>',
@@ -169,39 +174,6 @@ EOT
redo if $OPT{'all'} and $attachments->Count == ($OPT{'limit'} || 100);
}
-sub attachments {
- my $res = RT::Attachments->new( RT->SystemUser );
- my $txn_alias = $res->Join(
- ALIAS1 => 'main',
- FIELD1 => 'TransactionId',
- TABLE2 => 'Transactions',
- FIELD2 => 'id',
- );
- $res->Limit(
- ALIAS => $txn_alias,
- FIELD => 'ObjectType',
- VALUE => 'RT::Ticket',
- );
- my $ticket_alias = $res->Join(
- ALIAS1 => $txn_alias,
- FIELD1 => 'ObjectId',
- TABLE2 => 'Tickets',
- FIELD2 => 'id',
- );
- $res->Limit(
- ALIAS => $ticket_alias,
- FIELD => 'Status',
- OPERATOR => '!=',
- VALUE => 'deleted'
- );
- $res->Limit(
- FIELD => 'ContentType',
- OPERATOR => 'IN',
- VALUE => ['text/plain', 'text/html'],
- );
- return $res;
-}
-
sub last_indexed {
return goto_specific(
suffix => $db_type,
@@ -236,7 +208,12 @@ sub process_mysql {
}
sub last_indexed_pg {
- my $attachments = attachments();
+ my $attachments = RT::Attachments->new( RT->SystemUser );
+ $attachments->Limit(
+ FIELD => 'ContentType',
+ OPERATOR => 'IN',
+ VALUE => ['text/plain', 'text/html'],
+ );
my $alias = 'main';
if ( $fts_config->{'Table'} && $fts_config->{'Table'} ne 'Attachments' ) {
$alias = $attachments->Join(
commit 18250bde09893ba9d936bb4b88f96886a66ba5f0
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 00:58:24 2014 -0400
mysql and pg share the same last_indexed; unify the method
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index f869b56..ea4b032 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -175,39 +175,6 @@ EOT
}
sub last_indexed {
- return goto_specific(
- suffix => $db_type,
- error => "Don't know how to find last indexed attachment for $db_type DB",
- arguments => \@_,
- );
-}
-
-sub process {
- return goto_specific(
- suffix => $db_type,
- error => "No processer for $db_type DB",
- arguments => \@_,
- );
-}
-
-sub last_indexed_mysql { last_indexed_pg(@_); }
-sub process_mysql {
- my ($attachment, $text) = (@_);
-
- my $dbh = $RT::Handle->dbh;
- my $table = $fts_config->{'Table'};
-
- my $query;
- if ( my ($id) = $dbh->selectrow_array("SELECT id FROM $table WHERE id = ?", undef, $attachment->id) ) {
- $query = "UPDATE $table SET Content = ? WHERE id = ?";
- } else {
- $query = "INSERT INTO $table(Content, id) VALUES(?, ?)";
- }
-
- $dbh->do( $query, undef, $$text, $attachment->id );
-}
-
-sub last_indexed_pg {
my $attachments = RT::Attachments->new( RT->SystemUser );
$attachments->Limit(
FIELD => 'ContentType',
@@ -236,6 +203,31 @@ sub last_indexed_pg {
return $res->id;
}
+sub process {
+ return goto_specific(
+ suffix => $db_type,
+ error => "No processer for $db_type DB",
+ arguments => \@_,
+ );
+}
+
+
+sub process_mysql {
+ my ($attachment, $text) = (@_);
+
+ my $dbh = $RT::Handle->dbh;
+ my $table = $fts_config->{'Table'};
+
+ my $query;
+ if ( my ($id) = $dbh->selectrow_array("SELECT id FROM $table WHERE id = ?", undef, $attachment->id) ) {
+ $query = "UPDATE $table SET Content = ? WHERE id = ?";
+ } else {
+ $query = "INSERT INTO $table(Content, id) VALUES(?, ?)";
+ }
+
+ $dbh->do( $query, undef, $$text, $attachment->id );
+}
+
sub process_pg {
my ($attachment, $text) = (@_);
commit eb95ffabd8e682f722e51d46cfb2524a9d961697
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 01:05:22 2014 -0400
Replace the last use of goto_specific with explicit function calls
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index ea4b032..8b4922e 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -168,7 +168,11 @@ EOT
my $text = $a->Content;
next unless defined $text && length($text);
$found++;
- process( $a, \$text );
+ if ($db_type eq 'mysql') {
+ process_mysql( $a, \$text );
+ } elsif ($db_type eq 'Pg') {
+ process_pg( $a, \$text );
+ }
debug("Processed attachment #". $a->id );
}
redo if $OPT{'all'} and $attachments->Count == ($OPT{'limit'} || 100);
@@ -203,14 +207,6 @@ sub last_indexed {
return $res->id;
}
-sub process {
- return goto_specific(
- suffix => $db_type,
- error => "No processer for $db_type DB",
- arguments => \@_,
- );
-}
-
sub process_mysql {
my ($attachment, $text) = (@_);
@@ -263,20 +259,6 @@ sub process_pg {
}
}
-sub goto_specific {
- my %args = (@_);
-
- my $func = (caller(1))[3];
- $func =~ s/.*:://;
- my $call = $func ."_". lc $args{'suffix'};
- unless ( defined &$call ) {
- return undef unless $args{'error'};
- require Carp; Carp::croak( $args{'error'} );
- }
- @_ = @{ $args{'arguments'} };
- goto &$call;
-}
-
# helper functions
sub debug { print @_, "\n" if $OPT{debug}; 1 }
commit f706f2bea12d7302c64b5c2c31e6e5711a322a83
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 15:31:36 2014 -0400
Simplify last_indexed
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index 8b4922e..645ec12 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -179,35 +179,17 @@ EOT
}
sub last_indexed {
- my $attachments = RT::Attachments->new( RT->SystemUser );
- $attachments->Limit(
- FIELD => 'ContentType',
- OPERATOR => 'IN',
- VALUE => ['text/plain', 'text/html'],
- );
- my $alias = 'main';
- if ( $fts_config->{'Table'} && $fts_config->{'Table'} ne 'Attachments' ) {
- $alias = $attachments->Join(
- TYPE => 'left',
- FIELD1 => 'id',
- TABLE2 => $fts_config->{'Table'},
- FIELD2 => 'id',
- );
+ if ( $db_type eq "mysql" ) {
+ return $dbh->selectrow_arrayref("SELECT MAX(id) FROM $table")->[0];
+ } elsif ( $db_type eq "pg" ) {
+ if ( $fts_config->{'Table'} && $fts_config->{'Table'} ne 'Attachments' ) {
+ return $dbh->selectrow_array("SELECT MAX(id) FROM $table")->[0];
+ } else {
+ return $dbh->selectrow_array("SELECT MAX(id) FROM Attachments WHERE $column IS NOT NULL")->[0];
+ }
}
- $attachments->Limit(
- ALIAS => $alias,
- FIELD => $fts_config->{'Column'},
- OPERATOR => 'IS NOT',
- VALUE => 'NULL',
- );
- $attachments->OrderBy( FIELD => 'id', ORDER => 'desc' );
- $attachments->RowsPerPage( 1 );
- my $res = $attachments->First;
- return 0 unless $res;
- return $res->id;
}
-
sub process_mysql {
my ($attachment, $text) = (@_);
commit 197874301fb4a3e82c0d48602abcc4efced6e5aa
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 15:33:58 2014 -0400
Only call last_indexed once, as it may be heavy
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index 645ec12..d0e636a 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -147,6 +147,7 @@ EOT
*RT::Attachments::Next = \&DBIx::SearchBuilder::Next;
}
+my $LAST = last_indexed();
{
my $attachments = RT::Attachments->new( RT->SystemUser );
$attachments->Limit(
@@ -157,7 +158,7 @@ EOT
$attachments->Limit(
FIELD => 'id',
OPERATOR => '>',
- VALUE => last_indexed()
+ VALUE => $LAST,
);
$attachments->OrderBy( FIELD => 'id', ORDER => 'asc' );
$attachments->RowsPerPage( $OPT{'limit'} || 100 );
@@ -175,6 +176,7 @@ EOT
}
debug("Processed attachment #". $a->id );
}
+ $LAST = $attachments->Last->id if $attachments->Count;
redo if $OPT{'all'} and $attachments->Count == ($OPT{'limit'} || 100);
}
commit 975d6f68b5079e50ee606cae696a4d7b6824e8ee
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 15:36:21 2014 -0400
Index even empty attachments
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index d0e636a..188eb67 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -49,6 +49,7 @@
use strict;
use warnings;
no warnings 'once';
+use 5.010;
# fix lib paths, some may be relative
BEGIN { # BEGIN RT CMD BOILERPLATE
@@ -166,8 +167,7 @@ my $LAST = last_indexed();
my $found = 0;
while ( my $a = $attachments->Next ) {
debug("Found attachment #". $a->id );
- my $text = $a->Content;
- next unless defined $text && length($text);
+ my $text = $a->Content // "";
$found++;
if ($db_type eq 'mysql') {
process_mysql( $a, \$text );
commit 9f41fc6f0330fa4276eb5feee6f4e6c47810e382
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 15:42:19 2014 -0400
As last_indexed is based on the highest insert, there will never be an UPDATE needed
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index 188eb67..f544131 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -198,12 +198,7 @@ sub process_mysql {
my $dbh = $RT::Handle->dbh;
my $table = $fts_config->{'Table'};
- my $query;
- if ( my ($id) = $dbh->selectrow_array("SELECT id FROM $table WHERE id = ?", undef, $attachment->id) ) {
- $query = "UPDATE $table SET Content = ? WHERE id = ?";
- } else {
- $query = "INSERT INTO $table(Content, id) VALUES(?, ?)";
- }
+ my $query = "INSERT INTO $table(Content, id) VALUES(?, ?)";
$dbh->do( $query, undef, $$text, $attachment->id );
}
@@ -216,12 +211,8 @@ sub process_pg {
my $column = $fts_config->{'Column'};
my $query;
- if ( $table ) {
- if ( my ($id) = $dbh->selectrow_array("SELECT id FROM $table WHERE id = ?", undef, $attachment->id) ) {
- $query = "UPDATE $table SET $column = to_tsvector(?) WHERE id = ?";
- } else {
- $query = "INSERT INTO $table($column, id) VALUES(to_tsvector(?), ?)";
- }
+ if ( $table ne 'Attachments' ) {
+ $query = "INSERT INTO $table($column, id) VALUES(to_tsvector(?), ?)";
} else {
$query = "UPDATE Attachments SET $column = to_tsvector(?) WHERE id = ?";
}
commit 5a8a4dbb0b596e90a534450272abc07fe26f98df
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 15:45:15 2014 -0400
Inversion of control of main indexing loops
Rather than having database-dependent if-statements, followed by a
standard loop which contains further database-dependent if-statements,
instead turn the loop into a function which cane be called from one of
two database-specific functions. This is important because MySQL's
iteration will further diverge from PostgreSQL's in the following
commits.
This commit is best viewed with:
git diff --patience --ignore-all-space
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index f544131..c7fd745 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -48,7 +48,6 @@
# END BPS TAGGED BLOCK }}}
use strict;
use warnings;
-no warnings 'once';
use 5.010;
# fix lib paths, some may be relative
@@ -148,68 +147,68 @@ EOT
*RT::Attachments::Next = \&DBIx::SearchBuilder::Next;
}
-my $LAST = last_indexed();
-{
- my $attachments = RT::Attachments->new( RT->SystemUser );
- $attachments->Limit(
- FIELD => 'ContentType',
- OPERATOR => 'IN',
- VALUE => ['text/plain', 'text/html'],
- );
- $attachments->Limit(
- FIELD => 'id',
- OPERATOR => '>',
- VALUE => $LAST,
- );
- $attachments->OrderBy( FIELD => 'id', ORDER => 'asc' );
- $attachments->RowsPerPage( $OPT{'limit'} || 100 );
-
- my $found = 0;
- while ( my $a = $attachments->Next ) {
- debug("Found attachment #". $a->id );
- my $text = $a->Content // "";
- $found++;
- if ($db_type eq 'mysql') {
- process_mysql( $a, \$text );
- } elsif ($db_type eq 'Pg') {
- process_pg( $a, \$text );
- }
- debug("Processed attachment #". $a->id );
- }
- $LAST = $attachments->Last->id if $attachments->Count;
- redo if $OPT{'all'} and $attachments->Count == ($OPT{'limit'} || 100);
+my $LAST;
+if ($db_type eq 'mysql') {
+ process_mysql();
+} elsif ($db_type eq 'Pg') {
+ process_pg();
}
-sub last_indexed {
- if ( $db_type eq "mysql" ) {
- return $dbh->selectrow_arrayref("SELECT MAX(id) FROM $table")->[0];
- } elsif ( $db_type eq "pg" ) {
- if ( $fts_config->{'Table'} && $fts_config->{'Table'} ne 'Attachments' ) {
- return $dbh->selectrow_array("SELECT MAX(id) FROM $table")->[0];
- } else {
- return $dbh->selectrow_array("SELECT MAX(id) FROM Attachments WHERE $column IS NOT NULL")->[0];
- }
+sub attachment_loop {
+ my $subref = shift;
+ my $table = $fts_config->{'Table'};
+ $LAST //= 0;
+
+ # Fetch in batches of size --limit
+ {
+ # Indexes all text/plain and text/html attachments
+ my $attachments = RT::Attachments->new( RT->SystemUser );
+ $attachments->Limit(
+ FIELD => 'ContentType',
+ OPERATOR => 'IN',
+ VALUE => ['text/plain', 'text/html'],
+ );
+ $attachments->Limit( FIELD => 'id', OPERATOR => '>', VALUE => $LAST );
+ $attachments->OrderBy( FIELD => 'id', ORDER => 'asc' );
+ $attachments->RowsPerPage( $OPT{'limit'} || 100 );
+
+ # Call back to the DB-specific part
+ $subref->($attachments);
+
+ $LAST = $attachments->Last->id if $attachments->Count;
+
+ redo if $OPT{'all'} and $attachments->Count == ($OPT{'limit'} || 100);
}
}
sub process_mysql {
- my ($attachment, $text) = (@_);
-
my $dbh = $RT::Handle->dbh;
my $table = $fts_config->{'Table'};
+ ($LAST) = $dbh->selectrow_array("SELECT MAX(id) FROM $table");
+
my $query = "INSERT INTO $table(Content, id) VALUES(?, ?)";
- $dbh->do( $query, undef, $$text, $attachment->id );
+ attachment_loop( sub {
+ my ($attachments) = @_;
+ while ( my $a = $attachments->Next ) {
+ debug("Found attachment #". $a->id );
+ $dbh->do( $query, undef, ($a->Content // ""), $a->id );
+ }
+ });
}
sub process_pg {
- my ($attachment, $text) = (@_);
-
my $dbh = $RT::Handle->dbh;
my $table = $fts_config->{'Table'};
my $column = $fts_config->{'Column'};
+ if ( $table ne 'Attachments' ) {
+ ($LAST) = $dbh->selectrow_array("SELECT MAX(id) FROM $table");
+ } else {
+ ($LAST) = $dbh->selectrow_array("SELECT MAX(id) FROM Attachments WHERE $column IS NOT NULL");
+ }
+
my $query;
if ( $table ne 'Attachments' ) {
$query = "INSERT INTO $table($column, id) VALUES(to_tsvector(?), ?)";
@@ -217,21 +216,27 @@ sub process_pg {
$query = "UPDATE Attachments SET $column = to_tsvector(?) WHERE id = ?";
}
- my $status = eval { $dbh->do( $query, undef, $$text, $attachment->id ) };
- unless ( $status ) {
- if ( $dbh->err == 7 && $dbh->state eq '54000' ) {
- warn "Attachment @{[$attachment->id]} cannot be indexed. Most probably it contains too many unique words. Error: ". $dbh->errstr;
- } elsif ( $dbh->err == 7 && $dbh->state eq '22021' ) {
- warn "Attachment @{[$attachment->id]} cannot be indexed. Most probably it contains invalid UTF8 bytes. Error: ". $dbh->errstr;
- } else {
- die "error: ". $dbh->errstr;
+ attachment_loop( sub {
+ my ($attachments) = @_;
+ while ( my $a = $attachments->Next ) {
+ debug("Found attachment #". $a->id );
+ my $status = eval { $dbh->do( $query, undef, ($a->Content // ""), $a->id ) };
+ unless ( $status ) {
+ if ( $dbh->err == 7 && $dbh->state eq '54000' ) {
+ warn "Attachment @{[$a->id]} cannot be indexed. Most probably it contains too many unique words. Error: ". $dbh->errstr;
+ } elsif ( $dbh->err == 7 && $dbh->state eq '22021' ) {
+ warn "Attachment @{[$a->id]} cannot be indexed. Most probably it contains invalid UTF8 bytes. Error: ". $dbh->errstr;
+ } else {
+ die "error: ". $dbh->errstr;
+ }
+
+ # Insert an empty tsvector, so we count this row as "indexed"
+ # for purposes of knowing where to pick up
+ eval { $dbh->do( $query, undef, "", $a->id ) }
+ or die "Failed to insert empty row: " . $dbh->errstr;
+ }
}
-
- # Insert an empty tsvector, so we count this row as "indexed"
- # for purposes of knowing where to pick up
- eval { $dbh->do( $query, undef, "", $attachment->id ) }
- or die "Failed to insert empty tsvector: " . $dbh->errstr;
- }
+ });
}
commit c569765a85f231be0d87d128ead2835da2e2f0f5
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 01:37:10 2014 -0400
Switch to preparing statements, rather than just setting strings
Prepared statements provide a small speed benefit, removing the need for
the database to re-parse the query string.
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index c7fd745..647b8f9 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -187,13 +187,13 @@ sub process_mysql {
($LAST) = $dbh->selectrow_array("SELECT MAX(id) FROM $table");
- my $query = "INSERT INTO $table(Content, id) VALUES(?, ?)";
+ my $sth = $dbh->prepare("INSERT INTO $table(Content, id) VALUES(?, ?)");
attachment_loop( sub {
my ($attachments) = @_;
while ( my $a = $attachments->Next ) {
debug("Found attachment #". $a->id );
- $dbh->do( $query, undef, ($a->Content // ""), $a->id );
+ $sth->execute( ($a->Content // ""), $a->id );
}
});
}
@@ -209,18 +209,18 @@ sub process_pg {
($LAST) = $dbh->selectrow_array("SELECT MAX(id) FROM Attachments WHERE $column IS NOT NULL");
}
- my $query;
+ my $sth;
if ( $table ne 'Attachments' ) {
- $query = "INSERT INTO $table($column, id) VALUES(to_tsvector(?), ?)";
+ $sth = $dbh->prepare("INSERT INTO $table($column, id) VALUES(to_tsvector(?), ?)");
} else {
- $query = "UPDATE Attachments SET $column = to_tsvector(?) WHERE id = ?";
+ $sth = $dbh->prepare("UPDATE Attachments SET $column = to_tsvector(?) WHERE id = ?");
}
attachment_loop( sub {
my ($attachments) = @_;
while ( my $a = $attachments->Next ) {
debug("Found attachment #". $a->id );
- my $status = eval { $dbh->do( $query, undef, ($a->Content // ""), $a->id ) };
+ my $status = eval { $sth->execute( ($a->Content // ""), $a->id ) };
unless ( $status ) {
if ( $dbh->err == 7 && $dbh->state eq '54000' ) {
warn "Attachment @{[$a->id]} cannot be indexed. Most probably it contains too many unique words. Error: ". $dbh->errstr;
@@ -232,7 +232,7 @@ sub process_pg {
# Insert an empty tsvector, so we count this row as "indexed"
# for purposes of knowing where to pick up
- eval { $dbh->do( $query, undef, "", $a->id ) }
+ eval { $sth->execute( "", $a->id ) }
or die "Failed to insert empty row: " . $dbh->errstr;
}
}
commit 440a429c8a69f6efabf8210fab29837a4aaece82
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 01:39:22 2014 -0400
INSERT DELAYED provides notable speed benefits on MyISAM
MySQL supports INSERT DELAYED[1] for MyISAM (but not InnoDB) tables. This
allows the server to defer inserts until it has a good opportunity to
write them, and to write them in bulk. While there is a small risk of
data loss (if the server is terminated before the data is written) this
poses no problem for an AttachmentsIndex table, for which all inserted
data is trivial to re-generate.
[1] http://dev.mysql.com/doc/refman/5.1/en/insert-delayed.html
diff --git a/lib/RT/Config.pm b/lib/RT/Config.pm
index 1f68749..0dc54cf 100644
--- a/lib/RT/Config.pm
+++ b/lib/RT/Config.pm
@@ -604,6 +604,7 @@ our %META;
} else {
# Internal, one-column table
$v->{Column} = 'Content';
+ $v->{Engine} = $engine;
}
}
} else {
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index 647b8f9..e9bea72 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -187,7 +187,8 @@ sub process_mysql {
($LAST) = $dbh->selectrow_array("SELECT MAX(id) FROM $table");
- my $sth = $dbh->prepare("INSERT INTO $table(Content, id) VALUES(?, ?)");
+ my $insert = $fts_config->{Engine} eq "MyISAM" ? "INSERT DELAYED" : "INSERT";
+ my $sth = $dbh->prepare("$insert INTO $table(Content, id) VALUES(?, ?)");
attachment_loop( sub {
my ($attachments) = @_;
commit ae4ab619b5c84f6b5252aa969ff12f73b780161b
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 01:40:16 2014 -0400
Improve MySQL insert speed by batching inserts into one statement
MySQL must flush buffers after every insert statement; as such,
providing large numbers of INSERT statements is quite inefficient, as
most time is spent in disk I/O. Instead, store rows to be inserted, and
insert them batch-by-batch.
This technique is not applicable to PostgreSQL because failure of the
to_tsvector call to convert a string would abort the entire insert.
Additionally, most installs use an additional column on the existing
table, which requires an UPDATE, and not an INSERT, which is not easily
batched.
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index e9bea72..13c1b12 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -187,15 +187,28 @@ sub process_mysql {
($LAST) = $dbh->selectrow_array("SELECT MAX(id) FROM $table");
+ # Doing large inserts is faster than individual statements, but
+ # comes at a parsing cost; cache the statement handles (99% of which
+ # will be the same size) for a notable (2x) speed gain.
+ my %sthandles;
my $insert = $fts_config->{Engine} eq "MyISAM" ? "INSERT DELAYED" : "INSERT";
- my $sth = $dbh->prepare("$insert INTO $table(Content, id) VALUES(?, ?)");
attachment_loop( sub {
my ($attachments) = @_;
+ my @insert;
+ my $found = 0;
while ( my $a = $attachments->Next ) {
debug("Found attachment #". $a->id );
- $sth->execute( ($a->Content // ""), $a->id );
+ push @insert, ($a->Content // ""), $a->id;
+ $found++;
}
+ return unless $found;
+
+ # $found should be the limit size on all but the last go-around.
+ $sthandles{$found} ||=
+ $dbh->prepare("$insert INTO $table(Content, id) VALUES "
+ . join(", ", ("(?,?)") x $found));
+ $sthandles{$found}->execute(@insert);
});
}
commit e1af270f7e7678e8bf4adbe2d41988fc0ad94770
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Jul 25 15:49:17 2014 -0400
Testing finds 200 is a good default batch size
Smaller batch sizes are slightly less efficient; the main trade-off is
in maximum memory consumption.
diff --git a/docs/full_text_indexing.pod b/docs/full_text_indexing.pod
index 378b629..79036f2 100644
--- a/docs/full_text_indexing.pod
+++ b/docs/full_text_indexing.pod
@@ -45,10 +45,9 @@ To keep the index up-to-date, you will need to run:
/opt/rt4/sbin/rt-fulltext-indexer
-...at regular intervals. By default, this will only tokenize up to 100
-tickets at a time; you can adjust this upwards by passing
-C<--limit 500>. Larger batch sizes will take longer and
-consume more memory.
+...at regular intervals. By default, this will only tokenize up to 200
+tickets at a time; you can adjust this upwards by passing C<--limit
+500>. Larger batch sizes will take longer and consume more memory.
If there is already an instances of C<rt-fulltext-indexer> running, new
ones will exit abnormally (with exit code 1) and the error message
@@ -95,10 +94,9 @@ To keep the index up-to-date, you will need to run:
/opt/rt4/sbin/rt-fulltext-indexer
-...at regular intervals. By default, this will only tokenize up to 100
-tickets at a time; you can adjust this upwards by passing
-C<--limit 500>. Larger batch sizes will take longer and
-consume more memory.
+...at regular intervals. By default, this will only tokenize up to 200
+tickets at a time; you can adjust this upwards by passing C<--limit
+500>. Larger batch sizes will take longer and consume more memory.
If there is already an instances of C<rt-fulltext-indexer> running, new
ones will exit abnormally (with exit code 1) and the error message
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index 13c1b12..e67915b 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -82,6 +82,7 @@ GetOptions( \%OPT,
"memory=s",
);
+$OPT{limit} ||= 200;
RT::Interface::CLI->ShowHelp if $OPT{help};
@@ -170,14 +171,14 @@ sub attachment_loop {
);
$attachments->Limit( FIELD => 'id', OPERATOR => '>', VALUE => $LAST );
$attachments->OrderBy( FIELD => 'id', ORDER => 'asc' );
- $attachments->RowsPerPage( $OPT{'limit'} || 100 );
+ $attachments->RowsPerPage( $OPT{'limit'} );
# Call back to the DB-specific part
$subref->($attachments);
$LAST = $attachments->Last->id if $attachments->Count;
- redo if $OPT{'all'} and $attachments->Count == ($OPT{'limit'} || 100);
+ redo if $OPT{'all'} and $attachments->Count == $OPT{'limit'};
}
}
commit 05a16d1d88a09e66c66fa218aacdfa7f033e214f
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Fri Aug 1 14:13:00 2014 -0400
Indexing of "text" may fail for content with invalid UTF-8 byte sequences
Prior to f04f561f, content that claimed to be UTF-8 was inserted into
the database verbatim. This errored immediately upon insertion for
PostgreSQL, as the Content column is of type "TEXT"; MySQL, however,
did not attempt to validate the bytes, as its Content column is of type
"LONGBLOB".
This thus only causes errors when attempting to insert the characters
into a "TEXT" column. As multiple rows are inserted in batches, use the
row number from the error message to determine which row to blank, and
re-attempt insertion.
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index e67915b..9b25717 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -209,7 +209,22 @@ sub process_mysql {
$sthandles{$found} ||=
$dbh->prepare("$insert INTO $table(Content, id) VALUES "
. join(", ", ("(?,?)") x $found));
- $sthandles{$found}->execute(@insert);
+ TRY: {
+ my $status = eval { $sthandles{$found}->execute(@insert); };
+ unless ( $status ) {
+ my ($row) = $dbh->errstr =~ /\brow (\d+)\b/;
+ die $dbh->errstr unless $row;
+
+ my ($content, $id) = ($insert[($row - 1)*2], $insert[($row - 1)*2 +1]);
+ if ($dbh->err == 1366 and $dbh->state eq "HY000") {
+ warn "Attachment $id cannot be indexed. Most probably it contains invalid UTF8 bytes. Error: ". $dbh->errstr;
+ $insert[($row - 1)*2] = "";
+ redo TRY;
+ } else {
+ die "Attachment $id: ".$dbh->errstr;
+ }
+ }
+ }
});
}
commit 45bfd41f127883cda76075848c1c2e59723e6b78
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Thu Nov 6 16:29:56 2014 -0800
Refactor PostgreSQL's insert to also do bulk insertion
PostgreSQL's full-text search can use a separate column, or a separate
table. In the case of a separate table, the same bulk-insertion trick
as MySQL uses can also be used to noticeably speed up indexing time.
Refactor the MySQL loop to be used in both cases. Note that instead of
attempting to parse the error message (which only works in English
locales for MySQL, and does not work at all in PostgreSQL) the rows are
instead attempted one-at-a-time. This also better catches the case
where multiple errors occurred in one bulk insert.
This provides notable speed benefits: 5x for GIN, and 30x for GiST. Note
that this does not speed up the default configuration for PostgreSQL,
wherein the index lies in the Attachments table, and thus UPDATE
statements must be used.
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index 9b25717..eac3a70 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -182,17 +182,17 @@ sub attachment_loop {
}
}
-sub process_mysql {
+sub process_bulk_insert {
my $dbh = $RT::Handle->dbh;
- my $table = $fts_config->{'Table'};
-
- ($LAST) = $dbh->selectrow_array("SELECT MAX(id) FROM $table");
+ my ($statement, $error) = @_;
# Doing large inserts is faster than individual statements, but
# comes at a parsing cost; cache the statement handles (99% of which
# will be the same size) for a notable (2x) speed gain.
my %sthandles;
- my $insert = $fts_config->{Engine} eq "MyISAM" ? "INSERT DELAYED" : "INSERT";
+
+ $sthandles{1} =
+ $dbh->prepare($statement->(1));
attachment_loop( sub {
my ($attachments) = @_;
@@ -206,45 +206,95 @@ sub process_mysql {
return unless $found;
# $found should be the limit size on all but the last go-around.
- $sthandles{$found} ||=
- $dbh->prepare("$insert INTO $table(Content, id) VALUES "
- . join(", ", ("(?,?)") x $found));
- TRY: {
- my $status = eval { $sthandles{$found}->execute(@insert); };
- unless ( $status ) {
- my ($row) = $dbh->errstr =~ /\brow (\d+)\b/;
- die $dbh->errstr unless $row;
-
- my ($content, $id) = ($insert[($row - 1)*2], $insert[($row - 1)*2 +1]);
- if ($dbh->err == 1366 and $dbh->state eq "HY000") {
- warn "Attachment $id cannot be indexed. Most probably it contains invalid UTF8 bytes. Error: ". $dbh->errstr;
- $insert[($row - 1)*2] = "";
- redo TRY;
- } else {
- die "Attachment $id: ".$dbh->errstr;
- }
- }
+ $sthandles{$found} ||= $dbh->prepare($statement->($found));
+
+ return if eval { $sthandles{$found}->execute(@insert); };
+
+ # We can catch and recover from some errors; re-do row-by-row to
+ # know which row had which errors
+ while (@insert) {
+ my ($content, $id) = splice(@insert,0,2);
+ next if eval { $sthandles{1}->execute($content, $id); };
+ $error->($id, $content);
+
+ # If this was a semi-expected error, insert an empty
+ # tsvector, so we count this row as "indexed" for
+ # purposes of knowing where to pick up
+ eval { $sthandles{1}->execute( "", $id ) }
+ or die "Failed to insert empty row for attachment $id: " . $dbh->errstr;
}
});
}
-sub process_pg {
+sub process_mysql {
my $dbh = $RT::Handle->dbh;
my $table = $fts_config->{'Table'};
- my $column = $fts_config->{'Column'};
- if ( $table ne 'Attachments' ) {
- ($LAST) = $dbh->selectrow_array("SELECT MAX(id) FROM $table");
- } else {
- ($LAST) = $dbh->selectrow_array("SELECT MAX(id) FROM Attachments WHERE $column IS NOT NULL");
- }
+ ($LAST) = $dbh->selectrow_array("SELECT MAX(id) FROM $table");
- my $sth;
- if ( $table ne 'Attachments' ) {
- $sth = $dbh->prepare("INSERT INTO $table($column, id) VALUES(to_tsvector(?), ?)");
+ my $insert = $fts_config->{Engine} eq "MyISAM" ? "INSERT DELAYED" : "INSERT";
+
+ process_bulk_insert(
+ sub {
+ my ($n) = @_;
+ return "$insert INTO $table(Content, id) VALUES "
+ . join(", ", ("(?,?)") x $n);
+ },
+ sub {
+ my ($id) = @_;
+ if ($dbh->err == 1366 and $dbh->state eq "HY000") {
+ warn "Attachment $id cannot be indexed. Most probably it contains invalid UTF8 bytes. ".
+ "Error: ". $dbh->errstr;
+ } else {
+ die "Attachment $id cannot be indexed: " . $dbh->errstr;
+ }
+ }
+ );
+}
+
+
+sub process_pg {
+ if ( $fts_config->{'Table'} ne 'Attachments' ) {
+ process_pg_insert();
} else {
- $sth = $dbh->prepare("UPDATE Attachments SET $column = to_tsvector(?) WHERE id = ?");
+ process_pg_update();
}
+}
+
+sub process_pg_insert {
+ my $dbh = $RT::Handle->dbh;
+ my $table = $fts_config->{'Table'};
+ my $column = $fts_config->{'Column'};
+ ($LAST) = $dbh->selectrow_array("SELECT MAX(id) FROM $table");
+
+ process_bulk_insert(
+ sub {
+ my ($n) = @_;
+ return "INSERT INTO $table($column, id) VALUES "
+ . join(", ", ("(TO_TSVECTOR(?),?)") x $n);
+ },
+ sub {
+ my ($id) = @_;
+ if ( $dbh->err == 7 && $dbh->state eq '54000' ) {
+ warn "Attachment $id cannot be indexed. Most probably it contains too many unique words. ".
+ "Error: ". $dbh->errstr;
+ } elsif ( $dbh->err == 7 && $dbh->state eq '22021' ) {
+ warn "Attachment $id cannot be indexed. Most probably it contains invalid UTF8 bytes. ".
+ "Error: ". $dbh->errstr;
+ } else {
+ die "Attachment $id cannot be indexed: " . $dbh->errstr;
+ }
+ }
+ );
+}
+
+sub process_pg_update {
+ my $dbh = $RT::Handle->dbh;
+ my $column = $fts_config->{'Column'};
+
+ ($LAST) = $dbh->selectrow_array("SELECT MAX(id) FROM Attachments WHERE $column IS NOT NULL");
+
+ $sth = $dbh->prepare("UPDATE Attachments SET $column = to_tsvector(?) WHERE id = ?");
attachment_loop( sub {
my ($attachments) = @_;
commit 4d21e316f03b885f32a360a7b5c591a82283d324
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Thu Nov 6 17:07:58 2014 -0800
Perform PostgreSQL UPDATE statements inside of a database transaction
This attempts to reduce the number of write operations necessary, by
allowing PostgreSQL to batch writes, doing them only at COMMIT time.
This requires optimistically assuming that all UPDATEs will succeed --
and of any one does not, redoing all of them one-by-one. In cases where
indexing errors are frequent (at least one per two batches), this will
lead to a decrease in performance. However, in most cases it results in
a notable performance increase: 3x for GIN, 10x for GiST.
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index eac3a70..458815d 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -294,27 +294,50 @@ sub process_pg_update {
($LAST) = $dbh->selectrow_array("SELECT MAX(id) FROM Attachments WHERE $column IS NOT NULL");
- $sth = $dbh->prepare("UPDATE Attachments SET $column = to_tsvector(?) WHERE id = ?");
+ my $sth = $dbh->prepare("UPDATE Attachments SET $column = TO_TSVECTOR(?) WHERE id = ?");
attachment_loop( sub {
my ($attachments) = @_;
+ my @insert;
while ( my $a = $attachments->Next ) {
debug("Found attachment #". $a->id );
- my $status = eval { $sth->execute( ($a->Content // ""), $a->id ) };
- unless ( $status ) {
- if ( $dbh->err == 7 && $dbh->state eq '54000' ) {
- warn "Attachment @{[$a->id]} cannot be indexed. Most probably it contains too many unique words. Error: ". $dbh->errstr;
- } elsif ( $dbh->err == 7 && $dbh->state eq '22021' ) {
- warn "Attachment @{[$a->id]} cannot be indexed. Most probably it contains invalid UTF8 bytes. Error: ". $dbh->errstr;
- } else {
- die "error: ". $dbh->errstr;
- }
-
- # Insert an empty tsvector, so we count this row as "indexed"
- # for purposes of knowing where to pick up
- eval { $sth->execute( "", $a->id ) }
- or die "Failed to insert empty row: " . $dbh->errstr;
+ push @insert, [($a->Content // ""), $a->id];
+ }
+
+ # Try in one database transaction; if it fails, we roll it back
+ # and try one statement at a time.
+ $dbh->begin_work;
+ my $ok = 1;
+ for (@insert) {
+ $ok = eval { $sth->execute( $_->[0], $_->[1] ) };
+ last unless $ok;
+ }
+ if ($ok) {
+ $dbh->commit;
+ return;
+ }
+ $dbh->rollback;
+
+ # Things didn't go well. Retry the UPDATE statements one row at
+ # a time, outside of the transaction.
+ for (@insert) {
+ my ($content, $id) = ($_->[0], $_->[1]);
+ next if eval { $sth->execute( $content, $id ) };
+ if ( $dbh->err == 7 && $dbh->state eq '54000' ) {
+ warn "Attachment $id cannot be indexed. Most probably it contains too many unique words. ".
+ "Error: ". $dbh->errstr;
+ } elsif ( $dbh->err == 7 && $dbh->state eq '22021' ) {
+ warn "Attachment $id cannot be indexed. Most probably it contains invalid UTF8 bytes. ".
+ "Error: ". $dbh->errstr;
+ } else {
+ die "Attachment $id cannot be indexed: " . $dbh->errstr;
}
+
+ # If this was a semi-expected error, insert an empty
+ # tsvector, so we count this row as "indexed" for
+ # purposes of knowing where to pick up
+ eval { $sth->execute( "", $id ) }
+ or die "Failed to insert empty row for attachment $id: " . $dbh->errstr;
}
});
}
commit d7bca97256c8e927dbef4dcde12f890233f1ec30
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Thu Nov 6 17:41:06 2014 -0800
If a new table is used for indexing, grant rights on it
Without this, the new table (created by the DBA user) cannot be read or
inserted into by the RT user, resulting in errors when indexing.
diff --git a/sbin/rt-setup-fulltext-index.in b/sbin/rt-setup-fulltext-index.in
index c4d8437..d39554d 100644
--- a/sbin/rt-setup-fulltext-index.in
+++ b/sbin/rt-setup-fulltext-index.in
@@ -307,16 +307,20 @@ elsif ( $DB{'type'} eq 'Pg' ) {
silent => !$OPT{'ask'},
);
- my $schema;
+ my @schema;
my $drop;
if ( lc($table) eq 'attachments' ) {
$drop = "ALTER TABLE $table DROP COLUMN $column";
- $schema = "ALTER TABLE $table ADD COLUMN $column tsvector";
+ push @schema, "ALTER TABLE $table ADD COLUMN $column tsvector";
} else {
$drop = "DROP TABLE $table";
- $schema = "CREATE TABLE $table ( "
- ."id INTEGER NOT NULL,"
- ."$column tsvector )";
+ push @schema, split /;\n+/, <<SCHEMA;
+CREATE TABLE $table (
+ id SERIAL,
+ $column tsvector
+);
+GRANT SELECT, INSERT, UPDATE, DELETE ON $table TO "$DB{user}"
+SCHEMA
}
my $index_type = lc($OPT{'index-type'} || '');
@@ -328,11 +332,11 @@ elsif ( $DB{'type'} eq 'Pg' ) {
silent => !$OPT{'ask'},
);
}
+ push @schema, "CREATE INDEX ${column}_idx ON $table USING $index_type($column)";
do_error_is_ok( dba_handle() => $drop )
unless $OPT{'dryrun'};
- insert_schema( $schema );
- insert_schema("CREATE INDEX ${column}_idx ON $table USING $index_type($column)");
+ insert_schema( $_ ) for @schema;
print_rt_config( Table => $table, Column => $column );
}
commit 1ac62462fcd6ff94e048c483432c7ae37f324a0e
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Tue Nov 11 00:35:53 2014 -0500
Insert data to index before creating the index
Having to update the index after every insertion is a notable
performance hit; it is much more performant to create the index after
all data has been inserted.
Move the initial run of rt-fulltext-indexer to after the requisite
tables have been set up, but prior to creating the index and displaying
the required RT configuration. This requires serializing the intended
%FullTextSearch configuration between rt-setup-fulltext-index and
rt-fulltext-indexer, as the RT_SiteConfig file will not have been
updated until after both have run; thus, serialize the configuration via
an environment variable.
The documentation is updated to note that running
rt-setup-fulltext-index is now a time-intensive operation.
diff --git a/docs/full_text_indexing.pod b/docs/full_text_indexing.pod
index 79036f2..ad49f74 100644
--- a/docs/full_text_indexing.pod
+++ b/docs/full_text_indexing.pod
@@ -28,16 +28,14 @@ you may need to pass the C<--dba> or C<--dba-password> options:
/opt/rt4/sbin/rt-setup-fulltext-index --dba postgres --dba-password secret
-This will also output an appropriate C<%FullTextSearch> configuration to
-add to your F<RT_SiteConfig.pm>; you will need to restart your webserver
-after making these changes. However, the index will also need to be
-filled before it can be used. To update the index initially, run:
+This will then tokenize and index all existing attachments in your
+database; it may take quite a while if your database already has a large
+number of tickets in it.
- /opt/rt4/sbin/rt-fulltext-indexer --all
+Finally, it will output an appropriate C<%FullTextSearch> configuration
+to add to your F<RT_SiteConfig.pm>; you will need to restart your
+webserver after making these changes.
-This will tokenize and index all existing attachments in your database;
-it may take quite a while if your database already has a large number of
-tickets in it.
=head2 Updating the index
@@ -77,16 +75,14 @@ you may need to pass the C<--dba> or C<--dba-password> options:
/opt/rt4/sbin/rt-setup-fulltext-index --dba root --dba-password secret
-This will also output an appropriate C<%FullTextSearch> configuration to
-add to your F<RT_SiteConfig.pm>; you will need to restart your webserver
-after making these changes. However, the index will also need to be
-filled before it can be used. To update the index initially, run:
+This will then tokenize and index all existing attachments in your
+database; it may take quite a while if your database already has a large
+number of tickets in it.
- /opt/rt4/sbin/rt-fulltext-indexer --all
+Finally, it will output an appropriate C<%FullTextSearch> configuration
+to add to your F<RT_SiteConfig.pm>; you will need to restart your
+webserver after making these changes.
-This will tokenize and index all existing attachments in your database;
-it may take quite a while if your database already has a large number of
-tickets in it.
=head3 Updating the index
diff --git a/sbin/rt-fulltext-indexer.in b/sbin/rt-fulltext-indexer.in
index 458815d..5bcd9a2 100644
--- a/sbin/rt-fulltext-indexer.in
+++ b/sbin/rt-fulltext-indexer.in
@@ -99,7 +99,8 @@ if ( !flock main::DATA, LOCK_EX | LOCK_NB ) {
}
my $db_type = RT->Config->Get('DatabaseType');
-my $fts_config = RT->Config->Get('FullTextSearch') || {};
+my $fts_config = $ENV{RT_FTS_CONFIG} ? JSON::from_json($ENV{RT_FTS_CONFIG})
+ : RT->Config->Get('FullTextSearch') || {};
unless ( $fts_config->{'Enable'} ) {
print STDERR <<EOT;
diff --git a/sbin/rt-setup-fulltext-index.in b/sbin/rt-setup-fulltext-index.in
index d39554d..b39b7b2 100644
--- a/sbin/rt-setup-fulltext-index.in
+++ b/sbin/rt-setup-fulltext-index.in
@@ -121,6 +121,7 @@ GetOptions(
'dba=s' => \$DB{'admin'},
'dba-password=s' => \$DB{'admin_password'},
+ 'limit=i' => \$DB{'batch-size'},
) or show_help();
if ( $OPT{'help'} || (!$DB{'admin'} && $DB{'type'} eq 'Oracle' ) ) {
@@ -175,9 +176,13 @@ if ( $DB{'type'} eq 'mysql' ) {
my $engine = $RT::Handle->dbh->{mysql_serverversion} < 50600 ? "MyISAM" : "InnoDB";
my $schema = "CREATE TABLE $table ( "
."id INT UNSIGNED AUTO_INCREMENT NOT NULL PRIMARY KEY,"
- ."Content LONGTEXT, FULLTEXT(Content) ) ENGINE=$engine CHARACTER SET utf8";
+ ."Content LONGTEXT ) ENGINE=$engine CHARACTER SET utf8";
insert_schema( $schema );
+ insert_data( Table => $table, Engine => $engine );
+
+ insert_schema( "CREATE FULLTEXT INDEX $table ON $table(Content)" );
+
print_rt_config( Table => $table );
} elsif ($DB{'type'} eq 'sphinx') {
check_sphinx();
@@ -332,12 +337,15 @@ SCHEMA
silent => !$OPT{'ask'},
);
}
- push @schema, "CREATE INDEX ${column}_idx ON $table USING $index_type($column)";
do_error_is_ok( dba_handle() => $drop )
unless $OPT{'dryrun'};
insert_schema( $_ ) for @schema;
+ insert_data( Table => $table, Column => $column );
+
+ insert_schema( "CREATE INDEX ${column}_idx ON $table USING $index_type($column)" );
+
print_rt_config( Table => $table, Column => $column );
}
elsif ( $DB{'type'} eq 'Oracle' ) {
@@ -729,6 +737,16 @@ sub insert_schema {
}
}
+sub insert_data {
+ return if $OPT{dryrun};
+
+ print "Indexing existing data...\n";
+
+ $ENV{RT_FTS_CONFIG} = JSON::to_json( {Enable => 1, Indexed => 1, @_});
+ system( "$RT::SbinPath/rt-fulltext-indexer", "--all",
+ ($DB{'batch-size'} ? ("--limit", $DB{'batch-size'}) : ()));
+}
+
=head1 NAME
rt-setup-fulltext-index - Create indexes for full text search
commit 4045f7d2faa5d22517113383dd80fc7a5e069b24
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Wed Nov 26 15:18:41 2014 -0500
Switch the default Postgres index to GIN
GIN indexes, while slower to index, provide a 10x speedup in query time;
they are suggested for static data, which the Attachments table
certainly is. The other improvements in indexing are sufficient to
still net a 6.5x improvement in indexing speed over the speed of the
prior GiST default. In addition, indexing time, which mostly impacts
only initial deployment, should take second priority behind query time,
which GIN improves notably.
diff --git a/sbin/rt-setup-fulltext-index.in b/sbin/rt-setup-fulltext-index.in
index b39b7b2..d5931cd 100644
--- a/sbin/rt-setup-fulltext-index.in
+++ b/sbin/rt-setup-fulltext-index.in
@@ -331,9 +331,9 @@ SCHEMA
my $index_type = lc($OPT{'index-type'} || '');
while ( $index_type ne 'gist' and $index_type ne 'gin' ) {
$index_type = lc prompt(
- message => "You may choose between GiST or GIN indexes; the former is several times\n"
- . "slower to search, but takes less space on disk and is faster to update.",
- default => 'GiST',
+ message => "You may choose between GiST or GIN indexes; the GiST takes less space on\n"
+ . "disk and is faster to update, but is an order of magnitude slower to query.",
+ default => 'GIN',
silent => !$OPT{'ask'},
);
}
commit dff4059ab03803ee0990573ca9397b034ccc998a
Author: Alex Vandiver <alexmv at bestpractical.com>
Date: Wed Nov 26 15:29:06 2014 -0500
Default to storing the tsvector in a new table, to speed indexing
The ability to perform bulk INSERT, rather than individual UPDATE
statements, increases indexing speeds by 1.5x. While this requires an
additional JOIN at query time, testing shows that this comes at no
notable cost of execution time -- in fact, queries with the tsvector in
a separate table perform 5-10% _faster_ than those with it in the
Attachments table, perhaps because it requires scans of a less
heavyweight table.
diff --git a/sbin/rt-setup-fulltext-index.in b/sbin/rt-setup-fulltext-index.in
index d5931cd..7f51b88 100644
--- a/sbin/rt-setup-fulltext-index.in
+++ b/sbin/rt-setup-fulltext-index.in
@@ -91,7 +91,7 @@ my %OPT = (
my %DEFAULT;
if ( $DB{'type'} eq 'Pg' ) {
%DEFAULT = (
- table => 'Attachments',
+ table => 'AttachmentsIndex',
column => 'ContentIndex',
);
}
@@ -302,7 +302,7 @@ elsif ( $DB{'type'} eq 'Pg' ) {
my $table = $OPT{'table'} || prompt(
message => "Enter the name of a DB table that will be used to store the Pg tsvector.\n"
. "You may either use the existing Attachments table, or create a new\n"
- . "table.",
+ . "table. Creating a new table makes initial indexing faster.",
default => $DEFAULT{'table'},
silent => !$OPT{'ask'},
);
diff --git a/t/fts/indexed_pg.t b/t/fts/indexed_pg.t
index b8d4b1b..b80cd12 100644
--- a/t/fts/indexed_pg.t
+++ b/t/fts/indexed_pg.t
@@ -11,7 +11,7 @@ plan skip_all => "Need Pg 8.2 or higher; we have $major.$minor"
plan tests => 36;
-RT->Config->Set( FullTextSearch => Enable => 1, Indexed => 1, Column => 'ContentIndex', Table => 'Attachments' );
+RT->Config->Set( FullTextSearch => Enable => 1, Indexed => 1, Column => 'ContentIndex', Table => 'AttachmentsIndex' );
setup_indexing();
-----------------------------------------------------------------------
More information about the rt-commit
mailing list