X-Git-Url: http://www.git.stargrave.org/?a=blobdiff_plain;f=lib%2FPublicInbox%2FSearchIdx.pm;h=3596972faa962ebec1160a51f1cbdbf9cd799b98;hb=fad9acd35e56a289ade90a62d056b2a6663d448c;hp=3ef444d67ea0af8f2fba679ea3a19360d556e13f;hpb=95bd8265dfb67ba90e7068bf5a4360168a1f30b6;p=public-inbox.git diff --git a/lib/PublicInbox/SearchIdx.pm b/lib/PublicInbox/SearchIdx.pm index 3ef444d6..3596972f 100644 --- a/lib/PublicInbox/SearchIdx.pm +++ b/lib/PublicInbox/SearchIdx.pm @@ -9,23 +9,19 @@ package PublicInbox::SearchIdx; use strict; use warnings; -use Fcntl qw(:flock :DEFAULT); +use base qw(PublicInbox::Search PublicInbox::Lock); use PublicInbox::MIME; -use base qw(PublicInbox::Search); +use PublicInbox::InboxWritable; use PublicInbox::MID qw/mid_clean id_compress mid_mime mids references/; use PublicInbox::MsgIter; use Carp qw(croak); use POSIX qw(strftime); +use PublicInbox::OverIdx; require PublicInbox::Git; +use Compress::Zlib qw(compress); use constant { - MAX_MID_SIZE => 244, # max term size (Xapian limitation) - length('Q') - PERM_UMASK => 0, - OLD_PERM_GROUP => 1, - OLD_PERM_EVERYBODY => 2, - PERM_GROUP => 0660, - PERM_EVERYBODY => 0664, - BATCH_BYTES => 1_000_000, + BATCH_BYTES => 10_000_000, DEBUG => !!$ENV{DEBUG}, }; @@ -64,28 +60,29 @@ sub new { PublicInbox::AltId->new($ibx, $_); } @$altid ]; } - $git = $ibx->git; - } else { - $git = PublicInbox::Git->new($git_dir); # v1 only + } else { # v1 + $ibx = { mainrepo => $git_dir, version => 1 }; } + $ibx = PublicInbox::InboxWritable->new($ibx); require Search::Xapian::WritableDatabase; my $self = bless { mainrepo => $mainrepo, - git => $git, + -inbox => $ibx, + git => $ibx->git, -altid => $altid, version => $version, }, $class; - my $perm = $self->_git_config_perm; - my $umask = _umask_for($perm); - $self->{umask} = $umask; + $ibx->umask_prepare; if ($version == 1) { $self->{lock_path} = "$mainrepo/ssoma.lock"; + my $dir = $self->xdir; + $self->{over_ro} = $self->{over} = + PublicInbox::OverIdx->new("$dir/over.sqlite3"); } elsif ($version == 2) { defined $part or die "partition is required for v2\n"; - # partition is a number or "all" + # partition is a number $self->{partition} = $part; $self->{lock_path} = undef; - $self->{msgmap_path} = "$mainrepo/msgmap.sqlite3"; } else { die "unsupported inbox version=$version\n"; } @@ -97,7 +94,7 @@ sub _xdb_release { my ($self) = @_; my $xdb = delete $self->{xdb} or croak 'not acquired'; $xdb->close; - _lock_release($self) if $self->{creat}; + $self->lock_release if $self->{creat}; undef; } @@ -108,58 +105,19 @@ sub _xdb_acquire { my $flag = Search::Xapian::DB_OPEN; if ($self->{creat}) { require File::Path; - _lock_acquire($self); + $self->lock_acquire; File::Path::mkpath($dir); $flag = Search::Xapian::DB_CREATE_OR_OPEN; } $self->{xdb} = Search::Xapian::WritableDatabase->new($dir, $flag); } -# we only acquire the flock if creating or reindexing; -# PublicInbox::Import already has the lock on its own. -sub _lock_acquire { - my ($self) = @_; - croak 'already locked' if $self->{lockfh}; - my $lock_path = $self->{lock_path} or return; - sysopen(my $lockfh, $lock_path, O_WRONLY|O_CREAT) or - die "failed to open lock $lock_path: $!\n"; - flock($lockfh, LOCK_EX) or die "lock failed: $!\n"; - $self->{lockfh} = $lockfh; -} - -sub _lock_release { - my ($self) = @_; - return unless $self->{lock_path}; - my $lockfh = delete $self->{lockfh} or croak 'not locked'; - flock($lockfh, LOCK_UN) or die "unlock failed: $!\n"; - close $lockfh or die "close failed: $!\n"; -} - sub add_val ($$$) { my ($doc, $col, $num) = @_; $num = Search::Xapian::sortable_serialise($num); $doc->add_value($col, $num); } -sub add_values ($$) { - my ($doc, $values) = @_; - - my $ts = $values->[PublicInbox::Search::TS]; - add_val($doc, PublicInbox::Search::TS, $ts); - - my $num = $values->[PublicInbox::Search::NUM]; - defined($num) and add_val($doc, PublicInbox::Search::NUM, $num); - - my $bytes = $values->[PublicInbox::Search::BYTES]; - defined($bytes) and add_val($doc, PublicInbox::Search::BYTES, $bytes); - - my $lines = $values->[PublicInbox::Search::LINES]; - add_val($doc, PublicInbox::Search::LINES, $lines); - - my $yyyymmdd = strftime('%Y%m%d', gmtime($ts)); - add_val($doc, PublicInbox::Search::YYYYMMDD, $yyyymmdd); -} - sub index_users ($$) { my ($tg, $smsg) = @_; @@ -303,20 +261,18 @@ sub index_body ($$$) { } sub add_message { - my ($self, $mime, $bytes, $num, $blob) = @_; # mime = Email::MIME object + # mime = Email::MIME object + my ($self, $mime, $bytes, $num, $oid, $mid0) = @_; my $doc_id; my $mids = mids($mime->header_obj); - my $skel = $self->{skeleton}; - + $mid0 = $mids->[0] unless defined $mid0; # v1 compatibility + unless (defined $num) { # v1 + my $mm = $self->_msgmap_init; + $num = $mm->mid_insert($mid0) || $mm->num_for($mid0); + } eval { my $smsg = PublicInbox::SearchMsg->new($mime); my $doc = $smsg->{doc}; - foreach my $mid (@$mids) { - # FIXME: may be abused to prevent archival - length($mid) > MAX_MID_SIZE and - die 'Message-ID too long'; - $doc->add_term('Q' . $mid); - } my $subj = $smsg->subject; my $xpath; if ($subj ne '') { @@ -324,9 +280,17 @@ sub add_message { $xpath = id_compress($xpath); } - my $lines = $mime->body_raw =~ tr!\n!\n!; - my @values = ($smsg->ts, $num, $bytes, $lines); - add_values($doc, \@values); + $smsg->{lines} = $mime->body_raw =~ tr!\n!\n!; + defined $bytes or $bytes = length($mime->as_string); + $smsg->{bytes} = $bytes; + + add_val($doc, PublicInbox::Search::TS(), $smsg->ts); + my @ds = gmtime($smsg->ds); + my $yyyymmdd = strftime('%Y%m%d', @ds); + add_val($doc, PublicInbox::Search::YYYYMMDD(), $yyyymmdd); + my $dt = strftime('%Y%m%d%H%M%S', @ds); + add_val($doc, PublicInbox::Search::DT(), $dt); + my @vals = ($smsg->{ts}, $smsg->{ds}); my $tg = $self->term_generator; @@ -376,8 +340,8 @@ sub add_message { }); # populates smsg->references for smsg->to_doc_data - my $refs = parse_references($smsg); - my $data = $smsg->to_doc_data($blob); + my $refs = parse_references($smsg, $mid0, $mids); + my $data = $smsg->to_doc_data($oid, $mid0); foreach my $mid (@$mids) { $tg->index_text($mid, 1, 'XM'); } @@ -392,14 +356,16 @@ sub add_message { } } } - if ($skel) { - push @values, $mids, $xpath, $data; - $skel->index_skeleton(\@values); - $doc_id = $self->{xdb}->add_document($doc); - } else { - $doc_id = link_and_save($self, $doc, $mids, $refs, - $num, $xpath); - } + + $self->delete_article($num) if defined $num; # for reindexing + + utf8::encode($data); + $data = compress($data); + push @vals, $num, $mids, $refs, $xpath, $data; + $self->{over}->add_over(\@vals); + $doc->add_boolean_term('Q' . $_) foreach @$mids; + $doc->add_boolean_term('XNUM' . $num) if defined $num; + $doc_id = $self->{xdb}->add_document($doc); }; if ($@) { @@ -409,29 +375,86 @@ sub add_message { $doc_id; } -# returns deleted doc_id on success, undef on missing +# returns begin and end PostingIterator +sub find_doc_ids { + my ($self, $termval) = @_; + my $db = $self->{xdb}; + + ($db->postlist_begin($termval), $db->postlist_end($termval)); +} + +sub batch_do { + my ($self, $termval, $cb) = @_; + my $batch_size = 1000; # don't let @ids grow too large to avoid OOM + while (1) { + my ($head, $tail) = $self->find_doc_ids($termval); + return if $head == $tail; + my @ids; + for (; $head != $tail && @ids < $batch_size; $head->inc) { + push @ids, $head->get_docid; + } + $cb->(\@ids); + } +} + sub remove_message { my ($self, $mid) = @_; my $db = $self->{xdb}; - my $doc_id; + my $called; $mid = mid_clean($mid); eval { - my ($head, $tail) = $self->find_doc_ids('Q' . $mid); - if ($head->equal($tail)) { - warn "cannot remove non-existent <$mid>\n"; - } - for (; $head != $tail; $head->inc) { - my $docid = $head->get_docid; - $db->delete_document($docid); - } + batch_do($self, 'Q' . $mid, sub { + my ($ids) = @_; + $db->delete_document($_) for @$ids; + $called = 1; + }); }; - if ($@) { warn "failed to remove message <$mid>: $@\n"; - return undef; + } elsif (!$called) { + warn "cannot remove non-existent <$mid>\n"; } - $doc_id; +} + +sub delete_article { + my ($self, $num) = @_; + my $ndel = 0; + batch_do($self, 'XNUM' . $num, sub { + my ($ids) = @_; + $ndel += scalar @$ids; + $self->{xdb}->delete_document($_) for @$ids; + }); +} + +# MID is a hint in V2 +sub remove_by_oid { + my ($self, $oid, $mid) = @_; + my $db = $self->{xdb}; + + # XXX careful, we cannot use batch_do here since we conditionally + # delete documents based on other factors, so we cannot call + # find_doc_ids twice. + my ($head, $tail) = $self->find_doc_ids('Q' . $mid); + return if $head == $tail; + + # there is only ONE element in @delete unless we + # have bugs in our v2writable deduplication check + my @delete; + my @over_del; + for (; $head != $tail; $head->inc) { + my $docid = $head->get_docid; + my $doc = $db->get_document($docid); + my $smsg = PublicInbox::SearchMsg->wrap($doc, $mid); + $smsg->load_expand; + if ($smsg->{blob} eq $oid) { + push(@delete, $docid); + push(@over_del, $smsg->num); + } + } + $db->delete_document($_) foreach @delete; + $self->{over}->remove_oid($oid, $mid); + scalar(@delete); } sub term_generator { # write-only @@ -446,100 +469,29 @@ sub term_generator { # write-only $self->{term_generator} = $tg; } -# increments last_thread_id counter -# returns a 64-bit integer represented as a decimal string -sub next_thread_id { - my ($self) = @_; - my $db = $self->{xdb}; - my $last_thread_id = int($db->get_metadata('last_thread_id') || 0); - - $db->set_metadata('last_thread_id', ++$last_thread_id); - - $last_thread_id; -} - -sub parse_references ($) { - my ($smsg) = @_; +sub parse_references ($$$) { + my ($smsg, $mid0, $mids) = @_; my $mime = $smsg->{mime}; my $hdr = $mime->header_obj; my $refs = references($hdr); + push(@$refs, @$mids) if scalar(@$mids) > 1; return $refs if scalar(@$refs) == 0; - # prevent circular references via References here: - my %mids = map { $_ => 1 } @{mids($hdr)}; + # prevent circular references here: + my %seen = ( $mid0 => 1 ); my @keep; foreach my $ref (@$refs) { - # FIXME: this is an archive-prevention vector like X-No-Archive - if (length($ref) > MAX_MID_SIZE) { + if (length($ref) > PublicInbox::MID::MAX_MID_SIZE) { warn "References: <$ref> too long, ignoring\n"; + next; } - next if $mids{$ref}; + next if $seen{$ref}++; push @keep, $ref; } $smsg->{references} = '<'.join('> <', @keep).'>' if @keep; \@keep; } -sub link_doc { - my ($self, $doc, $refs, $old_tid) = @_; - my $tid; - - if (@$refs) { - # first ref *should* be the thread root, - # but we can never trust clients to do the right thing - my $ref = shift @$refs; - $tid = resolve_mid_to_tid($self, $ref); - merge_threads($self, $tid, $old_tid) if defined $old_tid; - - # the rest of the refs should point to this tid: - foreach $ref (@$refs) { - my $ptid = resolve_mid_to_tid($self, $ref); - merge_threads($self, $tid, $ptid); - } - } else { - $tid = defined $old_tid ? $old_tid : $self->next_thread_id; - } - $doc->add_boolean_term('G' . $tid); - $tid; -} - -sub link_and_save { - my ($self, $doc, $mids, $refs, $num, $xpath) = @_; - my $db = $self->{xdb}; - my $old_tid; - my $doc_id; - $doc->add_boolean_term('XNUM' . $num) if defined $num; - $doc->add_boolean_term('XPATH' . $xpath) if defined $xpath; - my $vivified = 0; - foreach my $mid (@$mids) { - $self->each_smsg_by_mid($mid, sub { - my ($cur) = @_; - my $type = $cur->type; - my $cur_tid = $cur->thread_id; - $old_tid = $cur_tid unless defined $old_tid; - if ($type eq 'mail') { - # do not break existing mail messages, - # just merge the threads - merge_threads($self, $old_tid, $cur_tid); - return 1; - } - if ($type ne 'ghost') { - die "<$mid> has a bad type: $type\n"; - } - my $tid = link_doc($self, $doc, $refs, $old_tid); - $old_tid = $tid unless defined $old_tid; - $doc_id = $cur->{doc_id}; - $self->{xdb}->replace_document($doc_id, $doc); - ++$vivified; - 1; - }); - } - # not really important, but we return any vivified ghost docid, here: - return $doc_id if defined $doc_id; - link_doc($self, $doc, $refs, $old_tid); - $self->{xdb}->add_document($doc); -} - sub index_git_blob_id { my ($doc, $pfx, $objid) = @_; @@ -609,7 +561,7 @@ sub do_cat_mail { sub index_sync { my ($self, $opts) = @_; - with_umask($self, sub { $self->_index_sync($opts) }); + $self->{-inbox}->with_umask(sub { $self->_index_sync($opts) }) } sub batch_adjust ($$$$) { @@ -654,14 +606,10 @@ sub rlog { sub _msgmap_init { my ($self) = @_; + die "BUG: _msgmap_init is only for v1\n" if $self->{version} != 1; $self->{mm} ||= eval { require PublicInbox::Msgmap; - my $msgmap_path = $self->{msgmap_path}; - if (defined $msgmap_path) { # v2 - PublicInbox::Msgmap->new_file($msgmap_path, 1); - } else { - PublicInbox::Msgmap->new($self->{mainrepo}, 1); - } + PublicInbox::Msgmap->new($self->{mainrepo}, 1); }; } @@ -678,8 +626,7 @@ sub _index_sync { my $reindex = $opts->{reindex}; my ($mkey, $last_commit, $lx, $xlog); $self->{git}->batch_prepare; - my $xdb = _xdb_acquire($self); - $xdb->begin_transaction; + my $xdb = $self->begin_txn_lazy; do { $xlog = undef; $mkey = 'last_commit'; @@ -689,6 +636,9 @@ sub _index_sync { $lx = ''; $mkey = undef if $last_commit ne ''; } + $self->{over}->rollback_lazy; + $self->{over}->disconnect; + delete $self->{txn}; $xdb->cancel_transaction; $xdb = _xdb_release($self); @@ -696,8 +646,7 @@ sub _index_sync { my $range = $lx eq '' ? $tip : "$lx..$tip"; $xlog = _git_log($self, $range); - $xdb = _xdb_acquire($self); - $xdb->begin_transaction; + $xdb = $self->begin_txn_lazy; } while ($xdb->get_metadata('last_commit') ne $last_commit); my $mm = _msgmap_init($self); @@ -711,14 +660,12 @@ sub _index_sync { } if (!$mm_only) { $xdb->set_metadata($mkey, $commit) if $mkey && $commit; - $xdb->commit_transaction; - $xdb = _xdb_release($self); + $self->commit_txn_lazy; } # let another process do some work... < if ($more) { if (!$mm_only) { - $xdb = _xdb_acquire($self); - $xdb->begin_transaction; + $xdb = $self->begin_txn_lazy; } $dbh->begin_work if $dbh; } @@ -758,141 +705,68 @@ sub _index_sync { } } -# this will create a ghost as necessary -sub resolve_mid_to_tid { - my ($self, $mid) = @_; - my $tid; - $self->each_smsg_by_mid($mid, sub { - my ($smsg) = @_; - my $cur_tid = $smsg->thread_id; - if (defined $tid) { - merge_threads($self, $tid, $cur_tid); - } else { - $tid = $smsg->thread_id; - } - 1; - }); - return $tid if defined $tid; - - $self->create_ghost($mid)->thread_id; -} - -sub create_ghost { - my ($self, $mid) = @_; - - my $tid = $self->next_thread_id; - my $doc = Search::Xapian::Document->new; - $doc->add_boolean_term('Q' . $mid); - $doc->add_boolean_term('G' . $tid); - $doc->add_boolean_term('T' . 'ghost'); - - my $smsg = PublicInbox::SearchMsg->wrap($doc, $mid); - $self->{xdb}->add_document($doc); - - $smsg; +sub DESTROY { + # order matters for unlocking + $_[0]->{xdb} = undef; + $_[0]->{lockfh} = undef; } -sub merge_threads { - my ($self, $winner_tid, $loser_tid) = @_; - return if $winner_tid == $loser_tid; - my $db = $self->{xdb}; - - my $batch_size = 1000; # don't let @ids grow too large to avoid OOM - while (1) { - my ($head, $tail) = $self->find_doc_ids('G' . $loser_tid); - return if $head == $tail; - my @ids; - for (; $head != $tail && @ids < $batch_size; $head->inc) { - push @ids, $head->get_docid; - } - foreach my $docid (@ids) { - my $doc = $db->get_document($docid); - $doc->remove_term('G' . $loser_tid); - $doc->add_boolean_term('G' . $winner_tid); - $db->replace_document($docid, $doc); - } +# remote_* subs are only used by SearchIdxPart +sub remote_commit { + my ($self) = @_; + if (my $w = $self->{w}) { + print $w "commit\n" or die "failed to write commit: $!"; + } else { + $self->commit_txn_lazy; } } -sub _read_git_config_perm { +sub remote_close { my ($self) = @_; - my @cmd = qw(config); - if ($self->{version} == 2) { - push @cmd, "--file=$self->{mainrepo}/inbox-config"; - } - my $fh = $self->{git}->popen(@cmd, 'core.sharedRepository'); - local $/ = "\n"; - my $perm = <$fh>; - chomp $perm if defined $perm; - $perm; -} - -sub _git_config_perm { - my $self = shift; - my $perm = scalar @_ ? $_[0] : _read_git_config_perm($self); - return PERM_GROUP if (!defined($perm) || $perm eq ''); - return PERM_UMASK if ($perm eq 'umask'); - return PERM_GROUP if ($perm eq 'group'); - if ($perm =~ /\A(?:all|world|everybody)\z/) { - return PERM_EVERYBODY; - } - return PERM_GROUP if ($perm =~ /\A(?:true|yes|on|1)\z/); - return PERM_UMASK if ($perm =~ /\A(?:false|no|off|0)\z/); - - my $i = oct($perm); - return PERM_UMASK if ($i == PERM_UMASK); - return PERM_GROUP if ($i == OLD_PERM_GROUP); - return PERM_EVERYBODY if ($i == OLD_PERM_EVERYBODY); - - if (($i & 0600) != 0600) { - die "core.sharedRepository mode invalid: ". - sprintf('%.3o', $i) . "\nOwner must have permissions\n"; + if (my $w = delete $self->{w}) { + my $pid = delete $self->{pid} or die "no process to wait on\n"; + print $w "close\n" or die "failed to write to pid:$pid: $!\n"; + close $w or die "failed to close pipe for pid:$pid: $!\n"; + waitpid($pid, 0) == $pid or die "remote process did not finish"; + $? == 0 or die ref($self)." pid:$pid exited with: $?"; + } else { + die "transaction in progress $self\n" if $self->{txn}; + $self->_xdb_release if $self->{xdb}; } - ($i & 0666); -} - -sub _umask_for { - my ($perm) = @_; # _git_config_perm return value - my $rv = $perm; - return umask if $rv == 0; - - # set +x bit if +r or +w were set - $rv |= 0100 if ($rv & 0600); - $rv |= 0010 if ($rv & 0060); - $rv |= 0001 if ($rv & 0006); - (~$rv & 0777); } -sub with_umask { - my ($self, $cb) = @_; - my $old = umask $self->{umask}; - my $rv = eval { $cb->() }; - my $err = $@; - umask $old; - die $err if $err; - $rv; +sub remote_remove { + my ($self, $oid, $mid) = @_; + if (my $w = $self->{w}) { + # triggers remove_by_oid in a partition + print $w "D $oid $mid\n" or die "failed to write remove $!"; + } else { + $self->begin_txn_lazy; + $self->remove_by_oid($oid, $mid); + } } -sub DESTROY { - # order matters for unlocking - $_[0]->{xdb} = undef; - $_[0]->{lockfh} = undef; +sub begin_txn_lazy { + my ($self) = @_; + return if $self->{txn}; + my $xdb = $self->{xdb} || $self->_xdb_acquire; + $self->{over}->begin_lazy; + $xdb->begin_transaction; + $self->{txn} = 1; + $xdb; } -# remote_* subs are only used by SearchIdxPart and SearchIdxSkeleton -sub remote_commit { +sub commit_txn_lazy { my ($self) = @_; - print { $self->{w} } "commit\n" or die "failed to write commit: $!"; + delete $self->{txn} or return; + $self->{xdb}->commit_transaction; + $self->{over}->commit_lazy; } -sub remote_close { +sub worker_done { my ($self) = @_; - my $pid = delete $self->{pid} or die "no process to wait on\n"; - my $w = delete $self->{w} or die "no pipe to write to\n"; - print $w "close\n" or die "failed to write to pid:$pid: $!\n"; - close $w or die "failed to close pipe for pid:$pid: $!\n"; - waitpid($pid, 0) == $pid or die "remote process did not finish"; - $? == 0 or die ref($self)." pid:$pid exited with: $?"; + die "$$ $0 xdb not released\n" if $self->{xdb}; + die "$$ $0 still in transaction\n" if $self->{txn}; } 1;