X-Git-Url: http://www.git.stargrave.org/?p=public-inbox.git;a=blobdiff_plain;f=lib%2FPublicInbox%2FSearchIdx.pm;h=2239c90209579e42c590c8e0df113ab53617879b;hp=2e0b9a43e24750fce0e083c37120a47f34d8da8a;hb=017fed7bc4d33ac474a19356994be5bd0bfe68ba;hpb=7503aeb540af5afd5cb1b554b3c29f35f5fc918d diff --git a/lib/PublicInbox/SearchIdx.pm b/lib/PublicInbox/SearchIdx.pm index 2e0b9a43..2239c902 100644 --- a/lib/PublicInbox/SearchIdx.pm +++ b/lib/PublicInbox/SearchIdx.pm @@ -12,14 +12,17 @@ use warnings; use base qw(PublicInbox::Search PublicInbox::Lock); use PublicInbox::MIME; use PublicInbox::InboxWritable; -use PublicInbox::MID qw/mid_clean id_compress mid_mime mids references/; +use PublicInbox::MID qw/mid_clean id_compress mid_mime mids/; use PublicInbox::MsgIter; use Carp qw(croak); use POSIX qw(strftime); +use PublicInbox::OverIdx; +use PublicInbox::Spawn qw(spawn); require PublicInbox::Git; +use Compress::Zlib qw(compress); use constant { - BATCH_BYTES => 1_000_000, + BATCH_BYTES => 10_000_000, DEBUG => !!$ENV{DEBUG}, }; @@ -73,12 +76,13 @@ sub new { $ibx->umask_prepare; if ($version == 1) { $self->{lock_path} = "$mainrepo/ssoma.lock"; + my $dir = $self->xdir; + $self->{over} = PublicInbox::OverIdx->new("$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"; } @@ -114,14 +118,6 @@ sub add_val ($$$) { $doc->add_value($col, $num); } -sub add_values { - my ($doc, $ts, $ds, $num) = @_; - add_val($doc, PublicInbox::Search::TS, $ts); - my $yyyymmdd = strftime('%Y%m%d', gmtime($ds)); - add_val($doc, PublicInbox::Search::YYYYMMDD, $yyyymmdd); - defined($num) and add_val($doc, PublicInbox::Search::NUM, $num); -} - sub index_users ($$) { my ($tg, $smsg) = @_; @@ -269,23 +265,21 @@ sub add_message { my ($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 + $self->_msgmap_init; + $num = index_mm($self, $mime); + } eval { my $smsg = PublicInbox::SearchMsg->new($mime); my $doc = $smsg->{doc}; my $subj = $smsg->subject; - my $xpath; - if ($subj ne '') { - $xpath = $self->subject_path($subj); - $xpath = id_compress($xpath); - } - - my $lines = $mime->body_raw =~ tr!\n!\n!; - $smsg->{lines} = $mime->body_raw =~ tr!\n!\n!; - defined $bytes or $bytes = length($mime->as_string); - $smsg->{bytes} = $bytes; - add_values($doc, $smsg->ts, $smsg->ds, $num); + add_val($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 $tg = $self->term_generator; @@ -334,13 +328,12 @@ sub add_message { index_body($tg, \@orig, $doc) if @orig; }); - # populates smsg->references for smsg->to_doc_data - my $refs = parse_references($smsg); - $mid0 = $mids->[0] unless defined $mid0; # v1 compatibility - my $data = $smsg->to_doc_data($oid, $mid0); foreach my $mid (@$mids) { $tg->index_text($mid, 1, 'XM'); } + $smsg->{to} = $smsg->{cc} = ''; + PublicInbox::OverIdx::parse_references($smsg, $mid0, $mids); + my $data = $smsg->to_doc_data($oid, $mid0); $doc->set_data($data); if (my $altid = $self->{-altid}) { foreach my $alt (@$altid) { @@ -353,17 +346,11 @@ sub add_message { } } - $self->delete_article($num) if defined $num; # for reindexing - if ($skel) { - my @vals = ($smsg->ts, $num, $mids, $xpath, $data); - $skel->index_skeleton(\@vals); - $doc->add_boolean_term('Q' . $_) foreach @$mids; - $doc->add_boolean_term('XNUM' . $num) if defined $num; - $doc_id = $self->{xdb}->add_document($doc); - } else { - $doc_id = link_and_save($self, $doc, $mids, $refs, - $num, $xpath); + if (my $over = $self->{over}) { + $over->add_overview($mime, $bytes, $num, $oid, $mid0); } + $doc->add_boolean_term('Q' . $_) foreach @$mids; + $self->{xdb}->replace_document($doc_id = $num, $doc); }; if ($@) { @@ -400,11 +387,13 @@ sub remove_message { my $db = $self->{xdb}; my $called; $mid = mid_clean($mid); + my $over = $self->{over}; eval { batch_do($self, 'Q' . $mid, sub { my ($ids) = @_; $db->delete_document($_) for @$ids; + $over->delete_articles($ids) if $over; $called = 1; }); }; @@ -415,21 +404,13 @@ sub remove_message { } } -sub delete_article { - my ($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}; + $self->{over}->remove_oid($oid, $mid) if $self->{over}; + # 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. @@ -444,7 +425,9 @@ sub remove_by_oid { my $doc = $db->get_document($docid); my $smsg = PublicInbox::SearchMsg->wrap($doc, $mid); $smsg->load_expand; - push(@delete, $docid) if $smsg->{blob} eq $oid; + if ($smsg->{blob} eq $oid) { + push(@delete, $docid); + } } $db->delete_document($_) foreach @delete; scalar(@delete); @@ -462,105 +445,6 @@ 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) = @_; - my $mime = $smsg->{mime}; - my $hdr = $mime->header_obj; - my $refs = references($hdr); - return $refs if scalar(@$refs) == 0; - - # prevent circular references via References here: - my %mids = map { $_ => 1 } @{mids($hdr)}; - my @keep; - foreach my $ref (@$refs) { - if (length($ref) > PublicInbox::MID::MAX_MID_SIZE) { - warn "References: <$ref> too long, ignoring\n"; - next; - } - next if $mids{$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; - $doc->add_boolean_term('Q' . $_) foreach @$mids; - - $self->{skel} and die "Should not have read-only skel here\n";; - foreach my $mid (@$mids) { - my $vivified = 0; - $self->each_smsg_by_mid($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; - }); - $vivified > 1 and warn - "BUG: vivified multiple ($vivified) ghosts for $mid\n"; - } - # 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, $objid) = @_; @@ -581,8 +465,28 @@ sub index_mm { my ($self, $mime) = @_; my $mid = mid_clean(mid_mime($mime)); my $mm = $self->{mm}; - my $num = $mm->mid_insert($mid); - return $num if defined $num; + my $num; + + if (defined $self->{regen_down}) { + $num = $mm->num_for($mid) and return $num; + + while (($num = $self->{regen_down}--) > 0) { + if ($mm->mid_set($num, $mid) != 0) { + return $num; + } + } + } elsif (defined $self->{regen_up}) { + $num = $mm->num_for($mid) and return $num; + + # this is to fixup old bugs due to add-remove-add + while (($num = ++$self->{regen_up})) { + if ($mm->mid_set($num, $mid) != 0) { + return $num; + } + } + } + + $num = $mm->mid_insert($mid) and return $num; # fallback to num_for since filters like RubyLang set the number $mm->num_for($mid); @@ -593,18 +497,6 @@ sub unindex_mm { $self->{mm}->mid_delete(mid_clean(mid_mime($mime))); } -sub index_mm2 { - my ($self, $mime, $bytes, $blob) = @_; - my $num = $self->{mm}->num_for(mid_clean(mid_mime($mime))); - add_message($self, $mime, $bytes, $num, $blob); -} - -sub unindex_mm2 { - my ($self, $mime) = @_; - $self->{mm}->mid_delete(mid_clean(mid_mime($mime))); - unindex_blob($self, $mime); -} - sub index_both { my ($self, $mime, $bytes, $blob) = @_; my $num = index_mm($self, $mime); @@ -638,12 +530,12 @@ sub batch_adjust ($$$$) { $$max -= $bytes; if ($$max <= 0) { $$max = BATCH_BYTES; - $batch_cb->($latest, 1); + $batch_cb->($latest); } } # only for v1 -sub rlog { +sub read_log { my ($self, $log, $add_cb, $del_cb, $batch_cb) = @_; my $hex = '[a-f0-9]'; my $h40 = $hex .'{40}'; @@ -654,53 +546,108 @@ sub rlog { my $bytes; my $max = BATCH_BYTES; local $/ = "\n"; + my %D; my $line; + my $newest; + my $mid = '20170114215743.5igbjup6qpsh3jfg@genre.crustytoothpaste.net'; while (defined($line = <$log>)) { if ($line =~ /$addmsg/o) { my $blob = $1; + delete $D{$blob} and next; my $mime = do_cat_mail($git, $blob, \$bytes) or next; + my $mids = mids($mime->header_obj); + foreach (@$mids) { + warn "ADD $mid\n" if ($_ eq $mid); + } batch_adjust(\$max, $bytes, $batch_cb, $latest); $add_cb->($self, $mime, $bytes, $blob); } elsif ($line =~ /$delmsg/o) { my $blob = $1; - my $mime = do_cat_mail($git, $blob, \$bytes) or next; - batch_adjust(\$max, $bytes, $batch_cb, $latest); - $del_cb->($self, $mime); + $D{$blob} = 1; } elsif ($line =~ /^commit ($h40)/o) { $latest = $1; + $newest ||= $latest; } } - $batch_cb->($latest, 0); + # get the leftovers + foreach my $blob (keys %D) { + my $mime = do_cat_mail($git, $blob, \$bytes) or next; + my $mids = mids($mime->header_obj); + foreach (@$mids) { + warn "DEL $mid\n" if ($_ eq $mid); + } + $del_cb->($self, $mime); + } + $batch_cb->($latest, $newest); } sub _msgmap_init { my ($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); }; } sub _git_log { my ($self, $range) = @_; - $self->{git}->popen(qw/log --reverse --no-notes --no-color + my $git = $self->{git}; + + if (index($range, '..') < 0) { + my $regen_max = 0; + # can't use 'rev-list --count' if we use --diff-filter + my $fh = $git->popen(qw(log --pretty=tformat:%h + --no-notes --no-color --no-renames + --diff-filter=AM), $range); + ++$regen_max while <$fh>; + my (undef, $max) = $self->{mm}->minmax; + + if ($max && $max == $regen_max) { + # fix up old bugs in full indexes which caused messages to + # not appear in Msgmap + $self->{regen_up} = $max; + } else { + # normal regen is for for fresh data + $self->{regen_down} = $regen_max; + } + } + + $git->popen(qw/log --no-notes --no-color --no-renames --raw -r --no-abbrev/, $range); } -# indexes all unindexed messages +sub is_ancestor ($$$) { + my ($git, $cur, $tip) = @_; + return 0 unless $git->check($cur); + my $cmd = [ 'git', "--git-dir=$git->{git_dir}", + qw(merge-base --is-ancestor), $cur, $tip ]; + my $pid = spawn($cmd); + defined $pid or die "spawning ".join(' ', @$cmd)." failed: $!"; + waitpid($pid, 0) == $pid or die join(' ', @$cmd) .' did not finish'; + $? == 0; +} + +sub need_update ($$$) { + my ($self, $cur, $new) = @_; + my $git = $self->{git}; + return 1 if $cur && !is_ancestor($git, $cur, $new); + my $range = $cur eq '' ? $new : "$cur..$new"; + chomp(my $n = $git->qx(qw(rev-list --count), $range)); + ($n eq '' || $n > 0); +} + +# indexes all unindexed messages (v1 only) sub _index_sync { my ($self, $opts) = @_; my $tip = $opts->{ref} || 'HEAD'; my $reindex = $opts->{reindex}; my ($mkey, $last_commit, $lx, $xlog); - $self->{git}->batch_prepare; - my $xdb = _xdb_acquire($self); - $xdb->begin_transaction; + my $git = $self->{git}; + $git->batch_prepare; + + my $xdb = $self->begin_txn_lazy; + my $mm = _msgmap_init($self); do { $xlog = undef; $mkey = 'last_commit'; @@ -710,122 +657,54 @@ sub _index_sync { $lx = ''; $mkey = undef if $last_commit ne ''; } + + # use last_commit from msgmap if it is older or unset + my $lm = $mm->last_commit || ''; + if (!$lm || ($lm && $lx && is_ancestor($git, $lm, $lx))) { + $lx = $lm; + } + + $self->{over}->rollback_lazy; + $self->{over}->disconnect; + delete $self->{txn}; $xdb->cancel_transaction; $xdb = _xdb_release($self); - # ensure we leak no FDs to "git log" + # ensure we leak no FDs to "git log" with Xapian <= 1.2 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); my $dbh = $mm->{dbh} if $mm; - my $mm_only; my $cb = sub { - my ($commit, $more) = @_; + my ($commit, $newest) = @_; if ($dbh) { - $mm->last_commit($commit) if $commit; + if ($newest) { + my $cur = $mm->last_commit || ''; + if (need_update($self, $cur, $newest)) { + $mm->last_commit($newest); + } + } $dbh->commit; } - if (!$mm_only) { - $xdb->set_metadata($mkey, $commit) if $mkey && $commit; - $xdb->commit_transaction; - $xdb = _xdb_release($self); + if ($mkey && $newest) { + my $cur = $xdb->get_metadata($mkey); + if (need_update($self, $cur, $newest)) { + $xdb->set_metadata($mkey, $newest); + } } + $self->commit_txn_lazy; # let another process do some work... < - if ($more) { - if (!$mm_only) { - $xdb = _xdb_acquire($self); - $xdb->begin_transaction; - } + if (!$newest) { + $xdb = $self->begin_txn_lazy; $dbh->begin_work if $dbh; } }; - if ($mm) { - $dbh->begin_work; - my $lm = $mm->last_commit || ''; - if ($lm eq $lx) { - # Common case is the indexes are synced, - # we only need to run git-log once: - rlog($self, $xlog, *index_both, *unindex_both, $cb); - } else { - # Uncommon case, msgmap and xapian are out-of-sync - # do not care for performance (but git is fast :>) - # This happens if we have to reindex Xapian since - # msgmap is a frozen format and our Xapian format - # is evolving. - my $r = $lm eq '' ? $tip : "$lm..$tip"; - - # first, ensure msgmap is up-to-date: - my $mkey_prev = $mkey; - $mkey = undef; # ignore xapian, for now - my $mlog = _git_log($self, $r); - $mm_only = 1; - rlog($self, $mlog, *index_mm, *unindex_mm, $cb); - $mm_only = $mlog = undef; - - # now deal with Xapian - $mkey = $mkey_prev; - $dbh = undef; - rlog($self, $xlog, *index_mm2, *unindex_mm2, $cb); - } - } else { - # user didn't install DBD::SQLite and DBI - rlog($self, $xlog, *add_message, *unindex_blob, $cb); - } -} - -# this will create a ghost as necessary -sub resolve_mid_to_tid { - my ($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); - - $smsg; -} - -sub merge_threads { - my ($self, $winner_tid, $loser_tid) = @_; - return if $winner_tid == $loser_tid; - my $db = $self->{xdb}; - batch_do($self, 'G' . $loser_tid, sub { - my ($ids) = @_; - 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); - } - }); + $dbh->begin_work; + read_log($self, $xlog, *index_both, *unindex_both, $cb); } sub DESTROY { @@ -834,16 +713,13 @@ sub DESTROY { $_[0]->{lockfh} = undef; } -# remote_* subs are only used by SearchIdxPart and SearchIdxSkeleton +# remote_* subs are only used by SearchIdxPart sub remote_commit { my ($self) = @_; if (my $w = $self->{w}) { print $w "commit\n" or die "failed to write commit: $!"; } else { $self->commit_txn_lazy; - if (my $skel = $self->{skeleton}) { - $skel->commit_txn_lazy; - } } } @@ -864,7 +740,7 @@ sub remote_close { sub remote_remove { my ($self, $oid, $mid) = @_; if (my $w = $self->{w}) { - # triggers remove_by_oid in partition or skeleton + # triggers remove_by_oid in a partition print $w "D $oid $mid\n" or die "failed to write remove $!"; } else { $self->begin_txn_lazy; @@ -876,14 +752,17 @@ sub begin_txn_lazy { my ($self) = @_; return if $self->{txn}; my $xdb = $self->{xdb} || $self->_xdb_acquire; + $self->{over}->begin_lazy if $self->{over}; $xdb->begin_transaction; $self->{txn} = 1; + $xdb; } sub commit_txn_lazy { my ($self) = @_; delete $self->{txn} or return; $self->{xdb}->commit_transaction; + $self->{over}->commit_lazy if $self->{over}; } sub worker_done {