1 # Copyright (C) 2015-2020 all contributors <meta@public-inbox.org>
2 # License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
3 # based on notmuch, but with no concept of folders, files or flags
5 # Indexes mail with Xapian and our (SQLite-based) ::Msgmap for use
6 # with the web and NNTP interfaces. This index maintains thread
7 # relationships for use by PublicInbox::SearchThread.
8 # This writes to the search index.
9 package PublicInbox::SearchIdx;
12 use base qw(PublicInbox::Search PublicInbox::Lock);
13 use PublicInbox::MIME;
14 use PublicInbox::InboxWritable;
15 use PublicInbox::MID qw/mid_clean mid_mime mids_for_index/;
16 use PublicInbox::MsgIter;
18 use POSIX qw(strftime);
19 use PublicInbox::OverIdx;
20 use PublicInbox::Spawn qw(spawn);
21 use PublicInbox::Git qw(git_unquote);
22 use PublicInbox::MsgTime qw(msg_timestamp msg_datestamp);
23 my $X = \%PublicInbox::Search::X;
24 my ($DB_CREATE_OR_OPEN, $DB_OPEN);
26 BATCH_BYTES => defined($ENV{XAPIAN_FLUSH_THRESHOLD}) ?
27 0x7fffffff : 1_000_000,
28 DEBUG => !!$ENV{DEBUG},
31 my $xapianlevels = qr/\A(?:full|medium)\z/;
34 my ($class, $ibx, $creat, $shard) = @_;
35 ref $ibx or die "BUG: expected PublicInbox::Inbox object: $ibx";
36 my $levels = qr/\A(?:full|medium|basic)\z/;
37 my $inboxdir = $ibx->{inboxdir};
38 my $version = $ibx->version;
39 my $indexlevel = 'full';
40 my $altid = $ibx->{altid};
42 require PublicInbox::AltId;
43 $altid = [ map { PublicInbox::AltId->new($ibx, $_); } @$altid ];
45 if ($ibx->{indexlevel}) {
46 if ($ibx->{indexlevel} =~ $levels) {
47 $indexlevel = $ibx->{indexlevel};
49 die("Invalid indexlevel $ibx->{indexlevel}\n");
52 $ibx = PublicInbox::InboxWritable->new($ibx);
54 inboxdir => $inboxdir,
59 indexlevel => $indexlevel,
61 $self->{-set_indexlevel_once} = 1 if $indexlevel eq 'medium';
64 $self->{lock_path} = "$inboxdir/ssoma.lock";
65 my $dir = $self->xdir;
66 $self->{over} = PublicInbox::OverIdx->new("$dir/over.sqlite3");
67 } elsif ($version == 2) {
68 defined $shard or die "shard is required for v2\n";
70 $self->{shard} = $shard;
71 $self->{lock_path} = undef;
73 die "unsupported inbox version=$version\n";
75 $self->{creat} = ($creat || 0) == 1;
79 sub need_xapian ($) { $_[0]->{indexlevel} =~ $xapianlevels }
83 if (need_xapian($self)) {
84 my $xdb = delete $self->{xdb} or croak 'not acquired';
87 $self->lock_release if $self->{creat};
91 sub load_xapian_writable () {
92 return 1 if $X->{WritableDatabase};
93 PublicInbox::Search::load_xapian() or return;
94 my $xap = $PublicInbox::Search::Xap;
95 for (qw(Document TermGenerator WritableDatabase)) {
96 $X->{$_} = $xap.'::'.$_;
98 eval 'require '.$X->{WritableDatabase} or die;
99 *sortable_serialise = $xap.'::sortable_serialise';
100 $DB_CREATE_OR_OPEN = eval($xap.'::DB_CREATE_OR_OPEN()');
101 $DB_OPEN = eval($xap.'::DB_OPEN()');
108 my $dir = $self->xdir;
109 if (need_xapian($self)) {
110 croak 'already acquired' if $self->{xdb};
111 load_xapian_writable();
112 $flag = $self->{creat} ? $DB_CREATE_OR_OPEN : $DB_OPEN;
114 if ($self->{creat}) {
118 # don't create empty Xapian directories if we don't need Xapian
119 my $is_shard = defined($self->{shard});
120 if (!$is_shard || ($is_shard && need_xapian($self))) {
121 File::Path::mkpath($dir);
124 return unless defined $flag;
125 my $xdb = eval { ($X->{WritableDatabase})->new($dir, $flag) };
127 die "Failed opening $dir: ", $@;
133 my ($doc, $col, $num) = @_;
134 $num = sortable_serialise($num);
135 $doc->add_value($col, $num);
138 sub term_generator ($) { # write-only
141 $self->{term_generator} //= do {
142 my $tg = $X->{TermGenerator}->new;
143 $tg->set_stemmer($self->stemmer);
148 sub index_text ($$$$) {
149 my ($self, $text, $wdf_inc, $prefix) = @_;
150 my $tg = term_generator($self); # man Search::Xapian::TermGenerator
152 if ($self->{indexlevel} eq 'full') {
153 $tg->index_text($text, $wdf_inc, $prefix);
154 $tg->increase_termpos;
156 $tg->index_text_without_positions($text, $wdf_inc, $prefix);
160 sub index_users ($$) {
161 my ($self, $smsg) = @_;
163 my $from = $smsg->from;
167 index_text($self, $from, 1, 'A'); # A - author
168 index_text($self, $to, 1, 'XTO') if $to ne '';
169 index_text($self, $cc, 1, 'XCC') if $cc ne '';
172 sub index_diff_inc ($$$$) {
173 my ($self, $text, $pfx, $xnq) = @_;
175 index_text($self, join("\n", @$xnq), 1, 'XNQ');
178 index_text($self, $text, 1, $pfx);
181 sub index_old_diff_fn {
182 my ($self, $seen, $fa, $fb, $xnq) = @_;
184 # no renames or space support for traditional diffs,
185 # find the number of leading common paths to strip:
186 my @fa = split('/', $fa);
187 my @fb = split('/', $fb);
188 while (scalar(@fa) && scalar(@fb)) {
189 $fa = join('/', @fa);
190 $fb = join('/', @fb);
192 unless ($seen->{$fa}++) {
193 index_diff_inc($self, $fa, 'XDFN', $xnq);
203 sub index_diff ($$$) {
204 my ($self, $txt, $doc) = @_;
209 foreach (split(/\n/, $txt)) {
210 if ($in_diff && s/^ //) { # diff context
211 index_diff_inc($self, $_, 'XDFCTX', $xnq);
212 } elsif (/^-- $/) { # email signature begins
214 } elsif (m!^diff --git "?[^/]+/.+ "?[^/]+/.+\z!) {
215 # wait until "---" and "+++" to capture filenames
218 } elsif (m/^diff -(.+) (\S+) (\S+)$/) {
219 my ($opt, $fa, $fb) = ($1, $2, $3);
221 # only support unified:
222 next unless $opt =~ /[uU]/;
223 $in_diff = index_old_diff_fn($self, \%seen, $fa, $fb,
225 } elsif (m!^--- ("?[^/]+/.+)!) {
227 $fn = (split('/', git_unquote($fn), 2))[1];
228 $seen{$fn}++ or index_diff_inc($self, $fn, 'XDFN', $xnq);
230 } elsif (m!^\+\+\+ ("?[^/]+/.+)!) {
232 $fn = (split('/', git_unquote($fn), 2))[1];
233 $seen{$fn}++ or index_diff_inc($self, $fn, 'XDFN', $xnq);
235 } elsif (/^--- (\S+)/) {
238 } elsif (defined $in_diff && /^\+\+\+ (\S+)/) {
239 $in_diff = index_old_diff_fn($self, \%seen, $in_diff,
241 } elsif ($in_diff && s/^\+//) { # diff added
242 index_diff_inc($self, $_, 'XDFB', $xnq);
243 } elsif ($in_diff && s/^-//) { # diff removed
244 index_diff_inc($self, $_, 'XDFA', $xnq);
245 } elsif (m!^index ([a-f0-9]+)\.\.([a-f0-9]+)!) {
246 my ($ba, $bb) = ($1, $2);
247 index_git_blob_id($doc, 'XDFPRE', $ba);
248 index_git_blob_id($doc, 'XDFPOST', $bb);
250 } elsif (/^@@ (?:\S+) (?:\S+) @@\s*$/) {
251 # traditional diff w/o -p
252 } elsif (/^@@ (?:\S+) (?:\S+) @@\s*(\S+.*)$/) {
253 # hunk header context
254 index_diff_inc($self, $1, 'XDFHH', $xnq);
255 # ignore the following lines:
256 } elsif (/^(?:dis)similarity index/ ||
257 /^(?:old|new) mode/ ||
258 /^(?:deleted|new) file mode/ ||
259 /^(?:copy|rename) (?:from|to) / ||
260 /^(?:dis)?similarity index / ||
261 /^\\ No newline at end of file/ ||
262 /^Binary files .* differ/) {
265 # possible to be in diff context, some mail may be
266 # stripped by MUA or even GNU diff(1). "git apply"
267 # treats a bare "\n" as diff context, too
270 warn "non-diff line: $_\n" if DEBUG && $_ ne '';
275 index_text($self, join("\n", @xnq), 1, 'XNQ');
278 sub index_xapian { # msg_iter callback
279 my $part = $_[0]->[0]; # ignore $depth and @idx
280 my ($self, $doc) = @{$_[1]};
281 my $ct = $part->content_type || 'text/plain';
282 my $fn = $part->filename;
283 if (defined $fn && $fn ne '') {
284 index_text($self, $fn, 1, 'XFN');
287 my ($s, undef) = msg_part_text($part, $ct);
288 defined $s or return;
289 $_[0]->[0] = $part = undef; # free memory
291 # split off quoted and unquoted blocks:
292 my @sections = PublicInbox::MsgIter::split_quotes($s);
293 undef $s; # free memory
294 for my $txt (@sections) {
296 index_text($self, $txt, 0, 'XQUOT');
298 # does it look like a diff?
299 if ($txt =~ /^(?:diff|---|\+\+\+) /ms) {
300 index_diff($self, $txt, $doc);
302 index_text($self, $txt, 1, 'XNQ');
305 undef $txt; # free memory
309 sub add_xapian ($$$$) {
310 my ($self, $mime, $smsg, $mids) = @_;
311 $smsg->{mime} = $mime; # XXX dangerous
312 my $hdr = $mime->header_obj;
313 my $doc = $X->{Document}->new;
314 my $subj = $smsg->subject;
315 add_val($doc, PublicInbox::Search::TS(), $smsg->{ts});
316 my @ds = gmtime($smsg->{ds});
317 my $yyyymmdd = strftime('%Y%m%d', @ds);
318 add_val($doc, PublicInbox::Search::YYYYMMDD(), $yyyymmdd);
319 my $dt = strftime('%Y%m%d%H%M%S', @ds);
320 add_val($doc, PublicInbox::Search::DT(), $dt);
322 my $tg = term_generator($self);
324 $tg->set_document($doc);
325 index_text($self, $subj, 1, 'S') if $subj;
326 index_users($self, $smsg);
328 msg_iter($mime, \&index_xapian, [ $self, $doc ]);
329 foreach my $mid (@$mids) {
330 index_text($self, $mid, 1, 'XM');
332 # because too many Message-IDs are prefixed with
334 if ($mid =~ /\w{12,}/) {
335 my @long = ($mid =~ /(\w{3,}+)/g);
336 index_text($self, join(' ', @long), 1, 'XM');
339 $smsg->{to} = $smsg->{cc} = ''; # WWW doesn't need these, only NNTP
340 PublicInbox::OverIdx::parse_references($smsg, $hdr, $mids);
341 my $data = $smsg->to_doc_data;
342 $doc->set_data($data);
343 if (my $altid = $self->{-altid}) {
344 foreach my $alt (@$altid) {
345 my $pfx = $alt->{xprefix};
346 foreach my $mid (@$mids) {
347 my $id = $alt->mid2alt($mid);
348 next unless defined $id;
349 $doc->add_boolean_term($pfx . $id);
353 $doc->add_boolean_term('Q' . $_) foreach @$mids;
354 $self->{xdb}->replace_document($smsg->{num}, $doc);
357 sub _msgmap_init ($) {
359 die "BUG: _msgmap_init is only for v1\n" if $self->{ibx_ver} != 1;
360 $self->{mm} //= eval {
361 require PublicInbox::Msgmap;
362 PublicInbox::Msgmap->new($self->{inboxdir}, 1);
367 # mime = Email::MIME object
368 my ($self, $mime, $smsg) = @_;
369 my $hdr = $mime->header_obj;
370 my $mids = mids_for_index($hdr);
371 $smsg //= bless { blob => '' }, 'PublicInbox::Smsg'; # test-only compat
372 $smsg->{mid} //= $mids->[0]; # v1 compatibility
373 $smsg->{num} //= do { # v1
375 index_mm($self, $mime);
379 $smsg->{ds} //= msg_datestamp($hdr, $self->{autime});
380 $smsg->{ts} //= msg_timestamp($hdr, $self->{cotime});
383 # order matters, overview stores every possible piece of
384 # data in doc_data (deflated). Xapian only stores a subset
385 # of the fields which exist in over.sqlite3. We may stop
386 # storing doc_data in Xapian sometime after we get multi-inbox
388 if (my $over = $self->{over}) { # v1 only
389 $over->add_overview($mime, $smsg);
391 if (need_xapian($self)) {
392 add_xapian($self, $mime, $smsg, $mids);
397 warn "failed to index message <".join('> <',@$mids).">: $@\n";
403 # returns begin and end PostingIterator
405 my ($self, $termval) = @_;
406 my $db = $self->{xdb};
408 ($db->postlist_begin($termval), $db->postlist_end($termval));
413 my ($self, $termval, $cb) = @_;
414 my $batch_size = 1000; # don't let @ids grow too large to avoid OOM
416 my ($head, $tail) = $self->find_doc_ids($termval);
417 return if $head == $tail;
419 for (; $head != $tail && @ids < $batch_size; $head++) {
420 push @ids, $head->get_docid;
426 # v1 only, where $mid is unique
428 my ($self, $mid) = @_;
429 $mid = mid_clean($mid);
431 if (my $over = $self->{over}) {
432 my $nr = eval { $over->remove_oid(undef, $mid) };
434 warn "failed to remove <$mid> from overview: $@\n";
436 warn "<$mid> missing for removal from overview\n";
439 return unless need_xapian($self);
440 my $db = $self->{xdb};
443 batch_do($self, 'Q' . $mid, sub {
445 $db->delete_document($_) for @$ids;
450 warn "failed to remove <$mid> from Xapian: $@\n";
452 warn "<$mid> missing for removal from Xapian\n";
456 # MID is a hint in V2
458 my ($self, $oid, $mid) = @_;
460 $self->{over}->remove_oid($oid, $mid) if $self->{over};
462 return unless need_xapian($self);
463 my $db = $self->{xdb};
465 # XXX careful, we cannot use batch_do here since we conditionally
466 # delete documents based on other factors, so we cannot call
467 # find_doc_ids twice.
468 my ($head, $tail) = $self->find_doc_ids('Q' . $mid);
469 return if $head == $tail;
471 # there is only ONE element in @delete unless we
472 # have bugs in our v2writable deduplication check
474 for (; $head != $tail; $head++) {
475 my $docid = $head->get_docid;
476 my $doc = $db->get_document($docid);
477 my $smsg = PublicInbox::Smsg->wrap($mid);
478 $smsg->load_expand($doc);
479 if ($smsg->{blob} eq $oid) {
480 push(@delete, $docid);
483 $db->delete_document($_) foreach @delete;
487 sub index_git_blob_id {
488 my ($doc, $pfx, $objid) = @_;
490 my $len = length($objid);
491 for (my $len = length($objid); $len >= 7; ) {
492 $doc->add_term($pfx.$objid);
493 $objid = substr($objid, 0, --$len);
498 my ($self, $mime) = @_;
499 my $mid = eval { mid_mime($mime) };
500 $self->remove_message($mid) if defined $mid;
504 my ($self, $mime) = @_;
505 my $mid = mid_mime($mime);
506 my $mm = $self->{mm};
509 if (defined $self->{regen_down}) {
510 $num = $mm->num_for($mid) and return $num;
512 while (($num = $self->{regen_down}--) > 0) {
513 if ($mm->mid_set($num, $mid) != 0) {
517 } elsif (defined $self->{regen_up}) {
518 $num = $mm->num_for($mid) and return $num;
520 # this is to fixup old bugs due to add-remove-add
521 while (($num = ++$self->{regen_up})) {
522 if ($mm->mid_set($num, $mid) != 0) {
528 $num = $mm->mid_insert($mid) and return $num;
530 # fallback to num_for since filters like RubyLang set the number
535 my ($self, $mime) = @_;
536 $self->{mm}->mid_delete(mid_mime($mime));
540 my ($self, $mime, $smsg) = @_;
541 my $num = index_mm($self, $mime);
543 add_message($self, $mime, $smsg);
547 my ($self, $mime) = @_;
548 unindex_blob($self, $mime);
549 unindex_mm($self, $mime);
553 my ($git, $blob, $sizeref) = @_;
554 my $str = $git->cat_file($blob, $sizeref) or
555 die "BUG: $blob not found in $git->{git_dir}";
556 PublicInbox::MIME->new($str);
559 # called by public-inbox-index
561 my ($self, $opts) = @_;
562 delete $self->{lock_path} if $opts->{-skip_lock};
563 $self->{-inbox}->with_umask(sub { $self->_index_sync($opts) })
566 sub batch_adjust ($$$$$) {
567 my ($max, $bytes, $batch_cb, $latest, $nr) = @_;
571 $batch_cb->($nr, $latest);
577 my ($self, $log, $add_cb, $del_cb, $batch_cb) = @_;
578 my $hex = '[a-f0-9]';
579 my $h40 = $hex .'{40}';
580 my $addmsg = qr!^:000000 100644 \S+ ($h40) A\t${hex}{2}/${hex}{38}$!;
581 my $delmsg = qr!^:100644 000000 ($h40) \S+ D\t${hex}{2}/${hex}{38}$!;
582 my $git = $self->{git};
585 my $max = BATCH_BYTES;
591 while (defined($line = <$log>)) {
592 if ($line =~ /$addmsg/o) {
594 if (delete $D{$blob}) {
595 if (defined $self->{regen_down}) {
596 my $num = $self->{regen_down}--;
597 $self->{mm}->num_highwater($num);
601 my $mime = do_cat_mail($git, $blob, \$bytes);
602 my $smsg = bless {}, 'PublicInbox::Smsg';
603 batch_adjust(\$max, $bytes, $batch_cb, $latest, ++$nr);
604 $smsg->{blob} = $blob;
605 $smsg->{bytes} = $bytes;
606 $add_cb->($self, $mime, $smsg);
607 } elsif ($line =~ /$delmsg/o) {
610 } elsif ($line =~ /^commit ($h40)/o) {
613 } elsif ($line =~ /^author .*? ([0-9]+) [\-\+][0-9]+$/) {
614 $self->{autime} = $1;
615 } elsif ($line =~ /^committer .*? ([0-9]+) [\-\+][0-9]+$/) {
616 $self->{cotime} = $1;
619 close($log) or die "git log failed: \$?=$?";
621 foreach my $blob (keys %D) {
622 my $mime = do_cat_mail($git, $blob, \$bytes);
623 $del_cb->($self, $mime);
625 $batch_cb->($nr, $latest, $newest);
629 my ($self, $opts, $range) = @_;
630 my $git = $self->{git};
632 if (index($range, '..') < 0) {
633 # don't show annoying git errors to users who run -index
635 $git->qx(qw(rev-parse -q --verify), "$range^0");
637 open my $fh, '<', '/dev/null' or
638 die "failed to open /dev/null: $!\n";
643 # Count the new files so they can be added newest to oldest
644 # and still have numbers increasing from oldest to newest
646 my $pr = $opts->{-progress};
647 $pr->("counting changes\n\t$range ... ") if $pr;
648 # can't use 'rev-list --count' if we use --diff-filter
649 my $fh = $git->popen(qw(log --pretty=tformat:%h
650 --no-notes --no-color --no-renames
651 --diff-filter=AM), $range);
652 ++$fcount while <$fh>;
653 close $fh or die "git log failed: \$?=$?";
654 my $high = $self->{mm}->num_highwater;
655 $pr->("$fcount\n") if $pr; # continue previous line
656 $self->{ntodo} = $fcount;
658 if (index($range, '..') < 0) {
659 if ($high && $high == $fcount) {
660 # fix up old bugs in full indexes which caused messages to
661 # not appear in Msgmap
662 $self->{regen_up} = $high;
664 # normal regen is for for fresh data
665 $self->{regen_down} = $fcount;
668 # Give oldest messages the smallest numbers
669 $self->{regen_down} = $high + $fcount;
672 $git->popen(qw/log --pretty=raw --no-notes --no-color --no-renames
673 --raw -r --no-abbrev/, $range);
676 # --is-ancestor requires git 1.8.0+
677 sub is_ancestor ($$$) {
678 my ($git, $cur, $tip) = @_;
679 return 0 unless $git->check($cur);
680 my $cmd = [ 'git', "--git-dir=$git->{git_dir}",
681 qw(merge-base --is-ancestor), $cur, $tip ];
682 my $pid = spawn($cmd);
683 waitpid($pid, 0) == $pid or die join(' ', @$cmd) .' did not finish';
687 sub need_update ($$$) {
688 my ($self, $cur, $new) = @_;
689 my $git = $self->{git};
690 return 1 if $cur && !is_ancestor($git, $cur, $new);
691 my $range = $cur eq '' ? $new : "$cur..$new";
692 chomp(my $n = $git->qx(qw(rev-list --count), $range));
693 ($n eq '' || $n > 0);
696 # The last git commit we indexed with Xapian or SQLite (msgmap)
697 # This needs to account for cases where Xapian or SQLite is
698 # out-of-date with respect to the other.
700 my ($self, $mm) = @_;
701 my $lm = $mm->last_commit || '';
703 if (need_xapian($self)) {
704 $lx = $self->{xdb}->get_metadata('last_commit') || '';
708 # Use last_commit from msgmap if it is older or unset
709 if (!$lm || ($lx && $lm && is_ancestor($self->{git}, $lm, $lx))) {
715 sub reindex_from ($$) {
716 my ($reindex, $last_commit) = @_;
717 return $last_commit unless $reindex;
718 ref($reindex) eq 'HASH' ? $reindex->{from} : '';
721 # indexes all unindexed messages (v1 only)
723 my ($self, $opts) = @_;
724 my $tip = $opts->{ref} || 'HEAD';
725 my ($last_commit, $lx, $xlog);
726 my $git = $self->{git};
728 my $pr = $opts->{-progress};
730 my $xdb = $self->begin_txn_lazy;
731 my $mm = _msgmap_init($self);
734 close($xlog) or die "git log failed: \$?=$?";
737 $last_commit = _last_x_commit($self, $mm);
738 $lx = reindex_from($opts->{reindex}, $last_commit);
740 $self->{over}->rollback_lazy;
741 $self->{over}->disconnect;
744 $xdb->cancel_transaction if $xdb;
745 $xdb = _xdb_release($self);
747 # ensure we leak no FDs to "git log" with Xapian <= 1.2
748 my $range = $lx eq '' ? $tip : "$lx..$tip";
749 $xlog = _git_log($self, $opts, $range);
751 $xdb = $self->begin_txn_lazy;
752 } while (_last_x_commit($self, $mm) ne $last_commit);
754 my $dbh = $mm->{dbh} if $mm;
756 my ($nr, $commit, $newest) = @_;
759 my $cur = $mm->last_commit || '';
760 if (need_update($self, $cur, $newest)) {
761 $mm->last_commit($newest);
766 if ($newest && need_xapian($self)) {
767 my $cur = $xdb->get_metadata('last_commit');
768 if (need_update($self, $cur, $newest)) {
769 $xdb->set_metadata('last_commit', $newest);
772 $self->commit_txn_lazy;
774 $xdb = _xdb_release($self);
775 # let another process do some work... <
776 $pr->("indexed $nr/$self->{ntodo}\n") if $pr && $nr;
778 $xdb = $self->begin_txn_lazy;
779 $dbh->begin_work if $dbh;
784 read_log($self, $xlog, *index_both, *unindex_both, $cb);
788 # order matters for unlocking
789 $_[0]->{xdb} = undef;
790 $_[0]->{lockfh} = undef;
793 # remote_* subs are only used by SearchIdxPart
796 if (my $w = $self->{w}) {
797 print $w "commit\n" or die "failed to write commit: $!";
799 $self->commit_txn_lazy;
805 if (my $w = delete $self->{w}) {
806 my $pid = delete $self->{pid} or die "no process to wait on\n";
807 print $w "close\n" or die "failed to write to pid:$pid: $!\n";
808 close $w or die "failed to close pipe for pid:$pid: $!\n";
809 waitpid($pid, 0) == $pid or die "remote process did not finish";
810 $? == 0 or die ref($self)." pid:$pid exited with: $?";
812 die "transaction in progress $self\n" if $self->{txn};
813 $self->_xdb_release if $self->{xdb};
818 my ($self, $oid, $mid) = @_;
819 if (my $w = $self->{w}) {
820 # triggers remove_by_oid in a shard
821 print $w "D $oid $mid\n" or die "failed to write remove $!";
823 $self->begin_txn_lazy;
824 $self->remove_by_oid($oid, $mid);
830 return if $self->{txn};
832 $self->{-inbox}->with_umask(sub {
833 my $xdb = $self->{xdb} || $self->_xdb_acquire;
834 $self->{over}->begin_lazy if $self->{over};
835 $xdb->begin_transaction if $xdb;
841 # store 'indexlevel=medium' in v2 shard=0 and v1 (only one shard)
842 # This metadata is read by Admin::detect_indexlevel:
846 if (!$self->{shard} && # undef or 0, not >0
847 delete($self->{-set_indexlevel_once})) {
848 my $xdb = $self->{xdb};
849 my $level = $xdb->get_metadata('indexlevel');
850 if (!$level || $level ne 'medium') {
851 $xdb->set_metadata('indexlevel', 'medium');
856 sub commit_txn_lazy {
858 delete $self->{txn} or return;
859 $self->{-inbox}->with_umask(sub {
860 if (my $xdb = $self->{xdb}) {
861 set_indexlevel($self);
862 $xdb->commit_transaction;
864 $self->{over}->commit_lazy if $self->{over};
870 if (need_xapian($self)) {
871 die "$$ $0 xdb not released\n" if $self->{xdb};
873 die "$$ $0 still in transaction\n" if $self->{txn};