1 # Copyright (C) 2018-2021 all contributors <meta@public-inbox.org>
2 # License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
4 # This interface wraps and mimics PublicInbox::Import
5 # Used to write to V2 inboxes (see L<public-inbox-v2-format(5)>).
6 package PublicInbox::V2Writable;
9 use parent qw(PublicInbox::Lock PublicInbox::IPC);
10 use PublicInbox::SearchIdxShard;
14 use PublicInbox::Import;
15 use PublicInbox::MultiGit;
16 use PublicInbox::MID qw(mids references);
17 use PublicInbox::ContentHash qw(content_hash content_digest git_sha);
18 use PublicInbox::InboxWritable;
19 use PublicInbox::OverIdx;
20 use PublicInbox::Msgmap;
21 use PublicInbox::Spawn qw(spawn popen_rd run_die);
22 use PublicInbox::Search;
23 use PublicInbox::SearchIdx qw(log2stack is_ancestor check_size is_bad_blob);
24 use IO::Handle; # ->autoflush
28 my $OID = qr/[a-f0-9]{40,}/;
29 # an estimate of the post-packed size to the raw uncompressed size
30 our $PACKING_FACTOR = 0.4;
32 # SATA storage lags behind what CPUs are capable of, so relying on
33 # nproc(1) can be misleading and having extra Xapian shards is a
34 # waste of FDs and space. It can also lead to excessive IO latency
35 # and slow things down. Users on NVME or other fast storage can
36 # use the NPROC env or switches in our script/public-inbox-* programs
37 # to increase Xapian shards
38 our $NPROC_MAX_DEFAULT = 4;
40 sub nproc_shards ($) {
42 my $n = $creat_opt->{nproc} if ref($creat_opt) eq 'HASH';
45 # assume 2 cores if not detectable or zero
46 state $NPROC_DETECTED = PublicInbox::IPC::detect_nproc() || 2;
48 $n = $NPROC_MAX_DEFAULT if $n > $NPROC_MAX_DEFAULT;
51 # subtract for the main process and git-fast-import
56 sub count_shards ($) {
58 # always load existing shards in case core count changes:
59 # Also, shard count may change while -watch is running
60 if (my $ibx = $self->{ibx}) {
61 my $srch = $ibx->search or return 0;
62 delete $ibx->{search};
64 } else { # ExtSearchIdx
65 $self->{nshard} = scalar($self->xdb_shards_flat);
70 # $creat may be any true value, or 0/undef. A hashref is true,
71 # and $creat->{nproc} may be set to an integer
72 my ($class, $v2ibx, $creat) = @_;
73 $v2ibx = PublicInbox::InboxWritable->new($v2ibx);
74 my $dir = $v2ibx->assert_usable_dir;
76 die "$dir does not exist\n" if !$creat;
78 File::Path::mkpath($dir);
80 my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION;
83 mg => PublicInbox::MultiGit->new($dir, 'all.git', 'git'),
84 im => undef, # PublicInbox::Import
90 oidx => PublicInbox::OverIdx->new("$xpfx/over.sqlite3"),
91 lock_path => "$dir/inbox.lock",
92 # limit each git repo (epoch) to 1GB or so
93 rotate_bytes => int((1024 * 1024 * 1024) / $PACKING_FACTOR),
94 last_commit => [], # git epoch -> commit
96 $self->{oidx}->{-no_fsync} = 1 if $v2ibx->{-no_fsync};
97 $self->{shards} = count_shards($self) || nproc_shards($creat);
103 my ($self, $shards, $skip_epoch, $skip_artnum) = @_;
104 if (defined $shards) {
105 $self->{parallel} = 0 if $shards == 0;
106 $self->{shards} = $shards if $shards > 0;
109 $self->{mm}->skip_artnum($skip_artnum) if defined $skip_artnum;
110 my $max = $self->{ibx}->max_git_epoch;
111 $max = $skip_epoch if (defined($skip_epoch) && !defined($max));
112 $self->{mg}->add_epoch($max // 0);
116 # returns undef on duplicate or spam
117 # mimics Import::add and wraps it for v2
119 my ($self, $eml, $check_cb) = @_;
120 $self->{ibx}->with_umask(\&_add, $self, $eml, $check_cb);
124 my ($self, $num) = @_;
125 $self->{idx_shards}->[$num % scalar(@{$self->{idx_shards}})];
128 # indexes a message, returns true if checkpointing is needed
130 my ($self, $eml, $smsg) = @_;
131 $self->{oidx}->add_overview($eml, $smsg);
132 if ($self->{-need_xapian}) {
133 my $idx = idx_shard($self, $smsg->{num});
134 $idx->index_eml($eml, $smsg);
136 my $n = $self->{transact_bytes} += $smsg->{bytes};
137 $n >= $self->{batch_bytes};
141 my ($self, $mime, $check_cb) = @_;
145 $mime = $check_cb->($mime, $self->{ibx}) or return;
148 # All pipes (> $^F) known to Perl 5.6+ have FD_CLOEXEC set,
149 # as does SQLite 3.4.1+ (released in 2007-07-20), and
150 # Xapian 1.3.2+ (released 2015-03-15).
151 # For the most part, we can spawn git-fast-import without
152 # leaking FDs to it...
155 my ($num, $mid0) = v2_num_for($self, $mime);
156 defined $num or return; # duplicate
157 defined $mid0 or die "BUG: \$mid0 undefined\n";
158 my $im = $self->importer;
159 my $smsg = bless { mid => $mid0, num => $num }, 'PublicInbox::Smsg';
160 my $cmt = $im->add($mime, undef, $smsg); # sets $smsg->{ds|ts|blob}
161 $cmt = $im->get_mark($cmt);
162 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
164 if (do_idx($self, $mime, $smsg)) {
172 my ($self, $mime) = @_;
173 my $mids = mids($mime);
175 my $mid = $mids->[0];
176 my $num = $self->{mm}->mid_insert($mid);
177 if (defined $num) { # common case
181 # crap, Message-ID is already known, hope somebody just resent:
182 foreach my $m (@$mids) {
183 # read-only lookup now safe to do after above barrier
184 # easy, don't store duplicates
185 # note: do not add more diagnostic info here since
186 # it gets noisy on public-inbox-watch restarts
187 return () if content_exists($self, $mime, $m);
190 # AltId may pre-populate article numbers (e.g. X-Mail-Count
191 # or NNTP article number), use that article number if it's
193 my $altid = $self->{ibx}->{altid};
194 if ($altid && grep(/:file=msgmap\.sqlite3\z/, @$altid)) {
195 my $num = $self->{mm}->num_for($mid);
197 if (defined $num && !$self->{oidx}->get_art($num)) {
203 warn "<$mid> reused for mismatched content\n";
205 # try the rest of the mids
206 for(my $i = $#$mids; $i >= 1; $i--) {
208 $num = $self->{mm}->mid_insert($m);
210 warn "alternative <$m> for <$mid> found\n";
215 # none of the existing Message-IDs are good, generate a new one:
216 v2_num_for_harder($self, $mime);
219 sub v2_num_for_harder {
220 my ($self, $eml) = @_;
222 my $dig = content_digest($eml);
223 my $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
224 my $num = $self->{mm}->mid_insert($mid0);
225 unless (defined $num) {
226 # it's hard to spoof the last Received: header
227 my @recvd = $eml->header_raw('Received');
228 $dig->add("Received: $_") foreach (@recvd);
229 $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
230 $num = $self->{mm}->mid_insert($mid0);
232 # fall back to a random Message-ID and give up determinism:
233 until (defined($num)) {
235 $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
236 warn "using random Message-ID <$mid0> as fallback\n";
237 $num = $self->{mm}->mid_insert($mid0);
240 PublicInbox::Import::append_mid($eml, $mid0);
244 sub _idx_init { # with_umask callback
245 my ($self, $opt) = @_;
246 $self->lock_acquire unless $opt && $opt->{-skip_lock};
247 $self->{oidx}->create;
249 # xcpdb can change shard count while -watch is idle
250 my $nshards = count_shards($self);
251 $self->{shards} = $nshards if $nshards && $nshards != $self->{shards};
252 $self->{batch_bytes} = $opt->{batch_size} //
253 $PublicInbox::SearchIdx::BATCH_BYTES;
255 # need to create all shards before initializing msgmap FD
256 # idx_shards must be visible to all forked processes
257 my $max = $self->{shards} - 1;
258 my $idx = $self->{idx_shards} = [];
259 push @$idx, PublicInbox::SearchIdxShard->new($self, $_) for (0..$max);
260 $self->{-need_xapian} = $idx->[0]->need_xapian;
262 # SearchIdxShard may do their own flushing, so don't scale
263 # until after forking
264 $self->{batch_bytes} *= $self->{shards} if $self->{parallel};
266 my $ibx = $self->{ibx} or return; # ExtIdxSearch
268 # Now that all subprocesses are up, we can open the FDs
270 my $mm = $self->{mm} = PublicInbox::Msgmap->new_file(
271 "$ibx->{inboxdir}/msgmap.sqlite3",
272 $ibx->{-no_fsync} ? 2 : 1);
273 $mm->{dbh}->begin_work;
276 sub parallel_init ($$) {
277 my ($self, $indexlevel) = @_;
278 $self->{parallel} = 0 if ($indexlevel // 'full') eq 'basic';
283 my ($self, $opt) = @_;
284 return if $self->{idx_shards};
285 my $ibx = $self->{ibx};
287 # do not leak read-only FDs to child processes, we only have these
288 # FDs for duplicate detection so they should not be
289 # frequently activated.
290 delete @$ibx{qw(mm search)};
293 parallel_init($self, $ibx->{indexlevel});
294 $ibx->with_umask(\&_idx_init, $self, $opt);
297 # returns an array mapping [ epoch => latest_commit ]
298 # latest_commit may be undef if nothing was done to that epoch
299 # $replace_map = { $object_id => $strref, ... }
300 sub _replace_oids ($$$) {
301 my ($self, $mime, $replace_map) = @_;
303 my $ibx = $self->{ibx};
304 my $pfx = "$ibx->{inboxdir}/git";
305 my $rewrites = []; # epoch => commit
306 my $max = $self->{epoch_max} //= $ibx->max_git_epoch // return;
308 foreach my $i (0..$max) {
309 my $git_dir = "$pfx/$i.git";
311 my $git = PublicInbox::Git->new($git_dir);
312 my $im = $self->import_init($git, 0, 1);
313 $rewrites->[$i] = $im->replace_oids($mime, $replace_map);
319 sub content_hashes ($) {
321 my @chashes = ( content_hash($mime) );
323 # We still support Email::MIME, here, and
324 # Email::MIME->as_string doesn't always round-trip, so we may
325 # use a second content_hash
326 my $rt = content_hash(PublicInbox::Eml->new(\($mime->as_string)));
327 push @chashes, $rt if $chashes[0] ne $rt;
331 sub content_matches ($$) {
332 my ($chashes, $existing) = @_;
333 my $chash = content_hash($existing);
334 foreach (@$chashes) {
335 return 1 if $_ eq $chash
340 # used for removing or replacing (purging)
341 sub rewrite_internal ($$;$$$) {
342 my ($self, $old_eml, $cmt_msg, $new_eml, $sref) = @_;
344 my ($im, $need_reindex, $replace_map);
346 $replace_map = {}; # oid => sref
347 $need_reindex = [] if $new_eml;
349 $im = $self->importer;
351 my $oidx = $self->{oidx};
352 my $chashes = content_hashes($old_eml);
354 my $mids = mids($old_eml);
356 # We avoid introducing new blobs into git since the raw content
357 # can be slightly different, so we do not need the user-supplied
358 # message now that we have the mids and content_hash
362 foreach my $mid (@$mids) {
363 my %gone; # num => [ smsg, $mime, raw ]
365 while (my $smsg = $oidx->next_by_mid($mid, \$id, \$prev)) {
366 my $msg = get_blob($self, $smsg);
367 if (!defined($msg)) {
368 warn "broken smsg for $mid\n";
372 my $cur = PublicInbox::Eml->new($msg);
373 if (content_matches($chashes, $cur)) {
374 $gone{$smsg->{num}} = [ $smsg, $cur, \$orig ];
377 my $n = scalar keys %gone;
380 warn "BUG: multiple articles linked to <$mid>\n",
381 join(',', sort keys %gone), "\n";
383 foreach my $num (keys %gone) {
384 my ($smsg, $mime, $orig) = @{$gone{$num}};
385 # $removed should only be set once assuming
386 # no bugs in our deduplication code:
387 $removed = [ undef, $mime, $smsg ];
388 my $oid = $smsg->{blob};
390 $replace_map->{$oid} = $sref;
392 ($mark, undef) = $im->remove($orig, $cmt_msg);
393 $removed->[0] = $mark;
396 if ($need_reindex) { # ->replace
397 push @$need_reindex, $smsg;
398 } else { # ->purge or ->remove
399 $self->{mm}->num_delete($num);
401 unindex_oid_aux($self, $oid, $mid);
406 my $cmt = $im->get_mark($mark);
407 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
409 if ($replace_map && scalar keys %$replace_map) {
410 my $rewrites = _replace_oids($self, $new_eml, $replace_map);
411 return { rewrites => $rewrites, need_reindex => $need_reindex };
413 defined($mark) ? $removed : undef;
416 # public (see PublicInbox::Import->remove), but note the 3rd element
417 # (retval[2]) is not part of the stable API shared with Import->remove
419 my ($self, $eml, $cmt_msg) = @_;
420 my $r = $self->{ibx}->with_umask(\&rewrite_internal,
421 $self, $eml, $cmt_msg);
422 defined($r) && defined($r->[0]) ? @$r: undef;
425 sub _replace ($$;$$) {
426 my ($self, $old_eml, $new_eml, $sref) = @_;
427 my $arg = [ $self, $old_eml, undef, $new_eml, $sref ];
428 my $rewritten = $self->{ibx}->with_umask(\&rewrite_internal,
429 $self, $old_eml, undef, $new_eml, $sref) or return;
431 my $rewrites = $rewritten->{rewrites};
432 # ->done is called if there are rewrites since we gc+prune from git
433 $self->idx_init if @$rewrites;
435 for my $i (0..$#$rewrites) {
436 defined(my $cmt = $rewrites->[$i]) or next;
437 $self->{last_commit}->[$i] = $cmt;
444 my ($self, $mime) = @_;
445 my $rewritten = _replace($self, $mime, undef, \'') or return;
446 $rewritten->{rewrites}
449 sub _check_mids_match ($$$) {
450 my ($old_list, $new_list, $hdrs) = @_;
451 my %old_mids = map { $_ => 1 } @$old_list;
452 my %new_mids = map { $_ => 1 } @$new_list;
453 my @old = keys %old_mids;
454 my @new = keys %new_mids;
455 my $err = "$hdrs may not be changed when replacing\n";
456 die $err if scalar(@old) != scalar(@new);
457 delete @new_mids{@old};
458 delete @old_mids{@new};
459 die $err if (scalar(keys %old_mids) || scalar(keys %new_mids));
462 # Changing Message-IDs or References with ->replace isn't supported.
463 # The rules for dealing with messages with multiple or conflicting
464 # Message-IDs are pretty complex and rethreading hasn't been fully
466 sub check_mids_match ($$) {
467 my ($old, $new) = @_;
468 _check_mids_match(mids($old), mids($new), 'Message-ID(s)');
469 _check_mids_match(references($old), references($new),
470 'References/In-Reply-To');
475 my ($self, $old_mime, $new_mime) = @_;
477 check_mids_match($old_mime, $new_mime);
479 # mutt will always add Content-Length:, Status:, Lines: when editing
480 PublicInbox::Import::drop_unwanted_headers($new_mime);
482 my $raw = $new_mime->as_string;
483 my $expect_oid = git_sha(1, \$raw)->hexdigest;
484 my $rewritten = _replace($self, $old_mime, $new_mime, \$raw) or return;
485 my $need_reindex = $rewritten->{need_reindex};
487 # just in case we have bugs in deduplication code:
488 my $n = scalar(@$need_reindex);
490 my $list = join(', ', map {
491 "$_->{num}: <$_->{mid}>"
494 W: rewritten $n messages matching content of original message (expected: 1).
495 W: possible bug in public-inbox, NNTP article IDs and Message-IDs follow:
500 # make sure we really got the OID:
501 my ($blob, $type, $bytes) = $self->git->check($expect_oid);
502 $blob eq $expect_oid or die "BUG: $expect_oid not found after replace";
504 # don't leak FDs to Xapian:
507 # reindex modified messages:
508 for my $smsg (@$need_reindex) {
509 my $new_smsg = bless {
513 }, 'PublicInbox::Smsg';
514 my $sync = { autime => $smsg->{ds}, cotime => $smsg->{ts} };
515 $new_smsg->populate($new_mime, $sync);
516 $new_smsg->set_bytes($raw, $bytes);
517 do_idx($self, $new_mime, $new_smsg);
519 $rewritten->{rewrites};
522 sub last_epoch_commit ($$;$) {
523 my ($self, $i, $cmt) = @_;
524 my $v = PublicInbox::Search::SCHEMA_VERSION();
525 $self->{mm}->last_commit_xap($v, $i, $cmt);
528 sub set_last_commits ($) { # this is NOT for ExtSearchIdx
530 defined(my $epoch_max = $self->{epoch_max}) or return;
531 my $last_commit = $self->{last_commit};
532 foreach my $i (0..$epoch_max) {
533 defined(my $cmt = $last_commit->[$i]) or next;
534 $last_commit->[$i] = undef;
535 last_epoch_commit($self, $i, $cmt);
540 sub checkpoint ($;$) {
541 my ($self, $wait) = @_;
543 if (my $im = $self->{im}) {
550 my $shards = $self->{idx_shards};
552 my $mm = $self->{mm};
553 my $dbh = $mm->{dbh} if $mm;
555 # SQLite msgmap data is second in importance
556 $dbh->commit if $dbh;
558 # SQLite overview is third
559 $self->{oidx}->commit_lazy;
561 # Now deal with Xapian
563 # start commit_txn_lazy asynchronously on all parallel shards
564 # (non-parallel waits here)
565 $_->ipc_do('commit_txn_lazy') for @$shards;
567 # transactions started on parallel shards,
568 # wait for them by issuing an echo command (echo can only
569 # run after commit_txn_lazy is done)
570 if ($wait && $self->{parallel}) {
572 for my $shard (@$shards) {
573 my $echo = $shard->ipc_do('echo', $i);
574 $echo == $i or die <<"";
575 shard[$i] bad echo:$echo != $i waiting for txn commit
581 my $midx = $self->{midx}; # misc index
584 $PublicInbox::Search::X{CLOEXEC_UNSET} and
588 # last_commit is special, don't commit these until
589 # Xapian shards are done:
590 $dbh->begin_work if $dbh;
591 set_last_commits($self);
597 $self->{total_bytes} += $self->{transact_bytes};
598 $self->{transact_bytes} = 0;
601 # issue a write barrier to ensure all data is visible to other processes
602 # and read-only ops. Order of data importance is: git > SQLite > Xapian
604 sub barrier { checkpoint($_[0], 1) };
606 # true if locked and active
607 sub active { !!$_[0]->{im} }
613 if (my $im = delete $self->{im}) {
614 eval { $im->done }; # PublicInbox::Import::done
615 $err .= "import done: $@\n" if $@;
618 eval { checkpoint($self) };
619 $err .= "checkpoint: $@\n" if $@;
621 if (my $mm = delete $self->{mm}) {
622 my $m = $err ? 'rollback' : 'commit';
623 eval { $mm->{dbh}->$m };
624 $err .= "msgmap $m: $@\n" if $@;
626 my $shards = delete $self->{idx_shards};
629 eval { $_->shard_close };
630 $err .= "shard close: $@\n" if $@;
633 eval { $self->{oidx}->dbh_close };
634 $err .= "over close: $@\n" if $@;
635 delete $self->{midx};
636 my $nbytes = $self->{total_bytes};
637 $self->{total_bytes} = 0;
638 $self->lock_release(!!$nbytes) if $shards;
645 my $im = $self->{im};
647 if ($im->{bytes_added} < $self->{rotate_bytes}) {
654 my $dir = $self->{mg}->add_epoch(++$self->{epoch_max});
655 my $git = PublicInbox::Git->new($dir);
656 return $self->import_init($git, 0);
661 my $latest = $self->{ibx}->git_dir_latest(\$max);
662 if (defined $latest) {
663 my $git = PublicInbox::Git->new($latest);
664 my $packed_bytes = $git->packed_bytes;
665 my $unpacked_bytes = $packed_bytes / $PACKING_FACTOR;
667 if ($unpacked_bytes >= $self->{rotate_bytes}) {
670 $self->{epoch_max} = $max;
671 return $self->import_init($git, $packed_bytes);
674 $self->{epoch_max} = $epoch;
675 my $dir = $self->{mg}->add_epoch($epoch);
676 $self->import_init(PublicInbox::Git->new($dir), 0);
680 my ($self, $git, $packed_bytes, $tmp) = @_;
681 my $im = PublicInbox::Import->new($git, undef, undef, $self->{ibx});
682 $im->{bytes_added} = int($packed_bytes / $PACKING_FACTOR);
683 $im->{lock_path} = undef;
684 $im->{path_type} = 'v2';
685 $self->{im} = $im unless $tmp;
691 my ($mid, $cur, $new) = @_;
693 my $ah = File::Temp->new(TEMPLATE => 'email-cur-XXXX', TMPDIR => 1);
694 print $ah $cur->as_string or die "print: $!";
695 $ah->flush or die "flush: $!";
696 PublicInbox::Import::drop_unwanted_headers($new);
697 my $bh = File::Temp->new(TEMPLATE => 'email-new-XXXX', TMPDIR => 1);
698 print $bh $new->as_string or die "print: $!";
699 $bh->flush or die "flush: $!";
700 my $cmd = [ qw(diff -u), $ah->filename, $bh->filename ];
701 print STDERR "# MID conflict <$mid>\n";
702 my $pid = spawn($cmd, undef, { 1 => 2 });
703 waitpid($pid, 0) == $pid or die "diff did not finish";
707 my ($self, $smsg) = @_;
708 if (my $im = $self->{im}) {
709 my $msg = $im->cat_blob($smsg->{blob});
712 # older message, should be in alternates
713 $self->{ibx}->msg_by_smsg($smsg);
716 sub content_exists ($$$) {
717 my ($self, $mime, $mid) = @_;
718 my $oidx = $self->{oidx};
719 my $chashes = content_hashes($mime);
721 while (my $smsg = $oidx->next_by_mid($mid, \$id, \$prev)) {
722 my $msg = get_blob($self, $smsg);
723 if (!defined($msg)) {
724 warn "broken smsg for $mid\n";
727 my $cur = PublicInbox::Eml->new($msg);
728 return 1 if content_matches($chashes, $cur);
730 # XXX DEBUG_DIFF is experimental and may be removed
731 diff($mid, $cur, $mime) if $ENV{DEBUG_DIFF};
738 if (my $older_siblings = $self->{idx_shards}) {
739 $_->ipc_sibling_atfork_child for @$older_siblings;
741 if (my $im = $self->{im}) {
744 die "BUG: unexpected mm" if $self->{mm};
747 sub reindex_checkpoint ($$) {
748 my ($self, $sync) = @_;
750 $self->git->async_wait_all;
751 $self->update_last_commit($sync);
752 ${$sync->{need_checkpoint}} = 0;
753 my $mm_tmp = $sync->{mm_tmp};
754 $mm_tmp->atfork_prepare if $mm_tmp;
755 die 'BUG: {im} during reindex' if $self->{im};
756 if ($self->{ibx_map} && !$sync->{checkpoint_unlocks}) {
757 checkpoint($self, 1); # no need to release lock on pure index
759 $self->done; # release lock
762 if (my $pr = $sync->{-regen_fmt} ? $sync->{-opt}->{-progress} : undef) {
763 $pr->(sprintf($sync->{-regen_fmt}, ${$sync->{nr}}));
766 # allow -watch or -mda to write...
767 $self->idx_init($sync->{-opt}); # reacquire lock
768 if (my $intvl = $sync->{check_intvl}) { # eidx
769 $sync->{next_check} = PublicInbox::DS::now() + $intvl;
771 $mm_tmp->atfork_parent if $mm_tmp;
774 sub index_finalize ($$) {
775 my ($arg, $index) = @_;
776 ++$arg->{self}->{nidx};
777 if (defined(my $cur = $arg->{cur_cmt})) {
778 ${$arg->{latest_cmt}} = $cur;
780 die 'BUG: {cur_cmt} missing';
781 } # else { unindexing @leftovers doesn't set {cur_cmt}
784 sub index_oid { # cat_async callback
785 my ($bref, $oid, $type, $size, $arg) = @_;
786 is_bad_blob($oid, $type, $size, $arg->{oid}) and
787 return index_finalize($arg, 1); # size == 0 purged returns here
788 my $self = $arg->{self};
789 local $self->{current_info} = "$self->{current_info} $oid";
791 my $eml = PublicInbox::Eml->new($$bref);
792 my $mids = mids($eml);
793 my $chash = content_hash($eml);
795 if (scalar(@$mids) == 0) {
796 warn "E: $oid has no Message-ID, skipping\n";
800 # {unindexed} is unlikely
801 if (my $unindexed = $arg->{unindexed}) {
802 my $oidbin = pack('H*', $oid);
803 my $u = $unindexed->{$oidbin};
804 ($num, $mid0) = splice(@$u, 0, 2) if $u;
806 $self->{mm}->mid_set($num, $mid0);
807 if (scalar(@$u) == 0) { # done with current OID
808 delete $unindexed->{$oidbin};
809 delete($arg->{unindexed}) if !keys(%$unindexed);
813 if (!defined($num)) { # reuse if reindexing (or duplicates)
814 my $oidx = $self->{oidx};
815 for my $mid (@$mids) {
816 ($num, $mid0) = $oidx->num_mid0_for_oid($oid, $mid);
817 last if defined $num;
820 $mid0 //= do { # is this a number we got before?
821 $num = $arg->{mm_tmp}->num_for($mids->[0]);
822 defined($num) ? $mids->[0] : undef;
824 if (!defined($num)) {
825 for (my $i = $#$mids; $i >= 1; $i--) {
826 $num = $arg->{mm_tmp}->num_for($mids->[$i]);
834 $arg->{mm_tmp}->num_delete($num);
835 } else { # never seen
836 $num = $self->{mm}->mid_insert($mids->[0]);
839 } else { # rare, try the rest of them, backwards
840 for (my $i = $#$mids; $i >= 1; $i--) {
841 $num = $self->{mm}->mid_insert($mids->[$i]);
849 if (!defined($num)) {
850 warn "E: $oid <", join('> <', @$mids), "> is a duplicate\n";
858 }, 'PublicInbox::Smsg';
859 $smsg->populate($eml, $arg);
860 $smsg->set_bytes($$bref, $size);
861 if (do_idx($self, $eml, $smsg)) {
862 ${$arg->{need_checkpoint}} = 1;
864 index_finalize($arg, 1);
867 # only update last_commit for $i on reindex iff newer than current
868 sub update_last_commit {
869 my ($self, $sync, $stk) = @_;
870 my $unit = $sync->{unit} // return;
871 my $latest_cmt = $stk ? $stk->{latest_cmt} : ${$sync->{latest_cmt}};
872 defined($latest_cmt) or return;
873 my $last = last_epoch_commit($self, $unit->{epoch});
874 if (defined $last && is_ancestor($self->git, $last, $latest_cmt)) {
875 my @cmd = (qw(rev-list --count), "$last..$latest_cmt");
876 chomp(my $n = $unit->{git}->qx(@cmd));
877 return if $n ne '' && $n == 0;
879 last_epoch_commit($self, $unit->{epoch}, $latest_cmt);
883 my ($self, $sync) = @_;
885 for (my $i = $sync->{epoch_max}; $i >= 0; $i--) {
886 $heads->[$i] = last_epoch_commit($self, $i);
891 # returns a revision range for git-log(1)
892 sub log_range ($$$) {
893 my ($sync, $unit, $tip) = @_;
894 my $opt = $sync->{-opt};
895 my $pr = $opt->{-progress} if (($opt->{verbose} || 0) > 1);
896 my $i = $unit->{epoch};
897 my $cur = $sync->{ranges}->[$i] or do {
898 $pr->("$i.git indexing all of $tip\n") if $pr;
899 return $tip; # all of it
902 # fast equality check to avoid (v)fork+execve overhead
904 $sync->{ranges}->[$i] = undef;
908 my $range = "$cur..$tip";
909 $pr->("$i.git checking contiguity... ") if $pr;
910 my $git = $unit->{git};
911 if (is_ancestor($sync->{self}->git, $cur, $tip)) { # common case
912 $pr->("OK\n") if $pr;
913 my $n = $git->qx(qw(rev-list --count), $range);
916 $sync->{ranges}->[$i] = undef;
917 $pr->("$i.git has nothing new\n") if $pr;
918 return; # nothing to do
920 $pr->("$i.git has $n changes since $cur\n") if $pr;
922 $pr->("FAIL\n") if $pr;
924 discontiguous range: $range
925 Rewritten history? (in $git->{git_dir})
927 chomp(my $base = $git->qx('merge-base', $tip, $cur));
929 $range = "$base..$tip";
930 warn "found merge-base: $base\n"
933 warn "discarding history at $cur\n";
936 reindexing $git->{git_dir}
939 # $cur^0 may no longer exist if pruned by git
940 if ($git->qx(qw(rev-parse -q --verify), "$cur^0")) {
941 $unit->{unindex_range} = "$base..$cur";
942 } elsif ($base && $git->qx(qw(rev-parse -q --verify), $base)) {
943 $unit->{unindex_range} = "$base..";
945 warn "W: unable to unindex before $range\n";
951 # overridden by ExtSearchIdx
952 sub artnum_max { $_[0]->{mm}->num_highwater }
954 sub sync_prepare ($$) {
955 my ($self, $sync) = @_;
956 $sync->{ranges} = sync_ranges($self, $sync);
957 my $pr = $sync->{-opt}->{-progress};
959 my $head = $sync->{ibx}->{ref_head} || 'HEAD';
962 ($pfx) = ($sync->{ibx}->{inboxdir} =~ m!([^/]+)\z!g);
963 $pfx //= $sync->{ibx}->{inboxdir};
967 if ($self->{ibx_map}) {
968 # ExtSearchIdx won't index messages unless they're in
969 # over.sqlite3 for a given inbox, so don't read beyond
970 # what's in the per-inbox index.
972 my $v = PublicInbox::Search::SCHEMA_VERSION;
973 my $mm = $sync->{ibx}->mm;
974 for my $i (0..$sync->{epoch_max}) {
975 $reindex_heads->[$i] = $mm->last_commit_xap($v, $i);
977 } elsif ($sync->{reindex}) { # V2 inbox
978 # reindex stops at the current heads and we later
979 # rerun index_sync without {reindex}
980 $reindex_heads = $self->last_commits($sync);
982 if ($sync->{max_size} = $sync->{-opt}->{max_size}) {
983 $sync->{index_oid} = $self->can('index_oid');
985 my $git_pfx = "$sync->{ibx}->{inboxdir}/git";
986 for (my $i = $sync->{epoch_max}; $i >= 0; $i--) {
987 my $git_dir = "$git_pfx/$i.git";
988 -d $git_dir or next; # missing epochs are fine
989 my $git = PublicInbox::Git->new($git_dir);
990 my $unit = { git => $git, epoch => $i };
992 if ($reindex_heads) {
993 $tip = $head = $reindex_heads->[$i] or next;
995 $tip = $git->qx(qw(rev-parse -q --verify), $head);
996 next if $?; # new repo
999 my $range = log_range($sync, $unit, $tip) or next;
1000 # can't use 'rev-list --count' if we use --diff-filter
1001 $pr->("$pfx $i.git counting $range ... ") if $pr;
1002 # Don't bump num_highwater on --reindex by using {D}.
1003 # We intentionally do NOT use {D} in the non-reindex case
1004 # because we want NNTP article number gaps from unindexed
1005 # messages to show up in mirrors, too.
1006 $sync->{D} //= $sync->{reindex} ? {} : undef; # OID_BIN => NR
1007 my $stk = log2stack($sync, $git, $range);
1008 return 0 if $sync->{quit};
1009 my $nr = $stk ? $stk->num_records : 0;
1010 $pr->("$nr\n") if $pr;
1011 $unit->{stack} = $stk; # may be undef
1012 unshift @{$sync->{todo}}, $unit;
1015 return 0 if $sync->{quit};
1017 # XXX this should not happen unless somebody bypasses checks in
1018 # our code and blindly injects "d" file history into git repos
1019 if (my @leftovers = keys %{delete($sync->{D}) // {}}) {
1020 warn('W: unindexing '.scalar(@leftovers)." leftovers\n");
1021 local $self->{current_info} = 'leftover ';
1022 my $unindex_oid = $self->can('unindex_oid');
1023 for my $oid (@leftovers) {
1024 last if $sync->{quit};
1025 $oid = unpack('H*', $oid);
1026 my $req = { %$sync, oid => $oid };
1027 $self->git->cat_async($oid, $unindex_oid, $req);
1029 $self->git->cat_async_wait;
1031 return 0 if $sync->{quit};
1033 $sync->{-regen_fmt} = "%u/?\n";
1037 # reindex should NOT see new commits anymore, if we do,
1038 # it's a problem and we need to notice it via die()
1039 my $pad = length($regen_max) + 1;
1040 $sync->{-regen_fmt} = "% ${pad}u/$regen_max\n";
1041 $sync->{nr} = \(my $nr = 0);
1042 return -1 if $sync->{reindex};
1043 $regen_max + $self->artnum_max || 0;
1046 sub unindex_oid_aux ($$$) {
1047 my ($self, $oid, $mid) = @_;
1048 my @removed = $self->{oidx}->remove_oid($oid, $mid);
1049 return unless $self->{-need_xapian};
1050 for my $num (@removed) {
1051 idx_shard($self, $num)->ipc_do('xdb_remove', $num);
1055 sub unindex_oid ($$;$) { # git->cat_async callback
1056 my ($bref, $oid, $type, $size, $arg) = @_;
1057 is_bad_blob($oid, $type, $size, $arg->{oid}) and
1058 return index_finalize($arg, 0);
1059 my $self = $arg->{self};
1060 local $self->{current_info} = "$self->{current_info} $oid";
1061 my $unindexed = $arg->{in_unindex} ? $arg->{unindexed} : undef;
1062 my $mm = $self->{mm};
1063 my $mids = mids(PublicInbox::Eml->new($bref));
1065 my $oidx = $self->{oidx};
1066 foreach my $mid (@$mids) {
1069 while (my $smsg = $oidx->next_by_mid($mid, \$id, \$prev)) {
1070 $gone{$smsg->{num}} = 1 if $oid eq $smsg->{blob};
1072 my $n = scalar(keys(%gone)) or next;
1074 warn "BUG: multiple articles linked to $oid\n",
1075 join(',',sort keys %gone), "\n";
1077 # reuse (num => mid) mapping in ascending numeric order
1078 for my $num (sort { $a <=> $b } keys %gone) {
1081 my $mid0 = $mm->mid_for($num);
1082 my $oidbin = pack('H*', $oid);
1083 push @{$unindexed->{$oidbin}}, $num, $mid0;
1085 $mm->num_delete($num);
1087 unindex_oid_aux($self, $oid, $mid);
1089 index_finalize($arg, 0);
1092 sub git { $_[0]->{ibx}->git }
1094 # this is rare, it only happens when we get discontiguous history in
1095 # a mirror because the source used -purge or -edit
1096 sub unindex_todo ($$$) {
1097 my ($self, $sync, $unit) = @_;
1098 my $unindex_range = delete($unit->{unindex_range}) // return;
1099 my $unindexed = $sync->{unindexed} //= {}; # $oidbin => [$num, $mid0]
1100 my $before = scalar keys %$unindexed;
1101 # order does not matter, here:
1102 my $fh = $unit->{git}->popen(qw(log --raw -r --no-notes --no-color
1103 --no-abbrev --no-renames), $unindex_range);
1104 local $sync->{in_unindex} = 1;
1105 my $unindex_oid = $self->can('unindex_oid');
1107 /\A:\d{6} 100644 $OID ($OID) [AM]\tm$/o or next;
1108 $self->git->cat_async($1, $unindex_oid, { %$sync, oid => $1 });
1110 close $fh or die "git log failed: \$?=$?";
1111 $self->git->cat_async_wait;
1113 return unless $sync->{-opt}->{prune};
1114 my $after = scalar keys %$unindexed;
1115 return if $before == $after;
1117 # ensure any blob can not longer be accessed via dumb HTTP
1118 run_die(['git', "--git-dir=$unit->{git}->{git_dir}",
1119 qw(-c gc.reflogExpire=now gc --prune=all --quiet)]);
1122 sub sync_ranges ($$) {
1123 my ($self, $sync) = @_;
1124 my $reindex = $sync->{reindex};
1125 return $self->last_commits($sync) unless $reindex;
1126 return [] if ref($reindex) ne 'HASH';
1128 my $ranges = $reindex->{from}; # arrayref;
1129 if (ref($ranges) ne 'ARRAY') {
1130 die 'BUG: $reindex->{from} not an ARRAY';
1135 sub index_xap_only { # git->cat_async callback
1136 my ($bref, $oid, $type, $size, $smsg) = @_;
1137 my $self = delete $smsg->{self};
1138 my $idx = idx_shard($self, $smsg->{num});
1139 $idx->index_eml(PublicInbox::Eml->new($bref), $smsg);
1140 $self->{transact_bytes} += $smsg->{bytes};
1143 sub index_xap_step ($$$;$) {
1144 my ($self, $sync, $beg, $step) = @_;
1145 my $end = $sync->{art_end};
1146 return if $beg > $end; # nothing to do
1148 $step //= $self->{shards};
1149 my $ibx = $self->{ibx};
1150 if (my $pr = $sync->{-opt}->{-progress}) {
1151 $pr->("Xapian indexlevel=$ibx->{indexlevel} ".
1152 "$beg..$end (% $step)\n");
1154 for (my $num = $beg; $num <= $end; $num += $step) {
1155 last if $sync->{quit};
1156 my $smsg = $ibx->over->get_art($num) or next;
1157 $smsg->{self} = $self;
1158 $ibx->git->cat_async($smsg->{blob}, \&index_xap_only, $smsg);
1159 if ($self->{transact_bytes} >= $self->{batch_bytes}) {
1160 ${$sync->{nr}} = $num;
1161 reindex_checkpoint($self, $sync);
1166 sub index_todo ($$$) {
1167 my ($self, $sync, $unit) = @_;
1168 return if $sync->{quit};
1169 unindex_todo($self, $sync, $unit);
1170 my $stk = delete($unit->{stack}) or return;
1171 my $all = $self->git;
1172 my $index_oid = $self->can('index_oid');
1173 my $unindex_oid = $self->can('unindex_oid');
1175 if ($unit->{git}->{git_dir} =~ m!/([^/]+)/git/([0-9]+\.git)\z!) {
1176 $pfx = "$1 $2"; # v2
1178 ($pfx) = ($unit->{git}->{git_dir} =~ m!/([^/]+)\z!g);
1179 $pfx //= $unit->{git}->{git_dir};
1181 local $self->{current_info} = "$pfx ";
1182 local $sync->{latest_cmt} = \(my $latest_cmt);
1183 local $sync->{unit} = $unit;
1184 while (my ($f, $at, $ct, $oid, $cmt) = $stk->pop_rec) {
1185 if ($sync->{quit}) {
1186 warn "waiting to quit...\n";
1187 $all->async_wait_all;
1188 $self->update_last_commit($sync);
1199 if ($sync->{max_size}) {
1200 $all->check_async($oid, \&check_size, $req);
1202 $all->cat_async($oid, $index_oid, $req);
1204 } elsif ($f eq 'd') {
1205 $all->cat_async($oid, $unindex_oid, $req);
1207 if (${$sync->{need_checkpoint}}) {
1208 reindex_checkpoint($self, $sync);
1211 $all->async_wait_all;
1212 $self->update_last_commit($sync, $stk);
1216 my ($self, $opt, $sync, $art_beg) = @_;
1217 my $seq = $opt->{'sequential-shard'};
1219 local $self->{parallel} = 0 if $seq;
1220 $self->idx_init($opt); # acquire lock
1221 if (my $art_end = $self->{ibx}->mm->max) {
1223 need_checkpoint => \(my $bool = 0),
1226 nr => \(my $nr = 0),
1227 -regen_fmt => "%u/?\n",
1229 $sync->{art_end} = $art_end;
1230 if ($seq || !$self->{parallel}) {
1231 my $shard_end = $self->{shards} - 1;
1232 for my $i (0..$shard_end) {
1233 last if $sync->{quit};
1234 index_xap_step($self, $sync, $art_beg + $i);
1235 if ($i != $shard_end) {
1236 reindex_checkpoint($self, $sync);
1239 } else { # parallel (maybe)
1240 index_xap_step($self, $sync, $art_beg, 1);
1243 $self->git->cat_async_wait;
1244 $self->{ibx}->cleanup;
1248 # public, called by public-inbox-index
1250 my ($self, $opt) = @_;
1252 return xapian_only($self, $opt) if $opt->{xapian_only};
1255 my $latest = $self->{ibx}->git_dir_latest(\$epoch_max) // return;
1256 if ($opt->{'fast-noop'}) { # nanosecond (st_ctim) comparison
1257 use Time::HiRes qw(stat);
1258 if (my @mm = stat("$self->{ibx}->{inboxdir}/msgmap.sqlite3")) {
1259 my $c = $mm[10]; # 10 = ctime (nsec NV)
1260 my @hd = stat("$latest/refs/heads");
1261 my @pr = stat("$latest/packed-refs");
1262 return if $c > ($hd[10] // 0) && $c > ($pr[10] // 0);
1266 my $pr = $opt->{-progress};
1267 my $seq = $opt->{'sequential-shard'};
1268 my $art_beg; # the NNTP article number we start xapian_only at
1269 my $idxlevel = $self->{ibx}->{indexlevel};
1270 local $self->{ibx}->{indexlevel} = 'basic' if $seq;
1272 $self->idx_init($opt); # acquire lock
1273 $self->{mg}->fill_alternates;
1274 $self->{oidx}->rethread_prepare($opt);
1276 need_checkpoint => \(my $bool = 0),
1277 reindex => $opt->{reindex},
1280 ibx => $self->{ibx},
1281 epoch_max => $epoch_max,
1283 my $quit = PublicInbox::SearchIdx::quit_cb($sync);
1284 local $SIG{QUIT} = $quit;
1285 local $SIG{INT} = $quit;
1286 local $SIG{TERM} = $quit;
1288 if (sync_prepare($self, $sync)) {
1289 # tmp_clone seems to fail if inside a transaction, so
1290 # we rollback here (because we opened {mm} for reading)
1291 # Note: we do NOT rely on DBI transactions for atomicity;
1292 # only for batch performance.
1293 $self->{mm}->{dbh}->rollback;
1294 $self->{mm}->{dbh}->begin_work;
1296 $self->{mm}->tmp_clone($self->{ibx}->{inboxdir});
1298 # xapian_only works incrementally w/o --reindex
1299 if ($seq && !$opt->{reindex}) {
1300 $art_beg = $sync->{mm_tmp}->max || -1;
1304 # work forwards through history
1305 index_todo($self, $sync, $_) for @{delete($sync->{todo}) // []};
1306 $self->{oidx}->rethread_done($opt) unless $sync->{quit};
1309 if (my $nr = $sync->{nr}) {
1310 my $pr = $sync->{-opt}->{-progress};
1311 $pr->('all.git '.sprintf($sync->{-regen_fmt}, $$nr)) if $pr;
1315 # deal with Xapian shards sequentially
1316 if ($seq && delete($sync->{mm_tmp})) {
1317 if ($sync->{quit}) {
1320 $self->{ibx}->{indexlevel} = $idxlevel;
1321 xapian_only($self, $opt, $sync, $art_beg);
1322 $quit_warn = 1 if $sync->{quit};
1326 # --reindex on the command-line
1327 if (!$sync->{quit} && $opt->{reindex} &&
1328 !ref($opt->{reindex}) && $idxlevel ne 'basic') {
1329 $self->lock_acquire;
1330 my $s0 = PublicInbox::SearchIdx->new($self->{ibx}, 0, 0);
1331 if (my $xdb = $s0->idx_acquire) {
1332 my $n = $xdb->get_metadata('has_threadid');
1333 $xdb->set_metadata('has_threadid', '1') if $n ne '1';
1336 $self->lock_release;
1339 # reindex does not pick up new changes, so we rerun w/o it:
1340 if ($opt->{reindex} && !$sync->{quit}) {
1343 delete @again{qw(rethread reindex -skip_lock)};
1344 index_sync($self, \%again);
1345 $opt->{quit} = $again{quit}; # propagate to caller
1347 warn <<EOF if $quit_warn;
1348 W: interrupted, --xapian-only --reindex required upon restart
1352 sub ipc_atfork_child {
1354 if (my $lei = delete $self->{lei}) {
1355 $lei->_lei_atfork_child;
1356 my $pkt_op_p = delete $lei->{pkt_op_p};
1357 close($pkt_op_p->{op_p});
1359 $self->SUPER::ipc_atfork_child;