1 # Copyright (C) 2018-2020 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 base qw(PublicInbox::Lock);
11 use PublicInbox::SearchIdxShard;
14 use PublicInbox::Import;
15 use PublicInbox::MID qw(mids references);
16 use PublicInbox::ContentHash qw(content_hash content_digest);
17 use PublicInbox::Inbox;
18 use PublicInbox::OverIdx;
19 use PublicInbox::Msgmap;
20 use PublicInbox::Spawn qw(spawn popen_rd);
21 use PublicInbox::SearchIdx;
22 use PublicInbox::MultiMidQueue;
23 use IO::Handle; # ->autoflush
24 use File::Temp qw(tempfile);
26 # an estimate of the post-packed size to the raw uncompressed size
27 my $PACKING_FACTOR = 0.4;
29 # SATA storage lags behind what CPUs are capable of, so relying on
30 # nproc(1) can be misleading and having extra Xapian shards is a
31 # waste of FDs and space. It can also lead to excessive IO latency
32 # and slow things down. Users on NVME or other fast storage can
33 # use the NPROC env or switches in our script/public-inbox-* programs
34 # to increase Xapian shards
35 our $NPROC_MAX_DEFAULT = 4;
38 for my $nproc (qw(nproc gnproc)) { # GNU coreutils nproc
39 `$nproc 2>/dev/null` =~ /^(\d+)$/ and return $1;
42 # getconf(1) is POSIX, but *NPROCESSORS* vars are not
43 for (qw(_NPROCESSORS_ONLN NPROCESSORS_ONLN)) {
44 `getconf $_ 2>/dev/null` =~ /^(\d+)$/ and return $1;
47 # should we bother with `sysctl hw.ncpu`? Those only give
48 # us total processor count, not online processor count.
52 sub nproc_shards ($) {
54 my $n = $creat_opt->{nproc} if ref($creat_opt) eq 'HASH';
57 # assume 2 cores if not detectable or zero
58 state $NPROC_DETECTED = detect_nproc() || 2;
60 $n = $NPROC_MAX_DEFAULT if $n > $NPROC_MAX_DEFAULT;
63 # subtract for the main process and git-fast-import
68 sub count_shards ($) {
71 my $xpfx = $self->{xpfx};
73 # always load existing shards in case core count changes:
74 # Also, shard count may change while -watch is running
75 # due to "xcpdb --reshard"
78 foreach my $shard (<$xpfx/*>) {
79 -d $shard && $shard =~ m!/[0-9]+\z! or next;
80 $XapianDatabase //= do {
81 require PublicInbox::Search;
82 PublicInbox::Search::load_xapian();
83 $PublicInbox::Search::X{Database};
86 $XapianDatabase->new($shard)->close;
95 # $creat may be any true value, or 0/undef. A hashref is true,
96 # and $creat->{nproc} may be set to an integer
97 my ($class, $v2ibx, $creat) = @_;
98 $v2ibx = PublicInbox::InboxWritable->new($v2ibx);
99 my $dir = $v2ibx->assert_usable_dir;
103 File::Path::mkpath($dir);
105 die "$dir does not exist\n";
108 $v2ibx->umask_prepare;
110 my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION;
113 im => undef, # PublicInbox::Import
118 over => PublicInbox::OverIdx->new("$xpfx/over.sqlite3", 1),
119 lock_path => "$dir/inbox.lock",
120 # limit each git repo (epoch) to 1GB or so
121 rotate_bytes => int((1024 * 1024 * 1024) / $PACKING_FACTOR),
122 last_commit => [], # git repo -> commit
124 $self->{shards} = count_shards($self) || nproc_shards($creat);
125 $self->{index_max_size} = $v2ibx->{index_max_size};
131 my ($self, $shards, $skip_epoch) = @_;
132 if (defined $shards) {
133 $self->{parallel} = 0 if $shards == 0;
134 $self->{shards} = $shards if $shards > 0;
138 git_dir_latest($self, \$epoch_max);
139 if (defined $skip_epoch && $epoch_max == -1) {
140 $epoch_max = $skip_epoch;
142 $self->git_init($epoch_max >= 0 ? $epoch_max : 0);
146 # returns undef on duplicate or spam
147 # mimics Import::add and wraps it for v2
149 my ($self, $mime, $check_cb) = @_;
150 $self->{-inbox}->with_umask(sub {
151 _add($self, $mime, $check_cb)
155 # indexes a message, returns true if checkpointing is needed
157 my ($self, $msgref, $mime, $smsg) = @_;
158 $self->{over}->add_overview($mime, $smsg);
159 my $idx = idx_shard($self, $smsg->{num} % $self->{shards});
160 $idx->index_raw($msgref, $mime, $smsg);
161 my $n = $self->{transact_bytes} += $smsg->{bytes};
162 $n >= ($PublicInbox::SearchIdx::BATCH_BYTES * $self->{shards});
166 my ($self, $mime, $check_cb) = @_;
170 $mime = $check_cb->($mime) or return;
173 # All pipes (> $^F) known to Perl 5.6+ have FD_CLOEXEC set,
174 # as does SQLite 3.4.1+ (released in 2007-07-20), and
175 # Xapian 1.3.2+ (released 2015-03-15).
176 # For the most part, we can spawn git-fast-import without
177 # leaking FDs to it...
180 my ($num, $mid0) = v2_num_for($self, $mime);
181 defined $num or return; # duplicate
182 defined $mid0 or die "BUG: \$mid0 undefined\n";
183 my $im = $self->importer;
184 my $smsg = bless { mid => $mid0, num => $num }, 'PublicInbox::Smsg';
185 my $cmt = $im->add($mime, undef, $smsg); # sets $smsg->{ds|ts|blob}
186 $cmt = $im->get_mark($cmt);
187 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
189 my $msgref = delete $smsg->{-raw_email};
190 if (do_idx($self, $msgref, $mime, $smsg)) {
198 my ($self, $mime) = @_;
199 my $mids = mids($mime->header_obj);
201 my $mid = $mids->[0];
202 my $num = $self->{mm}->mid_insert($mid);
203 if (defined $num) { # common case
207 # crap, Message-ID is already known, hope somebody just resent:
208 foreach my $m (@$mids) {
209 # read-only lookup now safe to do after above barrier
210 # easy, don't store duplicates
211 # note: do not add more diagnostic info here since
212 # it gets noisy on public-inbox-watch restarts
213 return () if content_exists($self, $mime, $m);
216 # AltId may pre-populate article numbers (e.g. X-Mail-Count
217 # or NNTP article number), use that article number if it's
219 my $altid = $self->{-inbox}->{altid};
220 if ($altid && grep(/:file=msgmap\.sqlite3\z/, @$altid)) {
221 my $num = $self->{mm}->num_for($mid);
223 if (defined $num && !$self->{over}->get_art($num)) {
229 warn "<$mid> reused for mismatched content\n";
231 # try the rest of the mids
232 for(my $i = $#$mids; $i >= 1; $i--) {
234 $num = $self->{mm}->mid_insert($m);
236 warn "alternative <$m> for <$mid> found\n";
241 # none of the existing Message-IDs are good, generate a new one:
242 v2_num_for_harder($self, $mime);
245 sub v2_num_for_harder {
246 my ($self, $mime) = @_;
248 my $hdr = $mime->header_obj;
249 my $dig = content_digest($mime);
250 my $mid0 = PublicInbox::Import::digest2mid($dig, $hdr);
251 my $num = $self->{mm}->mid_insert($mid0);
252 unless (defined $num) {
253 # it's hard to spoof the last Received: header
254 my @recvd = $hdr->header_raw('Received');
255 $dig->add("Received: $_") foreach (@recvd);
256 $mid0 = PublicInbox::Import::digest2mid($dig, $hdr);
257 $num = $self->{mm}->mid_insert($mid0);
259 # fall back to a random Message-ID and give up determinism:
260 until (defined($num)) {
262 $mid0 = PublicInbox::Import::digest2mid($dig, $hdr);
263 warn "using random Message-ID <$mid0> as fallback\n";
264 $num = $self->{mm}->mid_insert($mid0);
267 PublicInbox::Import::append_mid($hdr, $mid0);
272 my ($self, $shard_i) = @_;
273 $self->{idx_shards}->[$shard_i];
278 my ($self, $opt) = @_;
279 return if $self->{idx_shards};
280 my $ibx = $self->{-inbox};
282 # do not leak read-only FDs to child processes, we only have these
283 # FDs for duplicate detection so they should not be
284 # frequently activated.
285 delete $ibx->{$_} foreach (qw(git mm search));
287 my $indexlevel = $ibx->{indexlevel};
288 if ($indexlevel && $indexlevel eq 'basic') {
289 $self->{parallel} = 0;
292 if ($self->{parallel}) {
293 pipe(my ($r, $w)) or die "pipe failed: $!";
294 # pipe for barrier notifications doesn't need to be big,
296 fcntl($w, 1031, 4096) if $^O eq 'linux';
297 $self->{bnote} = [ $r, $w ];
301 my $over = $self->{over};
303 $ibx->with_umask(sub {
304 $self->lock_acquire unless ($opt && $opt->{-skip_lock});
307 # xcpdb can change shard count while -watch is idle
308 my $nshards = count_shards($self);
309 if ($nshards && $nshards != $self->{shards}) {
310 $self->{shards} = $nshards;
313 # need to create all shards before initializing msgmap FD
314 my $max = $self->{shards} - 1;
316 # idx_shards must be visible to all forked processes
317 my $idx = $self->{idx_shards} = [];
318 for my $i (0..$max) {
319 push @$idx, PublicInbox::SearchIdxShard->new($self, $i);
322 # Now that all subprocesses are up, we can open the FDs
324 my $mm = $self->{mm} = PublicInbox::Msgmap->new_file(
325 "$self->{-inbox}->{inboxdir}/msgmap.sqlite3", 1);
326 $mm->{dbh}->begin_work;
330 # returns an array mapping [ epoch => latest_commit ]
331 # latest_commit may be undef if nothing was done to that epoch
332 # $replace_map = { $object_id => $strref, ... }
333 sub _replace_oids ($$$) {
334 my ($self, $mime, $replace_map) = @_;
336 my $pfx = "$self->{-inbox}->{inboxdir}/git";
337 my $rewrites = []; # epoch => commit
338 my $max = $self->{epoch_max};
340 unless (defined($max)) {
341 defined(my $latest = git_dir_latest($self, \$max)) or return;
342 $self->{epoch_max} = $max;
345 foreach my $i (0..$max) {
346 my $git_dir = "$pfx/$i.git";
348 my $git = PublicInbox::Git->new($git_dir);
349 my $im = $self->import_init($git, 0, 1);
350 $rewrites->[$i] = $im->replace_oids($mime, $replace_map);
356 sub content_hashes ($) {
358 my @chashes = ( content_hash($mime) );
360 # We still support Email::MIME, here, and
361 # Email::MIME->as_string doesn't always round-trip, so we may
362 # use a second content_hash
363 my $rt = content_hash(PublicInbox::Eml->new(\($mime->as_string)));
364 push @chashes, $rt if $chashes[0] ne $rt;
368 sub content_matches ($$) {
369 my ($chashes, $existing) = @_;
370 my $chash = content_hash($existing);
371 foreach (@$chashes) {
372 return 1 if $_ eq $chash
377 # used for removing or replacing (purging)
378 sub rewrite_internal ($$;$$$) {
379 my ($self, $old_mime, $cmt_msg, $new_mime, $sref) = @_;
381 my ($im, $need_reindex, $replace_map);
383 $replace_map = {}; # oid => sref
384 $need_reindex = [] if $new_mime;
386 $im = $self->importer;
388 my $over = $self->{over};
389 my $chashes = content_hashes($old_mime);
391 my $mids = mids($old_mime->header_obj);
393 # We avoid introducing new blobs into git since the raw content
394 # can be slightly different, so we do not need the user-supplied
395 # message now that we have the mids and content_hash
399 foreach my $mid (@$mids) {
400 my %gone; # num => [ smsg, $mime, raw ]
402 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
403 my $msg = get_blob($self, $smsg);
404 if (!defined($msg)) {
405 warn "broken smsg for $mid\n";
409 my $cur = PublicInbox::Eml->new($msg);
410 if (content_matches($chashes, $cur)) {
411 $gone{$smsg->{num}} = [ $smsg, $cur, \$orig ];
414 my $n = scalar keys %gone;
417 warn "BUG: multiple articles linked to <$mid>\n",
418 join(',', sort keys %gone), "\n";
420 foreach my $num (keys %gone) {
421 my ($smsg, $mime, $orig) = @{$gone{$num}};
422 # @removed should only be set once assuming
423 # no bugs in our deduplication code:
424 @removed = (undef, $mime, $smsg);
425 my $oid = $smsg->{blob};
427 $replace_map->{$oid} = $sref;
429 ($mark, undef) = $im->remove($orig, $cmt_msg);
433 if ($need_reindex) { # ->replace
434 push @$need_reindex, $smsg;
435 } else { # ->purge or ->remove
436 $self->{mm}->num_delete($num);
438 unindex_oid_remote($self, $oid, $mid);
443 my $cmt = $im->get_mark($mark);
444 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
446 if ($replace_map && scalar keys %$replace_map) {
447 my $rewrites = _replace_oids($self, $new_mime, $replace_map);
448 return { rewrites => $rewrites, need_reindex => $need_reindex };
450 defined($mark) ? @removed : undef;
453 # public (see PublicInbox::Import->remove), but note the 3rd element
454 # (retval[2]) is not part of the stable API shared with Import->remove
456 my ($self, $mime, $cmt_msg) = @_;
458 $self->{-inbox}->with_umask(sub {
459 @ret = rewrite_internal($self, $mime, $cmt_msg);
461 defined($ret[0]) ? @ret : undef;
464 sub _replace ($$;$$) {
465 my ($self, $old_mime, $new_mime, $sref) = @_;
466 my $rewritten = $self->{-inbox}->with_umask(sub {
467 rewrite_internal($self, $old_mime, undef, $new_mime, $sref);
470 my $rewrites = $rewritten->{rewrites};
471 # ->done is called if there are rewrites since we gc+prune from git
472 $self->idx_init if @$rewrites;
474 for my $i (0..$#$rewrites) {
475 defined(my $cmt = $rewrites->[$i]) or next;
476 $self->{last_commit}->[$i] = $cmt;
483 my ($self, $mime) = @_;
484 my $rewritten = _replace($self, $mime, undef, \'') or return;
485 $rewritten->{rewrites}
488 # returns the git object_id of $fh, does not write the object to FS
489 sub git_hash_raw ($$) {
490 my ($self, $raw) = @_;
491 # grab the expected OID we have to reindex:
492 open my $tmp_fh, '+>', undef or die "failed to open tmp: $!";
493 $tmp_fh->autoflush(1);
494 print $tmp_fh $$raw or die "print \$tmp_fh: $!";
495 sysseek($tmp_fh, 0, 0) or die "seek failed: $!";
497 my $git_dir = $self->{-inbox}->git->{git_dir};
498 my $cmd = ['git', "--git-dir=$git_dir", qw(hash-object --stdin)];
499 my $r = popen_rd($cmd, undef, { 0 => $tmp_fh });
501 chomp(my $oid = <$r>);
502 close $r or die "git hash-object failed: $?";
503 $oid =~ /\A[a-f0-9]{40}\z/ or die "OID not expected: $oid";
507 sub _check_mids_match ($$$) {
508 my ($old_list, $new_list, $hdrs) = @_;
509 my %old_mids = map { $_ => 1 } @$old_list;
510 my %new_mids = map { $_ => 1 } @$new_list;
511 my @old = keys %old_mids;
512 my @new = keys %new_mids;
513 my $err = "$hdrs may not be changed when replacing\n";
514 die $err if scalar(@old) != scalar(@new);
515 delete @new_mids{@old};
516 delete @old_mids{@new};
517 die $err if (scalar(keys %old_mids) || scalar(keys %new_mids));
520 # Changing Message-IDs or References with ->replace isn't supported.
521 # The rules for dealing with messages with multiple or conflicting
522 # Message-IDs are pretty complex and rethreading hasn't been fully
524 sub check_mids_match ($$) {
525 my ($old_mime, $new_mime) = @_;
526 my $old = $old_mime->header_obj;
527 my $new = $new_mime->header_obj;
528 _check_mids_match(mids($old), mids($new), 'Message-ID(s)');
529 _check_mids_match(references($old), references($new),
530 'References/In-Reply-To');
535 my ($self, $old_mime, $new_mime) = @_;
537 check_mids_match($old_mime, $new_mime);
539 # mutt will always add Content-Length:, Status:, Lines: when editing
540 PublicInbox::Import::drop_unwanted_headers($new_mime);
542 my $raw = $new_mime->as_string;
543 my $expect_oid = git_hash_raw($self, \$raw);
544 my $rewritten = _replace($self, $old_mime, $new_mime, \$raw) or return;
545 my $need_reindex = $rewritten->{need_reindex};
547 # just in case we have bugs in deduplication code:
548 my $n = scalar(@$need_reindex);
550 my $list = join(', ', map {
551 "$_->{num}: <$_->{mid}>"
554 W: rewritten $n messages matching content of original message (expected: 1).
555 W: possible bug in public-inbox, NNTP article IDs and Message-IDs follow:
560 # make sure we really got the OID:
561 my ($blob, $type, $bytes) = $self->{-inbox}->git->check($expect_oid);
562 $blob eq $expect_oid or die "BUG: $expect_oid not found after replace";
564 # don't leak FDs to Xapian:
565 $self->{-inbox}->git->cleanup;
567 # reindex modified messages:
568 for my $smsg (@$need_reindex) {
569 my $new_smsg = bless {
574 }, 'PublicInbox::Smsg';
575 my $v2w = { autime => $smsg->{ds}, cotime => $smsg->{ts} };
576 $new_smsg->populate($new_mime, $v2w);
577 do_idx($self, \$raw, $new_mime, $new_smsg);
579 $rewritten->{rewrites};
582 sub last_epoch_commit ($$;$) {
583 my ($self, $i, $cmt) = @_;
584 my $v = PublicInbox::Search::SCHEMA_VERSION();
585 $self->{mm}->last_commit_xap($v, $i, $cmt);
588 sub set_last_commits ($) {
590 defined(my $epoch_max = $self->{epoch_max}) or return;
591 my $last_commit = $self->{last_commit};
592 foreach my $i (0..$epoch_max) {
593 defined(my $cmt = $last_commit->[$i]) or next;
594 $last_commit->[$i] = undef;
595 last_epoch_commit($self, $i, $cmt);
601 $self->{bnote} or return;
603 my $barrier = { map { $_ => 1 } (0..$n) };
607 my ($self, $barrier) = @_;
608 my $bnote = $self->{bnote} or return;
610 while (scalar keys %$barrier) {
611 defined(my $l = readline($r)) or die "EOF on barrier_wait: $!";
612 $l =~ /\Abarrier (\d+)/ or die "bad line on barrier_wait: $l";
613 delete $barrier->{$1} or die "bad shard[$1] on barrier wait";
618 sub checkpoint ($;$) {
619 my ($self, $wait) = @_;
621 if (my $im = $self->{im}) {
628 my $shards = $self->{idx_shards};
630 my $dbh = $self->{mm}->{dbh};
632 # SQLite msgmap data is second in importance
635 # SQLite overview is third
636 $self->{over}->commit_lazy;
638 # Now deal with Xapian
640 my $barrier = $self->barrier_init(scalar @$shards);
642 # each shard needs to issue a barrier command
643 $_->remote_barrier for @$shards;
645 # wait for each Xapian shard
646 $self->barrier_wait($barrier);
648 $_->remote_commit for @$shards;
651 # last_commit is special, don't commit these until
652 # remote shards are done:
654 set_last_commits($self);
659 $self->{transact_bytes} = 0;
662 # issue a write barrier to ensure all data is visible to other processes
663 # and read-only ops. Order of data importance is: git > SQLite > Xapian
665 sub barrier { checkpoint($_[0], 1) };
670 my $im = delete $self->{im};
671 $im->done if $im; # PublicInbox::Import::done
673 my $mm = delete $self->{mm};
674 $mm->{dbh}->commit if $mm;
675 my $shards = delete $self->{idx_shards};
677 $_->remote_close for @$shards;
679 $self->{over}->disconnect;
680 delete $self->{bnote};
681 $self->{transact_bytes} = 0;
682 $self->lock_release if $shards;
683 $self->{-inbox}->git->cleanup;
686 sub fill_alternates ($$) {
687 my ($self, $epoch) = @_;
689 my $pfx = "$self->{-inbox}->{inboxdir}/git";
690 my $all = "$self->{-inbox}->{inboxdir}/all.git";
693 PublicInbox::Import::init_bare($all);
695 my $info_dir = "$all/objects/info";
696 my $alt = "$info_dir/alternates";
700 open(my $fh, '<', $alt) or die "open < $alt: $!\n";
701 $mode = (stat($fh))[2] & 07777;
703 # we assign a sort score to every alternate and favor
704 # the newest (highest numbered) one when we
706 my $other = 0; # in case admin adds non-epoch repos
708 if (m!\A\Q../../\E([0-9]+)\.git/objects\z!) {
714 } split(/\n+/, do { local $/; <$fh> });
717 foreach my $i (0..$epoch) {
718 my $dir = "../../git/$i.git/objects";
719 if (!exists($alt{$dir}) && -d "$pfx/$i.git") {
726 my ($fh, $tmp) = tempfile('alt-XXXXXXXX', DIR => $info_dir);
727 print $fh join("\n", sort { $alt{$b} <=> $alt{$a} } keys %alt), "\n"
728 or die "print $tmp: $!\n";
729 chmod($mode, $fh) or die "fchmod $tmp: $!\n";
730 close $fh or die "close $tmp $!\n";
731 rename($tmp, $alt) or die "rename $tmp => $alt: $!\n";
735 my ($self, $epoch) = @_;
736 my $git_dir = "$self->{-inbox}->{inboxdir}/git/$epoch.git";
737 PublicInbox::Import::init_bare($git_dir);
738 my @cmd = (qw/git config/, "--file=$git_dir/config",
739 'include.path', '../../all.git/config');
740 PublicInbox::Import::run_die(\@cmd);
741 fill_alternates($self, $epoch);
746 my ($self, $max) = @_;
748 my $pfx = "$self->{-inbox}->{inboxdir}/git";
749 return unless -d $pfx;
751 opendir my $dh, $pfx or die "opendir $pfx: $!\n";
752 while (defined(my $git_dir = readdir($dh))) {
753 $git_dir =~ m!\A([0-9]+)\.git\z! or next;
756 $latest = "$pfx/$git_dir";
764 my $im = $self->{im};
766 if ($im->{bytes_added} < $self->{rotate_bytes}) {
773 my $git_dir = $self->git_init(++$self->{epoch_max});
774 my $git = PublicInbox::Git->new($git_dir);
775 return $self->import_init($git, 0);
780 my $latest = git_dir_latest($self, \$max);
781 if (defined $latest) {
782 my $git = PublicInbox::Git->new($latest);
783 my $packed_bytes = $git->packed_bytes;
784 my $unpacked_bytes = $packed_bytes / $PACKING_FACTOR;
786 if ($unpacked_bytes >= $self->{rotate_bytes}) {
789 $self->{epoch_max} = $max;
790 return $self->import_init($git, $packed_bytes);
793 $self->{epoch_max} = $epoch;
794 $latest = $self->git_init($epoch);
795 $self->import_init(PublicInbox::Git->new($latest), 0);
799 my ($self, $git, $packed_bytes, $tmp) = @_;
800 my $im = PublicInbox::Import->new($git, undef, undef, $self->{-inbox});
801 $im->{bytes_added} = int($packed_bytes / $PACKING_FACTOR);
802 $im->{lock_path} = undef;
803 $im->{path_type} = 'v2';
804 $self->{im} = $im unless $tmp;
810 my ($mid, $cur, $new) = @_;
812 my ($ah, $an) = tempfile('email-cur-XXXXXXXX', TMPDIR => 1);
813 print $ah $cur->as_string or die "print: $!";
814 close $ah or die "close: $!";
815 my ($bh, $bn) = tempfile('email-new-XXXXXXXX', TMPDIR => 1);
816 PublicInbox::Import::drop_unwanted_headers($new);
817 print $bh $new->as_string or die "print: $!";
818 close $bh or die "close: $!";
819 my $cmd = [ qw(diff -u), $an, $bn ];
820 print STDERR "# MID conflict <$mid>\n";
821 my $pid = spawn($cmd, undef, { 1 => 2 });
822 waitpid($pid, 0) == $pid or die "diff did not finish";
827 my ($self, $smsg) = @_;
828 if (my $im = $self->{im}) {
829 my $msg = $im->cat_blob($smsg->{blob});
832 # older message, should be in alternates
833 my $ibx = $self->{-inbox};
834 $ibx->msg_by_smsg($smsg);
837 sub content_exists ($$$) {
838 my ($self, $mime, $mid) = @_;
839 my $over = $self->{over};
840 my $chashes = content_hashes($mime);
842 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
843 my $msg = get_blob($self, $smsg);
844 if (!defined($msg)) {
845 warn "broken smsg for $mid\n";
848 my $cur = PublicInbox::Eml->new($msg);
849 return 1 if content_matches($chashes, $cur);
851 # XXX DEBUG_DIFF is experimental and may be removed
852 diff($mid, $cur, $mime) if $ENV{DEBUG_DIFF};
859 my $fh = delete $self->{reindex_pipe};
861 if (my $shards = $self->{idx_shards}) {
862 $_->atfork_child foreach @$shards;
864 if (my $im = $self->{im}) {
867 die "unexpected mm" if $self->{mm};
868 close $self->{bnote}->[0] or die "close bnote[0]: $!\n";
872 sub mark_deleted ($$$$) {
873 my ($self, $sync, $git, $oid) = @_;
874 return if PublicInbox::SearchIdx::too_big($self, $git, $oid);
875 my $msgref = $git->cat_file($oid);
876 my $mime = PublicInbox::Eml->new($$msgref);
877 my $mids = mids($mime->header_obj);
878 my $chash = content_hash($mime);
879 foreach my $mid (@$mids) {
880 $sync->{D}->{"$mid\0$chash"} = $oid;
884 sub reindex_checkpoint ($$$) {
885 my ($self, $sync, $git) = @_;
888 $sync->{mm_tmp}->atfork_prepare;
889 $self->done; # release lock
891 if (my $pr = $sync->{-opt}->{-progress}) {
892 my ($bn) = (split('/', $git->{git_dir}))[-1];
893 $pr->("$bn ".sprintf($sync->{-regen_fmt}, $sync->{nr}));
896 # allow -watch or -mda to write...
897 $self->idx_init; # reacquire lock
898 $sync->{mm_tmp}->atfork_parent;
901 # only for a few odd messages with multiple Message-IDs
902 sub reindex_oid_m ($$$$;$) {
903 my ($self, $sync, $git, $oid, $regen_num) = @_;
904 $self->{current_info} = "multi_mid $oid";
905 my ($num, $mid0, $len);
906 my $msgref = $git->cat_file($oid, \$len);
907 my $mime = PublicInbox::Eml->new($$msgref);
908 my $mids = mids($mime->header_obj);
909 my $chash = content_hash($mime);
910 die "BUG: reindex_oid_m called for <=1 mids" if scalar(@$mids) <= 1;
912 for my $mid (reverse @$mids) {
913 delete($sync->{D}->{"$mid\0$chash"}) and
914 die "BUG: reindex_oid should handle <$mid> delete";
916 my $over = $self->{over};
917 for my $mid (reverse @$mids) {
918 ($num, $mid0) = $over->num_mid0_for_oid($oid, $mid);
919 next unless defined $num;
920 if (defined($regen_num) && $regen_num != $num) {
921 die "BUG: regen(#$regen_num) != over(#$num)";
924 unless (defined($num)) {
925 for my $mid (reverse @$mids) {
926 # is this a number we got before?
927 my $n = $sync->{mm_tmp}->num_for($mid);
928 next unless defined $n;
929 next if defined($regen_num) && $regen_num != $n;
930 ($num, $mid0) = ($n, $mid);
935 $sync->{mm_tmp}->num_delete($num);
936 } elsif (defined $regen_num) {
938 for my $mid (reverse @$mids) {
939 $self->{mm}->mid_set($num, $mid) == 1 or next;
943 unless (defined $mid0) {
944 warn "E: cannot regen #$num\n";
947 } else { # fixup bugs in old mirrors on reindex
948 for my $mid (reverse @$mids) {
949 $num = $self->{mm}->mid_insert($mid);
950 next unless defined $num;
955 if ($sync->{reindex}) {
956 warn "reindex added #$num <$mid0>\n";
959 warn "E: cannot find article #\n";
969 }, 'PublicInbox::Smsg';
970 $smsg->populate($mime, $self);
971 if (do_idx($self, $msgref, $mime, $smsg)) {
972 reindex_checkpoint($self, $sync, $git);
976 sub check_unindexed ($$$) {
977 my ($self, $num, $mid0) = @_;
978 my $unindexed = $self->{unindexed} // {};
979 my $n = delete($unindexed->{$mid0});
980 defined $n or return;
982 die "BUG: unindexed $n != $num <$mid0>\n";
984 $self->{mm}->mid_set($num, $mid0);
988 sub multi_mid_q_push ($$$) {
989 my ($self, $sync, $oid) = @_;
990 my $multi_mid = $sync->{multi_mid} //= PublicInbox::MultiMidQueue->new;
991 if ($sync->{reindex}) { # no regen on reindex
992 $multi_mid->push_oid($oid, $self);
994 my $num = $sync->{regen}--;
995 die "BUG: ran out of article numbers" if $num <= 0;
996 $multi_mid->set_oid($num, $oid, $self);
1000 sub reindex_oid ($$$$) {
1001 my ($self, $sync, $git, $oid) = @_;
1002 return if PublicInbox::SearchIdx::too_big($self, $git, $oid);
1003 my ($num, $mid0, $len);
1004 my $msgref = $git->cat_file($oid, \$len);
1005 return if $len == 0; # purged
1006 my $mime = PublicInbox::Eml->new($$msgref);
1007 my $mids = mids($mime->header_obj);
1008 my $chash = content_hash($mime);
1010 if (scalar(@$mids) == 0) {
1011 warn "E: $oid has no Message-ID, skipping\n";
1013 } elsif (scalar(@$mids) == 1) {
1014 my $mid = $mids->[0];
1016 # was the file previously marked as deleted?, skip if so
1017 if (delete($sync->{D}->{"$mid\0$chash"})) {
1018 if (!$sync->{reindex}) {
1019 $num = $sync->{regen}--;
1020 $self->{mm}->num_highwater($num);
1025 # is this a number we got before?
1026 $num = $sync->{mm_tmp}->num_for($mid);
1029 check_unindexed($self, $num, $mid0);
1031 $num = $sync->{regen}--;
1032 die "BUG: ran out of article numbers" if $num <= 0;
1033 if ($self->{mm}->mid_set($num, $mid) != 1) {
1034 warn "E: unable to assign $num => <$mid>\n";
1039 } else { # multiple MIDs are a weird case:
1042 $del += delete($sync->{D}->{"$_\0$chash"}) // 0;
1045 unindex_oid_remote($self, $oid, $_) for @$mids;
1046 # do not delete from {mm_tmp}, since another
1047 # single-MID message may use it.
1048 } else { # handle them at the end:
1049 multi_mid_q_push($self, $sync, $oid);
1053 $sync->{mm_tmp}->mid_delete($mid0) or
1054 die "failed to delete <$mid0> for article #$num\n";
1061 }, 'PublicInbox::Smsg';
1062 $smsg->populate($mime, $self);
1063 if (do_idx($self, $msgref, $mime, $smsg)) {
1064 reindex_checkpoint($self, $sync, $git);
1068 # only update last_commit for $i on reindex iff newer than current
1069 sub update_last_commit ($$$$) {
1070 my ($self, $git, $i, $cmt) = @_;
1071 my $last = last_epoch_commit($self, $i);
1072 if (defined $last && is_ancestor($git, $last, $cmt)) {
1073 my @cmd = (qw(rev-list --count), "$last..$cmt");
1074 chomp(my $n = $git->qx(@cmd));
1075 return if $n ne '' && $n == 0;
1077 last_epoch_commit($self, $i, $cmt);
1080 sub git_dir_n ($$) { "$_[0]->{-inbox}->{inboxdir}/git/$_[1].git" }
1082 sub last_commits ($$) {
1083 my ($self, $epoch_max) = @_;
1085 for (my $i = $epoch_max; $i >= 0; $i--) {
1086 $heads->[$i] = last_epoch_commit($self, $i);
1091 *is_ancestor = *PublicInbox::SearchIdx::is_ancestor;
1093 # returns a revision range for git-log(1)
1094 sub log_range ($$$$$) {
1095 my ($self, $sync, $git, $i, $tip) = @_;
1096 my $opt = $sync->{-opt};
1097 my $pr = $opt->{-progress} if (($opt->{verbose} || 0) > 1);
1098 my $cur = $sync->{ranges}->[$i] or do {
1099 $pr->("$i.git indexing all of $tip") if $pr;
1100 return $tip; # all of it
1103 # fast equality check to avoid (v)fork+execve overhead
1105 $sync->{ranges}->[$i] = undef;
1109 my $range = "$cur..$tip";
1110 $pr->("$i.git checking contiguity... ") if $pr;
1111 if (is_ancestor($git, $cur, $tip)) { # common case
1112 $pr->("OK\n") if $pr;
1113 my $n = $git->qx(qw(rev-list --count), $range);
1116 $sync->{ranges}->[$i] = undef;
1117 $pr->("$i.git has nothing new\n") if $pr;
1118 return; # nothing to do
1120 $pr->("$i.git has $n changes since $cur\n") if $pr;
1122 $pr->("FAIL\n") if $pr;
1124 discontiguous range: $range
1125 Rewritten history? (in $git->{git_dir})
1127 chomp(my $base = $git->qx('merge-base', $tip, $cur));
1129 $range = "$base..$tip";
1130 warn "found merge-base: $base\n"
1133 warn "discarding history at $cur\n";
1136 reindexing $git->{git_dir} starting at
1139 $sync->{unindex_range}->{$i} = "$base..$cur";
1144 sub sync_prepare ($$$) {
1145 my ($self, $sync, $epoch_max) = @_;
1146 my $pr = $sync->{-opt}->{-progress};
1148 my $head = $self->{-inbox}->{ref_head} || 'refs/heads/master';
1150 # reindex stops at the current heads and we later rerun index_sync
1152 my $reindex_heads = last_commits($self, $epoch_max) if $sync->{reindex};
1154 for (my $i = $epoch_max; $i >= 0; $i--) {
1155 die 'BUG: already indexing!' if $self->{reindex_pipe};
1156 my $git_dir = git_dir_n($self, $i);
1157 -d $git_dir or next; # missing epochs are fine
1158 my $git = PublicInbox::Git->new($git_dir);
1159 if ($reindex_heads) {
1160 $head = $reindex_heads->[$i] or next;
1162 chomp(my $tip = $git->qx(qw(rev-parse -q --verify), $head));
1164 next if $?; # new repo
1165 my $range = log_range($self, $sync, $git, $i, $tip) or next;
1166 $sync->{ranges}->[$i] = $range;
1168 # can't use 'rev-list --count' if we use --diff-filter
1169 $pr->("$i.git counting $range ... ") if $pr;
1171 my $fh = $git->popen(qw(log --pretty=tformat:%H
1172 --no-notes --no-color --no-renames
1173 --diff-filter=AM), $range, '--', 'm');
1175 close $fh or die "git log failed: \$?=$?";
1176 $pr->("$n\n") if $pr;
1180 return 0 if (!$regen_max && !keys(%{$self->{unindex_range}}));
1182 # reindex should NOT see new commits anymore, if we do,
1183 # it's a problem and we need to notice it via die()
1184 my $pad = length($regen_max) + 1;
1185 $sync->{-regen_fmt} = "% ${pad}u/$regen_max\n";
1187 return -1 if $sync->{reindex};
1188 $regen_max + $self->{mm}->num_highwater() || 0;
1191 sub unindex_oid_remote ($$$) {
1192 my ($self, $oid, $mid) = @_;
1193 $_->remote_remove($oid, $mid) foreach @{$self->{idx_shards}};
1194 $self->{over}->remove_oid($oid, $mid);
1197 sub unindex_oid ($$$;$) {
1198 my ($self, $git, $oid, $unindexed) = @_;
1199 my $mm = $self->{mm};
1200 my $msgref = $git->cat_file($oid);
1201 my $mime = PublicInbox::Eml->new($msgref);
1202 my $mids = mids($mime->header_obj);
1203 $mime = $msgref = undef;
1204 my $over = $self->{over};
1205 foreach my $mid (@$mids) {
1208 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
1209 $gone{$smsg->{num}} = 1 if $oid eq $smsg->{blob};
1212 my $n = scalar keys %gone;
1215 warn "BUG: multiple articles linked to $oid\n",
1216 join(',',sort keys %gone), "\n";
1218 foreach my $num (keys %gone) {
1220 my $mid0 = $mm->mid_for($num);
1221 $unindexed->{$mid0} = $num;
1223 $mm->num_delete($num);
1225 unindex_oid_remote($self, $oid, $mid);
1229 my $x40 = qr/[a-f0-9]{40}/;
1230 sub unindex ($$$$) {
1231 my ($self, $sync, $git, $unindex_range) = @_;
1232 my $unindexed = $self->{unindexed} ||= {}; # $mid0 => $num
1233 my $before = scalar keys %$unindexed;
1234 # order does not matter, here:
1235 my @cmd = qw(log --raw -r
1236 --no-notes --no-color --no-abbrev --no-renames);
1237 my $fh = $self->{reindex_pipe} = $git->popen(@cmd, $unindex_range);
1239 /\A:\d{6} 100644 $x40 ($x40) [AM]\tm$/o or next;
1240 unindex_oid($self, $git, $1, $unindexed);
1242 delete $self->{reindex_pipe};
1243 close $fh or die "git log failed: \$?=$?";
1245 return unless $sync->{-opt}->{prune};
1246 my $after = scalar keys %$unindexed;
1247 return if $before == $after;
1249 # ensure any blob can not longer be accessed via dumb HTTP
1250 PublicInbox::Import::run_die(['git', "--git-dir=$git->{git_dir}",
1251 qw(-c gc.reflogExpire=now gc --prune=all --quiet)]);
1254 sub sync_ranges ($$$) {
1255 my ($self, $sync, $epoch_max) = @_;
1256 my $reindex = $sync->{reindex};
1258 return last_commits($self, $epoch_max) unless $reindex;
1259 return [] if ref($reindex) ne 'HASH';
1261 my $ranges = $reindex->{from}; # arrayref;
1262 if (ref($ranges) ne 'ARRAY') {
1263 die 'BUG: $reindex->{from} not an ARRAY';
1268 sub index_epoch ($$$) {
1269 my ($self, $sync, $i) = @_;
1271 my $git_dir = git_dir_n($self, $i);
1272 die 'BUG: already reindexing!' if $self->{reindex_pipe};
1273 -d $git_dir or return; # missing epochs are fine
1274 fill_alternates($self, $i);
1275 my $git = PublicInbox::Git->new($git_dir);
1276 if (my $unindex_range = delete $sync->{unindex_range}->{$i}) {
1277 unindex($self, $sync, $git, $unindex_range);
1279 defined(my $range = $sync->{ranges}->[$i]) or return;
1280 if (my $pr = $sync->{-opt}->{-progress}) {
1281 $pr->("$i.git indexing $range\n");
1284 my @cmd = qw(log --raw -r --pretty=tformat:%H.%at.%ct
1285 --no-notes --no-color --no-abbrev --no-renames);
1286 my $fh = $self->{reindex_pipe} = $git->popen(@cmd, $range);
1290 $self->{current_info} = "$i.git $_";
1291 if (/\A($x40)\.([0-9]+)\.([0-9]+)$/o) {
1293 $self->{autime} = $2;
1294 $self->{cotime} = $3;
1295 } elsif (/\A:\d{6} 100644 $x40 ($x40) [AM]\tm$/o) {
1296 reindex_oid($self, $sync, $git, $1);
1297 } elsif (/\A:\d{6} 100644 $x40 ($x40) [AM]\td$/o) {
1298 mark_deleted($self, $sync, $git, $1);
1301 close $fh or die "git log failed: \$?=$?";
1302 delete @$self{qw(reindex_pipe autime cotime)};
1303 update_last_commit($self, $git, $i, $cmt) if defined $cmt;
1306 # public, called by public-inbox-index
1308 my ($self, $opt) = @_;
1310 my $pr = $opt->{-progress};
1312 my $latest = git_dir_latest($self, \$epoch_max);
1313 return unless defined $latest;
1314 $self->idx_init($opt); # acquire lock
1316 D => {}, # "$mid\0$chash" => $oid
1317 unindex_range => {}, # EPOCH => oid_old..oid_new
1318 reindex => $opt->{reindex},
1321 $sync->{ranges} = sync_ranges($self, $sync, $epoch_max);
1322 $sync->{regen} = sync_prepare($self, $sync, $epoch_max);
1324 if ($sync->{regen}) {
1325 # tmp_clone seems to fail if inside a transaction, so
1326 # we rollback here (because we opened {mm} for reading)
1327 # Note: we do NOT rely on DBI transactions for atomicity;
1328 # only for batch performance.
1329 $self->{mm}->{dbh}->rollback;
1330 $self->{mm}->{dbh}->begin_work;
1331 $sync->{mm_tmp} = $self->{mm}->tmp_clone;
1334 # work backwards through history
1335 for (my $i = $epoch_max; $i >= 0; $i--) {
1336 index_epoch($self, $sync, $i);
1339 # unindex is required for leftovers if "deletes" affect messages
1340 # in a previous fetch+index window:
1342 if (my @leftovers = values %{delete $sync->{D}}) {
1343 $git = $self->{-inbox}->git;
1344 for my $oid (@leftovers) {
1345 $self->{current_info} = "leftover $oid";
1346 unindex_oid($self, $git, $oid);
1349 if (my $multi_mid = delete $sync->{multi_mid}) {
1350 $git //= $self->{-inbox}->git;
1351 my $min = $multi_mid->{min};
1352 my $max = $multi_mid->{max};
1353 if ($sync->{reindex}) {
1354 # we may need to create new Message-IDs if mirrors
1355 # were initially indexed with old versions
1356 for (my $i = $max; $i >= $min; $i--) {
1358 $oid = $multi_mid->get_oid($i, $self) or next;
1359 next unless defined $oid;
1360 reindex_oid_m($self, $sync, $git, $oid);
1362 } else { # regen on initial index
1363 for my $num ($min..$max) {
1365 $oid = $multi_mid->get_oid($num, $self) or next;
1366 reindex_oid_m($self, $sync, $git, $oid, $num);
1370 $git->cleanup if $git;
1373 if (my $nr = $sync->{nr}) {
1374 my $pr = $sync->{-opt}->{-progress};
1375 $pr->('all.git '.sprintf($sync->{-regen_fmt}, $nr)) if $pr;
1378 # reindex does not pick up new changes, so we rerun w/o it:
1379 if ($opt->{reindex}) {
1382 delete @again{qw(reindex -skip_lock)};
1383 index_sync($self, \%again);