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 parent qw(PublicInbox::Lock);
10 use PublicInbox::SearchIdxShard;
13 use PublicInbox::Import;
14 use PublicInbox::MID qw(mids references);
15 use PublicInbox::ContentHash qw(content_hash content_digest);
16 use PublicInbox::InboxWritable;
17 use PublicInbox::OverIdx;
18 use PublicInbox::Msgmap;
19 use PublicInbox::Spawn qw(spawn popen_rd);
20 use PublicInbox::SearchIdx;
21 use PublicInbox::MultiMidQueue;
22 use IO::Handle; # ->autoflush
23 use File::Temp qw(tempfile);
25 # an estimate of the post-packed size to the raw uncompressed size
26 my $PACKING_FACTOR = 0.4;
28 # SATA storage lags behind what CPUs are capable of, so relying on
29 # nproc(1) can be misleading and having extra Xapian shards is a
30 # waste of FDs and space. It can also lead to excessive IO latency
31 # and slow things down. Users on NVME or other fast storage can
32 # use the NPROC env or switches in our script/public-inbox-* programs
33 # to increase Xapian shards
34 our $NPROC_MAX_DEFAULT = 4;
37 for my $nproc (qw(nproc gnproc)) { # GNU coreutils nproc
38 `$nproc 2>/dev/null` =~ /^(\d+)$/ and return $1;
41 # getconf(1) is POSIX, but *NPROCESSORS* vars are not
42 for (qw(_NPROCESSORS_ONLN NPROCESSORS_ONLN)) {
43 `getconf $_ 2>/dev/null` =~ /^(\d+)$/ and return $1;
46 # should we bother with `sysctl hw.ncpu`? Those only give
47 # us total processor count, not online processor count.
51 sub nproc_shards ($) {
53 my $n = $creat_opt->{nproc} if ref($creat_opt) eq 'HASH';
56 # assume 2 cores if not detectable or zero
57 state $NPROC_DETECTED = detect_nproc() || 2;
59 $n = $NPROC_MAX_DEFAULT if $n > $NPROC_MAX_DEFAULT;
62 # subtract for the main process and git-fast-import
67 sub count_shards ($) {
70 my $xpfx = $self->{xpfx};
72 # always load existing shards in case core count changes:
73 # Also, shard count may change while -watch is running
74 # due to "xcpdb --reshard"
77 foreach my $shard (<$xpfx/*>) {
78 -d $shard && $shard =~ m!/[0-9]+\z! or next;
79 $XapianDatabase //= do {
80 require PublicInbox::Search;
81 PublicInbox::Search::load_xapian();
82 $PublicInbox::Search::X{Database};
85 $XapianDatabase->new($shard)->close;
94 # $creat may be any true value, or 0/undef. A hashref is true,
95 # and $creat->{nproc} may be set to an integer
96 my ($class, $v2ibx, $creat) = @_;
97 $v2ibx = PublicInbox::InboxWritable->new($v2ibx);
98 my $dir = $v2ibx->assert_usable_dir;
102 File::Path::mkpath($dir);
104 die "$dir does not exist\n";
107 $v2ibx->umask_prepare;
109 my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION;
112 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, $skip_artnum) = @_;
132 if (defined $shards) {
133 $self->{parallel} = 0 if $shards == 0;
134 $self->{shards} = $shards if $shards > 0;
137 $self->{mm}->skip_artnum($skip_artnum) if defined $skip_artnum;
139 git_dir_latest($self, \$epoch_max);
140 if (defined $skip_epoch && $epoch_max == -1) {
141 $epoch_max = $skip_epoch;
143 $self->git_init($epoch_max >= 0 ? $epoch_max : 0);
147 # returns undef on duplicate or spam
148 # mimics Import::add and wraps it for v2
150 my ($self, $mime, $check_cb) = @_;
151 $self->{-inbox}->with_umask(sub {
152 _add($self, $mime, $check_cb)
156 # indexes a message, returns true if checkpointing is needed
158 my ($self, $msgref, $mime, $smsg) = @_;
159 $smsg->{bytes} = $smsg->{raw_bytes} +
160 PublicInbox::SearchIdx::crlf_adjust($$msgref);
161 $self->{over}->add_overview($mime, $smsg);
162 my $idx = idx_shard($self, $smsg->{num} % $self->{shards});
163 $idx->index_raw($msgref, $mime, $smsg);
164 my $n = $self->{transact_bytes} += $smsg->{raw_bytes};
165 $n >= ($PublicInbox::SearchIdx::BATCH_BYTES * $self->{shards});
169 my ($self, $mime, $check_cb) = @_;
173 $mime = $check_cb->($mime, $self->{-inbox}) or return;
176 # All pipes (> $^F) known to Perl 5.6+ have FD_CLOEXEC set,
177 # as does SQLite 3.4.1+ (released in 2007-07-20), and
178 # Xapian 1.3.2+ (released 2015-03-15).
179 # For the most part, we can spawn git-fast-import without
180 # leaking FDs to it...
183 my ($num, $mid0) = v2_num_for($self, $mime);
184 defined $num or return; # duplicate
185 defined $mid0 or die "BUG: \$mid0 undefined\n";
186 my $im = $self->importer;
187 my $smsg = bless { mid => $mid0, num => $num }, 'PublicInbox::Smsg';
188 my $cmt = $im->add($mime, undef, $smsg); # sets $smsg->{ds|ts|blob}
189 $cmt = $im->get_mark($cmt);
190 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
192 my $msgref = delete $smsg->{-raw_email};
193 if (do_idx($self, $msgref, $mime, $smsg)) {
201 my ($self, $mime) = @_;
202 my $mids = mids($mime->header_obj);
204 my $mid = $mids->[0];
205 my $num = $self->{mm}->mid_insert($mid);
206 if (defined $num) { # common case
210 # crap, Message-ID is already known, hope somebody just resent:
211 foreach my $m (@$mids) {
212 # read-only lookup now safe to do after above barrier
213 # easy, don't store duplicates
214 # note: do not add more diagnostic info here since
215 # it gets noisy on public-inbox-watch restarts
216 return () if content_exists($self, $mime, $m);
219 # AltId may pre-populate article numbers (e.g. X-Mail-Count
220 # or NNTP article number), use that article number if it's
222 my $altid = $self->{-inbox}->{altid};
223 if ($altid && grep(/:file=msgmap\.sqlite3\z/, @$altid)) {
224 my $num = $self->{mm}->num_for($mid);
226 if (defined $num && !$self->{over}->get_art($num)) {
232 warn "<$mid> reused for mismatched content\n";
234 # try the rest of the mids
235 for(my $i = $#$mids; $i >= 1; $i--) {
237 $num = $self->{mm}->mid_insert($m);
239 warn "alternative <$m> for <$mid> found\n";
244 # none of the existing Message-IDs are good, generate a new one:
245 v2_num_for_harder($self, $mime);
248 sub v2_num_for_harder {
249 my ($self, $mime) = @_;
251 my $hdr = $mime->header_obj;
252 my $dig = content_digest($mime);
253 my $mid0 = PublicInbox::Import::digest2mid($dig, $hdr);
254 my $num = $self->{mm}->mid_insert($mid0);
255 unless (defined $num) {
256 # it's hard to spoof the last Received: header
257 my @recvd = $hdr->header_raw('Received');
258 $dig->add("Received: $_") foreach (@recvd);
259 $mid0 = PublicInbox::Import::digest2mid($dig, $hdr);
260 $num = $self->{mm}->mid_insert($mid0);
262 # fall back to a random Message-ID and give up determinism:
263 until (defined($num)) {
265 $mid0 = PublicInbox::Import::digest2mid($dig, $hdr);
266 warn "using random Message-ID <$mid0> as fallback\n";
267 $num = $self->{mm}->mid_insert($mid0);
270 PublicInbox::Import::append_mid($hdr, $mid0);
275 my ($self, $shard_i) = @_;
276 $self->{idx_shards}->[$shard_i];
281 my ($self, $opt) = @_;
282 return if $self->{idx_shards};
283 my $ibx = $self->{-inbox};
285 # do not leak read-only FDs to child processes, we only have these
286 # FDs for duplicate detection so they should not be
287 # frequently activated.
288 delete $ibx->{$_} foreach (qw(git mm search));
290 my $indexlevel = $ibx->{indexlevel};
291 if ($indexlevel && $indexlevel eq 'basic') {
292 $self->{parallel} = 0;
295 if ($self->{parallel}) {
296 pipe(my ($r, $w)) or die "pipe failed: $!";
297 # pipe for barrier notifications doesn't need to be big,
299 fcntl($w, 1031, 4096) if $^O eq 'linux';
300 $self->{bnote} = [ $r, $w ];
304 my $over = $self->{over};
306 $ibx->with_umask(sub {
307 $self->lock_acquire unless ($opt && $opt->{-skip_lock});
310 # xcpdb can change shard count while -watch is idle
311 my $nshards = count_shards($self);
312 if ($nshards && $nshards != $self->{shards}) {
313 $self->{shards} = $nshards;
316 # need to create all shards before initializing msgmap FD
317 my $max = $self->{shards} - 1;
319 # idx_shards must be visible to all forked processes
320 my $idx = $self->{idx_shards} = [];
321 for my $i (0..$max) {
322 push @$idx, PublicInbox::SearchIdxShard->new($self, $i);
325 # Now that all subprocesses are up, we can open the FDs
327 my $mm = $self->{mm} = PublicInbox::Msgmap->new_file(
328 "$self->{-inbox}->{inboxdir}/msgmap.sqlite3", 1);
329 $mm->{dbh}->begin_work;
333 # returns an array mapping [ epoch => latest_commit ]
334 # latest_commit may be undef if nothing was done to that epoch
335 # $replace_map = { $object_id => $strref, ... }
336 sub _replace_oids ($$$) {
337 my ($self, $mime, $replace_map) = @_;
339 my $pfx = "$self->{-inbox}->{inboxdir}/git";
340 my $rewrites = []; # epoch => commit
341 my $max = $self->{epoch_max};
343 unless (defined($max)) {
344 defined(my $latest = git_dir_latest($self, \$max)) or return;
345 $self->{epoch_max} = $max;
348 foreach my $i (0..$max) {
349 my $git_dir = "$pfx/$i.git";
351 my $git = PublicInbox::Git->new($git_dir);
352 my $im = $self->import_init($git, 0, 1);
353 $rewrites->[$i] = $im->replace_oids($mime, $replace_map);
359 sub content_hashes ($) {
361 my @chashes = ( content_hash($mime) );
363 # We still support Email::MIME, here, and
364 # Email::MIME->as_string doesn't always round-trip, so we may
365 # use a second content_hash
366 my $rt = content_hash(PublicInbox::Eml->new(\($mime->as_string)));
367 push @chashes, $rt if $chashes[0] ne $rt;
371 sub content_matches ($$) {
372 my ($chashes, $existing) = @_;
373 my $chash = content_hash($existing);
374 foreach (@$chashes) {
375 return 1 if $_ eq $chash
380 # used for removing or replacing (purging)
381 sub rewrite_internal ($$;$$$) {
382 my ($self, $old_mime, $cmt_msg, $new_mime, $sref) = @_;
384 my ($im, $need_reindex, $replace_map);
386 $replace_map = {}; # oid => sref
387 $need_reindex = [] if $new_mime;
389 $im = $self->importer;
391 my $over = $self->{over};
392 my $chashes = content_hashes($old_mime);
394 my $mids = mids($old_mime->header_obj);
396 # We avoid introducing new blobs into git since the raw content
397 # can be slightly different, so we do not need the user-supplied
398 # message now that we have the mids and content_hash
402 foreach my $mid (@$mids) {
403 my %gone; # num => [ smsg, $mime, raw ]
405 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
406 my $msg = get_blob($self, $smsg);
407 if (!defined($msg)) {
408 warn "broken smsg for $mid\n";
412 my $cur = PublicInbox::Eml->new($msg);
413 if (content_matches($chashes, $cur)) {
414 $gone{$smsg->{num}} = [ $smsg, $cur, \$orig ];
417 my $n = scalar keys %gone;
420 warn "BUG: multiple articles linked to <$mid>\n",
421 join(',', sort keys %gone), "\n";
423 foreach my $num (keys %gone) {
424 my ($smsg, $mime, $orig) = @{$gone{$num}};
425 # @removed should only be set once assuming
426 # no bugs in our deduplication code:
427 @removed = (undef, $mime, $smsg);
428 my $oid = $smsg->{blob};
430 $replace_map->{$oid} = $sref;
432 ($mark, undef) = $im->remove($orig, $cmt_msg);
436 if ($need_reindex) { # ->replace
437 push @$need_reindex, $smsg;
438 } else { # ->purge or ->remove
439 $self->{mm}->num_delete($num);
441 unindex_oid_remote($self, $oid, $mid);
446 my $cmt = $im->get_mark($mark);
447 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
449 if ($replace_map && scalar keys %$replace_map) {
450 my $rewrites = _replace_oids($self, $new_mime, $replace_map);
451 return { rewrites => $rewrites, need_reindex => $need_reindex };
453 defined($mark) ? @removed : undef;
456 # public (see PublicInbox::Import->remove), but note the 3rd element
457 # (retval[2]) is not part of the stable API shared with Import->remove
459 my ($self, $mime, $cmt_msg) = @_;
461 $self->{-inbox}->with_umask(sub {
462 @ret = rewrite_internal($self, $mime, $cmt_msg);
464 defined($ret[0]) ? @ret : undef;
467 sub _replace ($$;$$) {
468 my ($self, $old_mime, $new_mime, $sref) = @_;
469 my $rewritten = $self->{-inbox}->with_umask(sub {
470 rewrite_internal($self, $old_mime, undef, $new_mime, $sref);
473 my $rewrites = $rewritten->{rewrites};
474 # ->done is called if there are rewrites since we gc+prune from git
475 $self->idx_init if @$rewrites;
477 for my $i (0..$#$rewrites) {
478 defined(my $cmt = $rewrites->[$i]) or next;
479 $self->{last_commit}->[$i] = $cmt;
486 my ($self, $mime) = @_;
487 my $rewritten = _replace($self, $mime, undef, \'') or return;
488 $rewritten->{rewrites}
491 # returns the git object_id of $fh, does not write the object to FS
492 sub git_hash_raw ($$) {
493 my ($self, $raw) = @_;
494 # grab the expected OID we have to reindex:
495 open my $tmp_fh, '+>', undef or die "failed to open tmp: $!";
496 $tmp_fh->autoflush(1);
497 print $tmp_fh $$raw or die "print \$tmp_fh: $!";
498 sysseek($tmp_fh, 0, 0) or die "seek failed: $!";
500 my $git_dir = $self->{-inbox}->git->{git_dir};
501 my $cmd = ['git', "--git-dir=$git_dir", qw(hash-object --stdin)];
502 my $r = popen_rd($cmd, undef, { 0 => $tmp_fh });
504 chomp(my $oid = <$r>);
505 close $r or die "git hash-object failed: $?";
506 $oid =~ /\A[a-f0-9]{40}\z/ or die "OID not expected: $oid";
510 sub _check_mids_match ($$$) {
511 my ($old_list, $new_list, $hdrs) = @_;
512 my %old_mids = map { $_ => 1 } @$old_list;
513 my %new_mids = map { $_ => 1 } @$new_list;
514 my @old = keys %old_mids;
515 my @new = keys %new_mids;
516 my $err = "$hdrs may not be changed when replacing\n";
517 die $err if scalar(@old) != scalar(@new);
518 delete @new_mids{@old};
519 delete @old_mids{@new};
520 die $err if (scalar(keys %old_mids) || scalar(keys %new_mids));
523 # Changing Message-IDs or References with ->replace isn't supported.
524 # The rules for dealing with messages with multiple or conflicting
525 # Message-IDs are pretty complex and rethreading hasn't been fully
527 sub check_mids_match ($$) {
528 my ($old_mime, $new_mime) = @_;
529 my $old = $old_mime->header_obj;
530 my $new = $new_mime->header_obj;
531 _check_mids_match(mids($old), mids($new), 'Message-ID(s)');
532 _check_mids_match(references($old), references($new),
533 'References/In-Reply-To');
538 my ($self, $old_mime, $new_mime) = @_;
540 check_mids_match($old_mime, $new_mime);
542 # mutt will always add Content-Length:, Status:, Lines: when editing
543 PublicInbox::Import::drop_unwanted_headers($new_mime);
545 my $raw = $new_mime->as_string;
546 my $expect_oid = git_hash_raw($self, \$raw);
547 my $rewritten = _replace($self, $old_mime, $new_mime, \$raw) or return;
548 my $need_reindex = $rewritten->{need_reindex};
550 # just in case we have bugs in deduplication code:
551 my $n = scalar(@$need_reindex);
553 my $list = join(', ', map {
554 "$_->{num}: <$_->{mid}>"
557 W: rewritten $n messages matching content of original message (expected: 1).
558 W: possible bug in public-inbox, NNTP article IDs and Message-IDs follow:
563 # make sure we really got the OID:
564 my ($blob, $type, $bytes) = $self->{-inbox}->git->check($expect_oid);
565 $blob eq $expect_oid or die "BUG: $expect_oid not found after replace";
567 # don't leak FDs to Xapian:
568 $self->{-inbox}->git->cleanup;
570 # reindex modified messages:
571 for my $smsg (@$need_reindex) {
572 my $new_smsg = bless {
577 }, 'PublicInbox::Smsg';
578 my $v2w = { autime => $smsg->{ds}, cotime => $smsg->{ts} };
579 $new_smsg->populate($new_mime, $v2w);
580 do_idx($self, \$raw, $new_mime, $new_smsg);
582 $rewritten->{rewrites};
585 sub last_epoch_commit ($$;$) {
586 my ($self, $i, $cmt) = @_;
587 my $v = PublicInbox::Search::SCHEMA_VERSION();
588 $self->{mm}->last_commit_xap($v, $i, $cmt);
591 sub set_last_commits ($) {
593 defined(my $epoch_max = $self->{epoch_max}) or return;
594 my $last_commit = $self->{last_commit};
595 foreach my $i (0..$epoch_max) {
596 defined(my $cmt = $last_commit->[$i]) or next;
597 $last_commit->[$i] = undef;
598 last_epoch_commit($self, $i, $cmt);
604 $self->{bnote} or return;
606 my $barrier = { map { $_ => 1 } (0..$n) };
610 my ($self, $barrier) = @_;
611 my $bnote = $self->{bnote} or return;
613 while (scalar keys %$barrier) {
614 defined(my $l = readline($r)) or die "EOF on barrier_wait: $!";
615 $l =~ /\Abarrier (\d+)/ or die "bad line on barrier_wait: $l";
616 delete $barrier->{$1} or die "bad shard[$1] on barrier wait";
621 sub checkpoint ($;$) {
622 my ($self, $wait) = @_;
624 if (my $im = $self->{im}) {
631 my $shards = $self->{idx_shards};
633 my $dbh = $self->{mm}->{dbh};
635 # SQLite msgmap data is second in importance
638 # SQLite overview is third
639 $self->{over}->commit_lazy;
641 # Now deal with Xapian
643 my $barrier = $self->barrier_init(scalar @$shards);
645 # each shard needs to issue a barrier command
646 $_->remote_barrier for @$shards;
648 # wait for each Xapian shard
649 $self->barrier_wait($barrier);
651 $_->remote_commit for @$shards;
654 # last_commit is special, don't commit these until
655 # remote shards are done:
657 set_last_commits($self);
662 $self->{total_bytes} += $self->{transact_bytes};
663 $self->{transact_bytes} = 0;
666 # issue a write barrier to ensure all data is visible to other processes
667 # and read-only ops. Order of data importance is: git > SQLite > Xapian
669 sub barrier { checkpoint($_[0], 1) };
674 my $im = delete $self->{im};
675 $im->done if $im; # PublicInbox::Import::done
677 my $mm = delete $self->{mm};
678 $mm->{dbh}->commit if $mm;
679 my $shards = delete $self->{idx_shards};
681 $_->remote_close for @$shards;
683 $self->{over}->disconnect;
684 delete $self->{bnote};
685 my $nbytes = $self->{total_bytes};
686 $self->{total_bytes} = 0;
687 $self->lock_release(!!$nbytes) if $shards;
688 $self->{-inbox}->git->cleanup;
691 sub fill_alternates ($$) {
692 my ($self, $epoch) = @_;
694 my $pfx = "$self->{-inbox}->{inboxdir}/git";
695 my $all = "$self->{-inbox}->{inboxdir}/all.git";
698 PublicInbox::Import::init_bare($all);
700 my $info_dir = "$all/objects/info";
701 my $alt = "$info_dir/alternates";
705 open(my $fh, '<', $alt) or die "open < $alt: $!\n";
706 $mode = (stat($fh))[2] & 07777;
708 # we assign a sort score to every alternate and favor
709 # the newest (highest numbered) one when we
711 my $other = 0; # in case admin adds non-epoch repos
713 if (m!\A\Q../../\E([0-9]+)\.git/objects\z!) {
719 } split(/\n+/, do { local $/; <$fh> });
722 foreach my $i (0..$epoch) {
723 my $dir = "../../git/$i.git/objects";
724 if (!exists($alt{$dir}) && -d "$pfx/$i.git") {
731 my ($fh, $tmp) = tempfile('alt-XXXXXXXX', DIR => $info_dir);
732 print $fh join("\n", sort { $alt{$b} <=> $alt{$a} } keys %alt), "\n"
733 or die "print $tmp: $!\n";
734 chmod($mode, $fh) or die "fchmod $tmp: $!\n";
735 close $fh or die "close $tmp $!\n";
736 rename($tmp, $alt) or die "rename $tmp => $alt: $!\n";
740 my ($self, $epoch) = @_;
741 my $git_dir = "$self->{-inbox}->{inboxdir}/git/$epoch.git";
742 PublicInbox::Import::init_bare($git_dir);
743 my @cmd = (qw/git config/, "--file=$git_dir/config",
744 'include.path', '../../all.git/config');
745 PublicInbox::Import::run_die(\@cmd);
746 fill_alternates($self, $epoch);
751 my ($self, $max) = @_;
753 my $pfx = "$self->{-inbox}->{inboxdir}/git";
754 return unless -d $pfx;
756 opendir my $dh, $pfx or die "opendir $pfx: $!\n";
757 while (defined(my $git_dir = readdir($dh))) {
758 $git_dir =~ m!\A([0-9]+)\.git\z! or next;
761 $latest = "$pfx/$git_dir";
769 my $im = $self->{im};
771 if ($im->{bytes_added} < $self->{rotate_bytes}) {
778 my $git_dir = $self->git_init(++$self->{epoch_max});
779 my $git = PublicInbox::Git->new($git_dir);
780 return $self->import_init($git, 0);
785 my $latest = git_dir_latest($self, \$max);
786 if (defined $latest) {
787 my $git = PublicInbox::Git->new($latest);
788 my $packed_bytes = $git->packed_bytes;
789 my $unpacked_bytes = $packed_bytes / $PACKING_FACTOR;
791 if ($unpacked_bytes >= $self->{rotate_bytes}) {
794 $self->{epoch_max} = $max;
795 return $self->import_init($git, $packed_bytes);
798 $self->{epoch_max} = $epoch;
799 $latest = $self->git_init($epoch);
800 $self->import_init(PublicInbox::Git->new($latest), 0);
804 my ($self, $git, $packed_bytes, $tmp) = @_;
805 my $im = PublicInbox::Import->new($git, undef, undef, $self->{-inbox});
806 $im->{bytes_added} = int($packed_bytes / $PACKING_FACTOR);
807 $im->{lock_path} = undef;
808 $im->{path_type} = 'v2';
809 $self->{im} = $im unless $tmp;
815 my ($mid, $cur, $new) = @_;
817 my ($ah, $an) = tempfile('email-cur-XXXXXXXX', TMPDIR => 1);
818 print $ah $cur->as_string or die "print: $!";
819 close $ah or die "close: $!";
820 my ($bh, $bn) = tempfile('email-new-XXXXXXXX', TMPDIR => 1);
821 PublicInbox::Import::drop_unwanted_headers($new);
822 print $bh $new->as_string or die "print: $!";
823 close $bh or die "close: $!";
824 my $cmd = [ qw(diff -u), $an, $bn ];
825 print STDERR "# MID conflict <$mid>\n";
826 my $pid = spawn($cmd, undef, { 1 => 2 });
827 waitpid($pid, 0) == $pid or die "diff did not finish";
832 my ($self, $smsg) = @_;
833 if (my $im = $self->{im}) {
834 my $msg = $im->cat_blob($smsg->{blob});
837 # older message, should be in alternates
838 my $ibx = $self->{-inbox};
839 $ibx->msg_by_smsg($smsg);
842 sub content_exists ($$$) {
843 my ($self, $mime, $mid) = @_;
844 my $over = $self->{over};
845 my $chashes = content_hashes($mime);
847 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
848 my $msg = get_blob($self, $smsg);
849 if (!defined($msg)) {
850 warn "broken smsg for $mid\n";
853 my $cur = PublicInbox::Eml->new($msg);
854 return 1 if content_matches($chashes, $cur);
856 # XXX DEBUG_DIFF is experimental and may be removed
857 diff($mid, $cur, $mime) if $ENV{DEBUG_DIFF};
864 my $fh = delete $self->{reindex_pipe};
866 if (my $shards = $self->{idx_shards}) {
867 $_->atfork_child foreach @$shards;
869 if (my $im = $self->{im}) {
872 die "unexpected mm" if $self->{mm};
873 close $self->{bnote}->[0] or die "close bnote[0]: $!\n";
877 sub mark_deleted ($$$$) {
878 my ($self, $sync, $git, $oid) = @_;
879 return if PublicInbox::SearchIdx::too_big($self, $git, $oid);
880 my $msgref = $git->cat_file($oid);
881 my $mime = PublicInbox::Eml->new($$msgref);
882 my $mids = mids($mime->header_obj);
883 my $chash = content_hash($mime);
884 foreach my $mid (@$mids) {
885 $sync->{D}->{"$mid\0$chash"} = $oid;
889 sub reindex_checkpoint ($$$) {
890 my ($self, $sync, $git) = @_;
893 $sync->{mm_tmp}->atfork_prepare;
894 $self->done; # release lock
896 if (my $pr = $sync->{-opt}->{-progress}) {
897 my ($bn) = (split('/', $git->{git_dir}))[-1];
898 $pr->("$bn ".sprintf($sync->{-regen_fmt}, $sync->{nr}));
901 # allow -watch or -mda to write...
902 $self->idx_init; # reacquire lock
903 $sync->{mm_tmp}->atfork_parent;
906 # only for a few odd messages with multiple Message-IDs
907 sub reindex_oid_m ($$$$;$) {
908 my ($self, $sync, $git, $oid, $regen_num) = @_;
909 $self->{current_info} = "multi_mid $oid";
910 my ($num, $mid0, $len);
911 my $msgref = $git->cat_file($oid, \$len);
912 my $mime = PublicInbox::Eml->new($$msgref);
913 my $mids = mids($mime->header_obj);
914 my $chash = content_hash($mime);
915 die "BUG: reindex_oid_m called for <=1 mids" if scalar(@$mids) <= 1;
917 for my $mid (reverse @$mids) {
918 delete($sync->{D}->{"$mid\0$chash"}) and
919 die "BUG: reindex_oid should handle <$mid> delete";
921 my $over = $self->{over};
922 for my $mid (reverse @$mids) {
923 ($num, $mid0) = $over->num_mid0_for_oid($oid, $mid);
924 next unless defined $num;
925 if (defined($regen_num) && $regen_num != $num) {
926 die "BUG: regen(#$regen_num) != over(#$num)";
929 unless (defined($num)) {
930 for my $mid (reverse @$mids) {
931 # is this a number we got before?
932 my $n = $sync->{mm_tmp}->num_for($mid);
933 next unless defined $n;
934 next if defined($regen_num) && $regen_num != $n;
935 ($num, $mid0) = ($n, $mid);
940 $sync->{mm_tmp}->num_delete($num);
941 } elsif (defined $regen_num) {
943 for my $mid (reverse @$mids) {
944 $self->{mm}->mid_set($num, $mid) == 1 or next;
948 unless (defined $mid0) {
949 warn "E: cannot regen #$num\n";
952 } else { # fixup bugs in old mirrors on reindex
953 for my $mid (reverse @$mids) {
954 $num = $self->{mm}->mid_insert($mid);
955 next unless defined $num;
960 if ($sync->{reindex}) {
961 warn "reindex added #$num <$mid0>\n";
964 warn "E: cannot find article #\n";
974 }, 'PublicInbox::Smsg';
975 $smsg->populate($mime, $self);
976 if (do_idx($self, $msgref, $mime, $smsg)) {
977 reindex_checkpoint($self, $sync, $git);
981 sub check_unindexed ($$$) {
982 my ($self, $num, $mid0) = @_;
983 my $unindexed = $self->{unindexed} // {};
984 my $n = delete($unindexed->{$mid0});
985 defined $n or return;
987 die "BUG: unindexed $n != $num <$mid0>\n";
989 $self->{mm}->mid_set($num, $mid0);
993 sub multi_mid_q_push ($$$) {
994 my ($self, $sync, $oid) = @_;
995 my $multi_mid = $sync->{multi_mid} //= PublicInbox::MultiMidQueue->new;
996 if ($sync->{reindex}) { # no regen on reindex
997 $multi_mid->push_oid($oid, $self);
999 my $num = $sync->{regen}--;
1000 die "BUG: ran out of article numbers" if $num <= 0;
1001 $multi_mid->set_oid($num, $oid, $self);
1005 sub reindex_oid ($$$$) {
1006 my ($self, $sync, $git, $oid) = @_;
1007 return if PublicInbox::SearchIdx::too_big($self, $git, $oid);
1008 my ($num, $mid0, $len);
1009 my $msgref = $git->cat_file($oid, \$len);
1010 return if $len == 0; # purged
1011 my $mime = PublicInbox::Eml->new($$msgref);
1012 my $mids = mids($mime->header_obj);
1013 my $chash = content_hash($mime);
1015 if (scalar(@$mids) == 0) {
1016 warn "E: $oid has no Message-ID, skipping\n";
1018 } elsif (scalar(@$mids) == 1) {
1019 my $mid = $mids->[0];
1021 # was the file previously marked as deleted?, skip if so
1022 if (delete($sync->{D}->{"$mid\0$chash"})) {
1023 if (!$sync->{reindex}) {
1024 $num = $sync->{regen}--;
1025 $self->{mm}->num_highwater($num);
1030 # is this a number we got before?
1031 $num = $sync->{mm_tmp}->num_for($mid);
1034 check_unindexed($self, $num, $mid0);
1036 $num = $sync->{regen}--;
1037 die "BUG: ran out of article numbers" if $num <= 0;
1038 if ($self->{mm}->mid_set($num, $mid) != 1) {
1039 warn "E: unable to assign $num => <$mid>\n";
1044 } else { # multiple MIDs are a weird case:
1047 $del += delete($sync->{D}->{"$_\0$chash"}) // 0;
1050 unindex_oid_remote($self, $oid, $_) for @$mids;
1051 # do not delete from {mm_tmp}, since another
1052 # single-MID message may use it.
1053 } else { # handle them at the end:
1054 multi_mid_q_push($self, $sync, $oid);
1058 $sync->{mm_tmp}->mid_delete($mid0) or
1059 die "failed to delete <$mid0> for article #$num\n";
1066 }, 'PublicInbox::Smsg';
1067 $smsg->populate($mime, $self);
1068 if (do_idx($self, $msgref, $mime, $smsg)) {
1069 reindex_checkpoint($self, $sync, $git);
1073 # only update last_commit for $i on reindex iff newer than current
1074 sub update_last_commit ($$$$) {
1075 my ($self, $git, $i, $cmt) = @_;
1076 my $last = last_epoch_commit($self, $i);
1077 if (defined $last && is_ancestor($git, $last, $cmt)) {
1078 my @cmd = (qw(rev-list --count), "$last..$cmt");
1079 chomp(my $n = $git->qx(@cmd));
1080 return if $n ne '' && $n == 0;
1082 last_epoch_commit($self, $i, $cmt);
1085 sub git_dir_n ($$) { "$_[0]->{-inbox}->{inboxdir}/git/$_[1].git" }
1087 sub last_commits ($$) {
1088 my ($self, $epoch_max) = @_;
1090 for (my $i = $epoch_max; $i >= 0; $i--) {
1091 $heads->[$i] = last_epoch_commit($self, $i);
1096 *is_ancestor = *PublicInbox::SearchIdx::is_ancestor;
1098 # returns a revision range for git-log(1)
1099 sub log_range ($$$$$) {
1100 my ($self, $sync, $git, $i, $tip) = @_;
1101 my $opt = $sync->{-opt};
1102 my $pr = $opt->{-progress} if (($opt->{verbose} || 0) > 1);
1103 my $cur = $sync->{ranges}->[$i] or do {
1104 $pr->("$i.git indexing all of $tip") if $pr;
1105 return $tip; # all of it
1108 # fast equality check to avoid (v)fork+execve overhead
1110 $sync->{ranges}->[$i] = undef;
1114 my $range = "$cur..$tip";
1115 $pr->("$i.git checking contiguity... ") if $pr;
1116 if (is_ancestor($git, $cur, $tip)) { # common case
1117 $pr->("OK\n") if $pr;
1118 my $n = $git->qx(qw(rev-list --count), $range);
1121 $sync->{ranges}->[$i] = undef;
1122 $pr->("$i.git has nothing new\n") if $pr;
1123 return; # nothing to do
1125 $pr->("$i.git has $n changes since $cur\n") if $pr;
1127 $pr->("FAIL\n") if $pr;
1129 discontiguous range: $range
1130 Rewritten history? (in $git->{git_dir})
1132 chomp(my $base = $git->qx('merge-base', $tip, $cur));
1134 $range = "$base..$tip";
1135 warn "found merge-base: $base\n"
1138 warn "discarding history at $cur\n";
1141 reindexing $git->{git_dir} starting at
1144 $sync->{unindex_range}->{$i} = "$base..$cur";
1149 sub sync_prepare ($$$) {
1150 my ($self, $sync, $epoch_max) = @_;
1151 my $pr = $sync->{-opt}->{-progress};
1153 my $head = $self->{-inbox}->{ref_head} || 'refs/heads/master';
1155 # reindex stops at the current heads and we later rerun index_sync
1157 my $reindex_heads = last_commits($self, $epoch_max) if $sync->{reindex};
1159 for (my $i = $epoch_max; $i >= 0; $i--) {
1160 die 'BUG: already indexing!' if $self->{reindex_pipe};
1161 my $git_dir = git_dir_n($self, $i);
1162 -d $git_dir or next; # missing epochs are fine
1163 my $git = PublicInbox::Git->new($git_dir);
1164 if ($reindex_heads) {
1165 $head = $reindex_heads->[$i] or next;
1167 chomp(my $tip = $git->qx(qw(rev-parse -q --verify), $head));
1169 next if $?; # new repo
1170 my $range = log_range($self, $sync, $git, $i, $tip) or next;
1171 $sync->{ranges}->[$i] = $range;
1173 # can't use 'rev-list --count' if we use --diff-filter
1174 $pr->("$i.git counting $range ... ") if $pr;
1176 my $fh = $git->popen(qw(log --pretty=tformat:%H
1177 --no-notes --no-color --no-renames
1178 --diff-filter=AM), $range, '--', 'm');
1180 close $fh or die "git log failed: \$?=$?";
1181 $pr->("$n\n") if $pr;
1185 return 0 if (!$regen_max && !keys(%{$self->{unindex_range}}));
1187 # reindex should NOT see new commits anymore, if we do,
1188 # it's a problem and we need to notice it via die()
1189 my $pad = length($regen_max) + 1;
1190 $sync->{-regen_fmt} = "% ${pad}u/$regen_max\n";
1192 return -1 if $sync->{reindex};
1193 $regen_max + $self->{mm}->num_highwater() || 0;
1196 sub unindex_oid_remote ($$$) {
1197 my ($self, $oid, $mid) = @_;
1198 $_->remote_remove($oid, $mid) foreach @{$self->{idx_shards}};
1199 $self->{over}->remove_oid($oid, $mid);
1202 sub unindex_oid ($$$;$) {
1203 my ($self, $git, $oid, $unindexed) = @_;
1204 my $mm = $self->{mm};
1205 my $msgref = $git->cat_file($oid);
1206 my $mime = PublicInbox::Eml->new($msgref);
1207 my $mids = mids($mime->header_obj);
1208 $mime = $msgref = undef;
1209 my $over = $self->{over};
1210 foreach my $mid (@$mids) {
1213 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
1214 $gone{$smsg->{num}} = 1 if $oid eq $smsg->{blob};
1217 my $n = scalar keys %gone;
1220 warn "BUG: multiple articles linked to $oid\n",
1221 join(',',sort keys %gone), "\n";
1223 foreach my $num (keys %gone) {
1225 my $mid0 = $mm->mid_for($num);
1226 $unindexed->{$mid0} = $num;
1228 $mm->num_delete($num);
1230 unindex_oid_remote($self, $oid, $mid);
1234 my $x40 = qr/[a-f0-9]{40}/;
1235 sub unindex ($$$$) {
1236 my ($self, $sync, $git, $unindex_range) = @_;
1237 my $unindexed = $self->{unindexed} ||= {}; # $mid0 => $num
1238 my $before = scalar keys %$unindexed;
1239 # order does not matter, here:
1240 my @cmd = qw(log --raw -r
1241 --no-notes --no-color --no-abbrev --no-renames);
1242 my $fh = $self->{reindex_pipe} = $git->popen(@cmd, $unindex_range);
1244 /\A:\d{6} 100644 $x40 ($x40) [AM]\tm$/o or next;
1245 unindex_oid($self, $git, $1, $unindexed);
1247 delete $self->{reindex_pipe};
1248 close $fh or die "git log failed: \$?=$?";
1250 return unless $sync->{-opt}->{prune};
1251 my $after = scalar keys %$unindexed;
1252 return if $before == $after;
1254 # ensure any blob can not longer be accessed via dumb HTTP
1255 PublicInbox::Import::run_die(['git', "--git-dir=$git->{git_dir}",
1256 qw(-c gc.reflogExpire=now gc --prune=all --quiet)]);
1259 sub sync_ranges ($$$) {
1260 my ($self, $sync, $epoch_max) = @_;
1261 my $reindex = $sync->{reindex};
1263 return last_commits($self, $epoch_max) unless $reindex;
1264 return [] if ref($reindex) ne 'HASH';
1266 my $ranges = $reindex->{from}; # arrayref;
1267 if (ref($ranges) ne 'ARRAY') {
1268 die 'BUG: $reindex->{from} not an ARRAY';
1273 sub index_epoch ($$$) {
1274 my ($self, $sync, $i) = @_;
1276 my $git_dir = git_dir_n($self, $i);
1277 die 'BUG: already reindexing!' if $self->{reindex_pipe};
1278 -d $git_dir or return; # missing epochs are fine
1279 fill_alternates($self, $i);
1280 my $git = PublicInbox::Git->new($git_dir);
1281 if (my $unindex_range = delete $sync->{unindex_range}->{$i}) {
1282 unindex($self, $sync, $git, $unindex_range);
1284 defined(my $range = $sync->{ranges}->[$i]) or return;
1285 if (my $pr = $sync->{-opt}->{-progress}) {
1286 $pr->("$i.git indexing $range\n");
1289 my @cmd = qw(log --raw -r --pretty=tformat:%H.%at.%ct
1290 --no-notes --no-color --no-abbrev --no-renames);
1291 my $fh = $self->{reindex_pipe} = $git->popen(@cmd, $range);
1295 $self->{current_info} = "$i.git $_";
1296 if (/\A($x40)\.([0-9]+)\.([0-9]+)$/o) {
1298 $self->{autime} = $2;
1299 $self->{cotime} = $3;
1300 } elsif (/\A:\d{6} 100644 $x40 ($x40) [AM]\tm$/o) {
1301 reindex_oid($self, $sync, $git, $1);
1302 } elsif (/\A:\d{6} 100644 $x40 ($x40) [AM]\td$/o) {
1303 mark_deleted($self, $sync, $git, $1);
1306 close $fh or die "git log failed: \$?=$?";
1307 delete @$self{qw(reindex_pipe autime cotime)};
1308 update_last_commit($self, $git, $i, $cmt) if defined $cmt;
1311 # public, called by public-inbox-index
1313 my ($self, $opt) = @_;
1315 my $pr = $opt->{-progress};
1317 my $latest = git_dir_latest($self, \$epoch_max);
1318 return unless defined $latest;
1319 $self->idx_init($opt); # acquire lock
1321 D => {}, # "$mid\0$chash" => $oid
1322 unindex_range => {}, # EPOCH => oid_old..oid_new
1323 reindex => $opt->{reindex},
1326 $sync->{ranges} = sync_ranges($self, $sync, $epoch_max);
1327 $sync->{regen} = sync_prepare($self, $sync, $epoch_max);
1329 if ($sync->{regen}) {
1330 # tmp_clone seems to fail if inside a transaction, so
1331 # we rollback here (because we opened {mm} for reading)
1332 # Note: we do NOT rely on DBI transactions for atomicity;
1333 # only for batch performance.
1334 $self->{mm}->{dbh}->rollback;
1335 $self->{mm}->{dbh}->begin_work;
1336 $sync->{mm_tmp} = $self->{mm}->tmp_clone;
1339 # work backwards through history
1340 for (my $i = $epoch_max; $i >= 0; $i--) {
1341 index_epoch($self, $sync, $i);
1344 # unindex is required for leftovers if "deletes" affect messages
1345 # in a previous fetch+index window:
1347 if (my @leftovers = values %{delete $sync->{D}}) {
1348 $git = $self->{-inbox}->git;
1349 for my $oid (@leftovers) {
1350 $self->{current_info} = "leftover $oid";
1351 unindex_oid($self, $git, $oid);
1354 if (my $multi_mid = delete $sync->{multi_mid}) {
1355 $git //= $self->{-inbox}->git;
1356 my $min = $multi_mid->{min};
1357 my $max = $multi_mid->{max};
1358 if ($sync->{reindex}) {
1359 # we may need to create new Message-IDs if mirrors
1360 # were initially indexed with old versions
1361 for (my $i = $max; $i >= $min; $i--) {
1363 $oid = $multi_mid->get_oid($i, $self) or next;
1364 next unless defined $oid;
1365 reindex_oid_m($self, $sync, $git, $oid);
1367 } else { # regen on initial index
1368 for my $num ($min..$max) {
1370 $oid = $multi_mid->get_oid($num, $self) or next;
1371 reindex_oid_m($self, $sync, $git, $oid, $num);
1375 $git->cleanup if $git;
1378 if (my $nr = $sync->{nr}) {
1379 my $pr = $sync->{-opt}->{-progress};
1380 $pr->('all.git '.sprintf($sync->{-regen_fmt}, $nr)) if $pr;
1383 # reindex does not pick up new changes, so we rerun w/o it:
1384 if ($opt->{reindex}) {
1387 delete @again{qw(reindex -skip_lock)};
1388 index_sync($self, \%again);