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::MsgTime qw(msg_timestamp msg_datestamp);
23 use PublicInbox::MultiMidQueue;
24 use IO::Handle; # ->autoflush
25 use File::Temp qw(tempfile);
27 # an estimate of the post-packed size to the raw uncompressed size
28 my $PACKING_FACTOR = 0.4;
30 # SATA storage lags behind what CPUs are capable of, so relying on
31 # nproc(1) can be misleading and having extra Xapian shards is a
32 # waste of FDs and space. It can also lead to excessive IO latency
33 # and slow things down. Users on NVME or other fast storage can
34 # use the NPROC env or switches in our script/public-inbox-* programs
35 # to increase Xapian shards
36 our $NPROC_MAX_DEFAULT = 4;
39 for my $nproc (qw(nproc gnproc)) { # GNU coreutils nproc
40 `$nproc 2>/dev/null` =~ /^(\d+)$/ and return $1;
43 # getconf(1) is POSIX, but *NPROCESSORS* vars are not
44 for (qw(_NPROCESSORS_ONLN NPROCESSORS_ONLN)) {
45 `getconf $_ 2>/dev/null` =~ /^(\d+)$/ and return $1;
48 # should we bother with `sysctl hw.ncpu`? Those only give
49 # us total processor count, not online processor count.
53 sub nproc_shards ($) {
55 my $n = $creat_opt->{nproc} if ref($creat_opt) eq 'HASH';
58 # assume 2 cores if not detectable or zero
59 state $NPROC_DETECTED = detect_nproc() || 2;
61 $n = $NPROC_MAX_DEFAULT if $n > $NPROC_MAX_DEFAULT;
64 # subtract for the main process and git-fast-import
69 sub count_shards ($) {
72 my $xpfx = $self->{xpfx};
74 # always load existing shards in case core count changes:
75 # Also, shard count may change while -watch is running
76 # due to "xcpdb --reshard"
79 foreach my $shard (<$xpfx/*>) {
80 -d $shard && $shard =~ m!/[0-9]+\z! or next;
81 $XapianDatabase //= do {
82 require PublicInbox::Search;
83 PublicInbox::Search::load_xapian();
84 $PublicInbox::Search::X{Database};
87 $XapianDatabase->new($shard)->close;
96 # $creat may be any true value, or 0/undef. A hashref is true,
97 # and $creat->{nproc} may be set to an integer
98 my ($class, $v2ibx, $creat) = @_;
99 $v2ibx = PublicInbox::InboxWritable->new($v2ibx);
100 my $dir = $v2ibx->assert_usable_dir;
104 File::Path::mkpath($dir);
106 die "$dir does not exist\n";
109 $v2ibx->umask_prepare;
111 my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION;
114 im => undef, # PublicInbox::Import
119 over => PublicInbox::OverIdx->new("$xpfx/over.sqlite3", 1),
120 lock_path => "$dir/inbox.lock",
121 # limit each git repo (epoch) to 1GB or so
122 rotate_bytes => int((1024 * 1024 * 1024) / $PACKING_FACTOR),
123 last_commit => [], # git repo -> commit
125 $self->{shards} = count_shards($self) || nproc_shards($creat);
126 $self->{index_max_size} = $v2ibx->{index_max_size};
132 my ($self, $shards, $skip_epoch) = @_;
133 if (defined $shards) {
134 $self->{parallel} = 0 if $shards == 0;
135 $self->{shards} = $shards if $shards > 0;
139 git_dir_latest($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 $smsg->{ds} //= msg_datestamp($mime->header_obj, $self->{autime});
160 $smsg->{ts} //= msg_timestamp($mime->header_obj, $self->{cotime});
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->{bytes};
165 $n >= ($PublicInbox::SearchIdx::BATCH_BYTES * $self->{shards});
169 my ($self, $mime, $check_cb) = @_;
173 $mime = $check_cb->($mime) 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 ($$) {
528 my ($old_mime, $new_mime) = @_;
529 my $old = $old_mime->header_obj;
530 my $new = $new_mime->header_obj;
531 _check_mids_match(mids($old), mids($new), 'Message-ID(s)');
532 _check_mids_match(references($old), references($new),
533 'References/In-Reply-To');
538 my ($self, $old_mime, $new_mime) = @_;
540 check_mids_match($old_mime, $new_mime);
542 # mutt will always add Content-Length:, Status:, Lines: when editing
543 PublicInbox::Import::drop_unwanted_headers($new_mime);
545 my $raw = $new_mime->as_string;
546 my $expect_oid = git_hash_raw($self, \$raw);
547 my $rewritten = _replace($self, $old_mime, $new_mime, \$raw) or return;
548 my $need_reindex = $rewritten->{need_reindex};
550 # just in case we have bugs in deduplication code:
551 my $n = scalar(@$need_reindex);
553 my $list = join(', ', map {
554 "$_->{num}: <$_->{mid}>"
557 W: rewritten $n messages matching content of original message (expected: 1).
558 W: possible bug in public-inbox, NNTP article IDs and Message-IDs follow:
563 # make sure we really got the OID:
564 my ($blob, $type, $bytes) = $self->{-inbox}->git->check($expect_oid);
565 $blob eq $expect_oid or die "BUG: $expect_oid not found after replace";
567 # don't leak FDs to Xapian:
568 $self->{-inbox}->git->cleanup;
570 # reindex modified messages:
571 for my $smsg (@$need_reindex) {
572 my $new_smsg = bless {
577 }, 'PublicInbox::Smsg';
578 do_idx($self, \$raw, $new_mime, $new_smsg);
580 $rewritten->{rewrites};
583 sub last_epoch_commit ($$;$) {
584 my ($self, $i, $cmt) = @_;
585 my $v = PublicInbox::Search::SCHEMA_VERSION();
586 $self->{mm}->last_commit_xap($v, $i, $cmt);
589 sub set_last_commits ($) {
591 defined(my $epoch_max = $self->{epoch_max}) or return;
592 my $last_commit = $self->{last_commit};
593 foreach my $i (0..$epoch_max) {
594 defined(my $cmt = $last_commit->[$i]) or next;
595 $last_commit->[$i] = undef;
596 last_epoch_commit($self, $i, $cmt);
602 $self->{bnote} or return;
604 my $barrier = { map { $_ => 1 } (0..$n) };
608 my ($self, $barrier) = @_;
609 my $bnote = $self->{bnote} or return;
611 while (scalar keys %$barrier) {
612 defined(my $l = readline($r)) or die "EOF on barrier_wait: $!";
613 $l =~ /\Abarrier (\d+)/ or die "bad line on barrier_wait: $l";
614 delete $barrier->{$1} or die "bad shard[$1] on barrier wait";
619 sub checkpoint ($;$) {
620 my ($self, $wait) = @_;
622 if (my $im = $self->{im}) {
629 my $shards = $self->{idx_shards};
631 my $dbh = $self->{mm}->{dbh};
633 # SQLite msgmap data is second in importance
636 # SQLite overview is third
637 $self->{over}->commit_lazy;
639 # Now deal with Xapian
641 my $barrier = $self->barrier_init(scalar @$shards);
643 # each shard needs to issue a barrier command
644 $_->remote_barrier for @$shards;
646 # wait for each Xapian shard
647 $self->barrier_wait($barrier);
649 $_->remote_commit for @$shards;
652 # last_commit is special, don't commit these until
653 # remote shards are done:
655 set_last_commits($self);
660 $self->{transact_bytes} = 0;
663 # issue a write barrier to ensure all data is visible to other processes
664 # and read-only ops. Order of data importance is: git > SQLite > Xapian
666 sub barrier { checkpoint($_[0], 1) };
671 my $im = delete $self->{im};
672 $im->done if $im; # PublicInbox::Import::done
674 my $mm = delete $self->{mm};
675 $mm->{dbh}->commit if $mm;
676 my $shards = delete $self->{idx_shards};
678 $_->remote_close for @$shards;
680 $self->{over}->disconnect;
681 delete $self->{bnote};
682 $self->{transact_bytes} = 0;
683 $self->lock_release if $shards;
684 $self->{-inbox}->git->cleanup;
687 sub fill_alternates ($$) {
688 my ($self, $epoch) = @_;
690 my $pfx = "$self->{-inbox}->{inboxdir}/git";
691 my $all = "$self->{-inbox}->{inboxdir}/all.git";
694 PublicInbox::Import::init_bare($all);
696 my $info_dir = "$all/objects/info";
697 my $alt = "$info_dir/alternates";
701 open(my $fh, '<', $alt) or die "open < $alt: $!\n";
702 $mode = (stat($fh))[2] & 07777;
704 # we assign a sort score to every alternate and favor
705 # the newest (highest numbered) one when we
707 my $other = 0; # in case admin adds non-epoch repos
709 if (m!\A\Q../../\E([0-9]+)\.git/objects\z!) {
715 } split(/\n+/, do { local $/; <$fh> });
718 foreach my $i (0..$epoch) {
719 my $dir = "../../git/$i.git/objects";
720 if (!exists($alt{$dir}) && -d "$pfx/$i.git") {
727 my ($fh, $tmp) = tempfile('alt-XXXXXXXX', DIR => $info_dir);
728 print $fh join("\n", sort { $alt{$b} <=> $alt{$a} } keys %alt), "\n"
729 or die "print $tmp: $!\n";
730 chmod($mode, $fh) or die "fchmod $tmp: $!\n";
731 close $fh or die "close $tmp $!\n";
732 rename($tmp, $alt) or die "rename $tmp => $alt: $!\n";
736 my ($self, $epoch) = @_;
737 my $git_dir = "$self->{-inbox}->{inboxdir}/git/$epoch.git";
738 PublicInbox::Import::init_bare($git_dir);
739 my @cmd = (qw/git config/, "--file=$git_dir/config",
740 'include.path', '../../all.git/config');
741 PublicInbox::Import::run_die(\@cmd);
742 fill_alternates($self, $epoch);
747 my ($self, $max) = @_;
749 my $pfx = "$self->{-inbox}->{inboxdir}/git";
750 return unless -d $pfx;
752 opendir my $dh, $pfx or die "opendir $pfx: $!\n";
753 while (defined(my $git_dir = readdir($dh))) {
754 $git_dir =~ m!\A([0-9]+)\.git\z! or next;
757 $latest = "$pfx/$git_dir";
765 my $im = $self->{im};
767 if ($im->{bytes_added} < $self->{rotate_bytes}) {
774 my $git_dir = $self->git_init(++$self->{epoch_max});
775 my $git = PublicInbox::Git->new($git_dir);
776 return $self->import_init($git, 0);
781 my $latest = git_dir_latest($self, \$max);
782 if (defined $latest) {
783 my $git = PublicInbox::Git->new($latest);
784 my $packed_bytes = $git->packed_bytes;
785 my $unpacked_bytes = $packed_bytes / $PACKING_FACTOR;
787 if ($unpacked_bytes >= $self->{rotate_bytes}) {
790 $self->{epoch_max} = $max;
791 return $self->import_init($git, $packed_bytes);
794 $self->{epoch_max} = $epoch;
795 $latest = $self->git_init($epoch);
796 $self->import_init(PublicInbox::Git->new($latest), 0);
800 my ($self, $git, $packed_bytes, $tmp) = @_;
801 my $im = PublicInbox::Import->new($git, undef, undef, $self->{-inbox});
802 $im->{bytes_added} = int($packed_bytes / $PACKING_FACTOR);
803 $im->{lock_path} = undef;
804 $im->{path_type} = 'v2';
805 $self->{im} = $im unless $tmp;
811 my ($mid, $cur, $new) = @_;
813 my ($ah, $an) = tempfile('email-cur-XXXXXXXX', TMPDIR => 1);
814 print $ah $cur->as_string or die "print: $!";
815 close $ah or die "close: $!";
816 my ($bh, $bn) = tempfile('email-new-XXXXXXXX', TMPDIR => 1);
817 PublicInbox::Import::drop_unwanted_headers($new);
818 print $bh $new->as_string or die "print: $!";
819 close $bh or die "close: $!";
820 my $cmd = [ qw(diff -u), $an, $bn ];
821 print STDERR "# MID conflict <$mid>\n";
822 my $pid = spawn($cmd, undef, { 1 => 2 });
823 waitpid($pid, 0) == $pid or die "diff did not finish";
828 my ($self, $smsg) = @_;
829 if (my $im = $self->{im}) {
830 my $msg = $im->cat_blob($smsg->{blob});
833 # older message, should be in alternates
834 my $ibx = $self->{-inbox};
835 $ibx->msg_by_smsg($smsg);
838 sub content_exists ($$$) {
839 my ($self, $mime, $mid) = @_;
840 my $over = $self->{over};
841 my $chashes = content_hashes($mime);
843 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
844 my $msg = get_blob($self, $smsg);
845 if (!defined($msg)) {
846 warn "broken smsg for $mid\n";
849 my $cur = PublicInbox::Eml->new($msg);
850 return 1 if content_matches($chashes, $cur);
852 # XXX DEBUG_DIFF is experimental and may be removed
853 diff($mid, $cur, $mime) if $ENV{DEBUG_DIFF};
860 my $fh = delete $self->{reindex_pipe};
862 if (my $shards = $self->{idx_shards}) {
863 $_->atfork_child foreach @$shards;
865 if (my $im = $self->{im}) {
868 die "unexpected mm" if $self->{mm};
869 close $self->{bnote}->[0] or die "close bnote[0]: $!\n";
873 sub mark_deleted ($$$$) {
874 my ($self, $sync, $git, $oid) = @_;
875 return if PublicInbox::SearchIdx::too_big($self, $git, $oid);
876 my $msgref = $git->cat_file($oid);
877 my $mime = PublicInbox::Eml->new($$msgref);
878 my $mids = mids($mime->header_obj);
879 my $chash = content_hash($mime);
880 foreach my $mid (@$mids) {
881 $sync->{D}->{"$mid\0$chash"} = $oid;
885 sub reindex_checkpoint ($$$) {
886 my ($self, $sync, $git) = @_;
889 $sync->{mm_tmp}->atfork_prepare;
890 $self->done; # release lock
892 if (my $pr = $sync->{-opt}->{-progress}) {
893 my ($bn) = (split('/', $git->{git_dir}))[-1];
894 $pr->("$bn ".sprintf($sync->{-regen_fmt}, $sync->{nr}));
897 # allow -watch or -mda to write...
898 $self->idx_init; # reacquire lock
899 $sync->{mm_tmp}->atfork_parent;
902 # only for a few odd messages with multiple Message-IDs
903 sub reindex_oid_m ($$$$;$) {
904 my ($self, $sync, $git, $oid, $regen_num) = @_;
905 $self->{current_info} = "multi_mid $oid";
906 my ($num, $mid0, $len);
907 my $msgref = $git->cat_file($oid, \$len);
908 my $mime = PublicInbox::Eml->new($$msgref);
909 my $mids = mids($mime->header_obj);
910 my $chash = content_hash($mime);
911 die "BUG: reindex_oid_m called for <=1 mids" if scalar(@$mids) <= 1;
913 for my $mid (reverse @$mids) {
914 delete($sync->{D}->{"$mid\0$chash"}) and
915 die "BUG: reindex_oid should handle <$mid> delete";
917 my $over = $self->{over};
918 for my $mid (reverse @$mids) {
919 ($num, $mid0) = $over->num_mid0_for_oid($oid, $mid);
920 next unless defined $num;
921 if (defined($regen_num) && $regen_num != $num) {
922 die "BUG: regen(#$regen_num) != over(#$num)";
925 unless (defined($num)) {
926 for my $mid (reverse @$mids) {
927 # is this a number we got before?
928 my $n = $sync->{mm_tmp}->num_for($mid);
929 next unless defined $n;
930 next if defined($regen_num) && $regen_num != $n;
931 ($num, $mid0) = ($n, $mid);
936 $sync->{mm_tmp}->num_delete($num);
937 } elsif (defined $regen_num) {
939 for my $mid (reverse @$mids) {
940 $self->{mm}->mid_set($num, $mid) == 1 or next;
944 unless (defined $mid0) {
945 warn "E: cannot regen #$num\n";
948 } else { # fixup bugs in old mirrors on reindex
949 for my $mid (reverse @$mids) {
950 $num = $self->{mm}->mid_insert($mid);
951 next unless defined $num;
956 if ($sync->{reindex}) {
957 warn "reindex added #$num <$mid0>\n";
960 warn "E: cannot find article #\n";
970 }, 'PublicInbox::Smsg';
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 if (do_idx($self, $msgref, $mime, $smsg)) {
1063 reindex_checkpoint($self, $sync, $git);
1067 # only update last_commit for $i on reindex iff newer than current
1068 sub update_last_commit ($$$$) {
1069 my ($self, $git, $i, $cmt) = @_;
1070 my $last = last_epoch_commit($self, $i);
1071 if (defined $last && is_ancestor($git, $last, $cmt)) {
1072 my @cmd = (qw(rev-list --count), "$last..$cmt");
1073 chomp(my $n = $git->qx(@cmd));
1074 return if $n ne '' && $n == 0;
1076 last_epoch_commit($self, $i, $cmt);
1079 sub git_dir_n ($$) { "$_[0]->{-inbox}->{inboxdir}/git/$_[1].git" }
1081 sub last_commits ($$) {
1082 my ($self, $epoch_max) = @_;
1084 for (my $i = $epoch_max; $i >= 0; $i--) {
1085 $heads->[$i] = last_epoch_commit($self, $i);
1090 *is_ancestor = *PublicInbox::SearchIdx::is_ancestor;
1092 # returns a revision range for git-log(1)
1093 sub log_range ($$$$$) {
1094 my ($self, $sync, $git, $i, $tip) = @_;
1095 my $opt = $sync->{-opt};
1096 my $pr = $opt->{-progress} if (($opt->{verbose} || 0) > 1);
1097 my $cur = $sync->{ranges}->[$i] or do {
1098 $pr->("$i.git indexing all of $tip") if $pr;
1099 return $tip; # all of it
1102 # fast equality check to avoid (v)fork+execve overhead
1104 $sync->{ranges}->[$i] = undef;
1108 my $range = "$cur..$tip";
1109 $pr->("$i.git checking contiguity... ") if $pr;
1110 if (is_ancestor($git, $cur, $tip)) { # common case
1111 $pr->("OK\n") if $pr;
1112 my $n = $git->qx(qw(rev-list --count), $range);
1115 $sync->{ranges}->[$i] = undef;
1116 $pr->("$i.git has nothing new\n") if $pr;
1117 return; # nothing to do
1119 $pr->("$i.git has $n changes since $cur\n") if $pr;
1121 $pr->("FAIL\n") if $pr;
1123 discontiguous range: $range
1124 Rewritten history? (in $git->{git_dir})
1126 chomp(my $base = $git->qx('merge-base', $tip, $cur));
1128 $range = "$base..$tip";
1129 warn "found merge-base: $base\n"
1132 warn "discarding history at $cur\n";
1135 reindexing $git->{git_dir} starting at
1138 $sync->{unindex_range}->{$i} = "$base..$cur";
1143 sub sync_prepare ($$$) {
1144 my ($self, $sync, $epoch_max) = @_;
1145 my $pr = $sync->{-opt}->{-progress};
1147 my $head = $self->{-inbox}->{ref_head} || 'refs/heads/master';
1149 # reindex stops at the current heads and we later rerun index_sync
1151 my $reindex_heads = last_commits($self, $epoch_max) if $sync->{reindex};
1153 for (my $i = $epoch_max; $i >= 0; $i--) {
1154 die 'BUG: already indexing!' if $self->{reindex_pipe};
1155 my $git_dir = git_dir_n($self, $i);
1156 -d $git_dir or next; # missing epochs are fine
1157 my $git = PublicInbox::Git->new($git_dir);
1158 if ($reindex_heads) {
1159 $head = $reindex_heads->[$i] or next;
1161 chomp(my $tip = $git->qx(qw(rev-parse -q --verify), $head));
1163 next if $?; # new repo
1164 my $range = log_range($self, $sync, $git, $i, $tip) or next;
1165 $sync->{ranges}->[$i] = $range;
1167 # can't use 'rev-list --count' if we use --diff-filter
1168 $pr->("$i.git counting $range ... ") if $pr;
1170 my $fh = $git->popen(qw(log --pretty=tformat:%H
1171 --no-notes --no-color --no-renames
1172 --diff-filter=AM), $range, '--', 'm');
1174 close $fh or die "git log failed: \$?=$?";
1175 $pr->("$n\n") if $pr;
1179 return 0 if (!$regen_max && !keys(%{$self->{unindex_range}}));
1181 # reindex should NOT see new commits anymore, if we do,
1182 # it's a problem and we need to notice it via die()
1183 my $pad = length($regen_max) + 1;
1184 $sync->{-regen_fmt} = "% ${pad}u/$regen_max\n";
1186 return -1 if $sync->{reindex};
1187 $regen_max + $self->{mm}->num_highwater() || 0;
1190 sub unindex_oid_remote ($$$) {
1191 my ($self, $oid, $mid) = @_;
1192 $_->remote_remove($oid, $mid) foreach @{$self->{idx_shards}};
1193 $self->{over}->remove_oid($oid, $mid);
1196 sub unindex_oid ($$$;$) {
1197 my ($self, $git, $oid, $unindexed) = @_;
1198 my $mm = $self->{mm};
1199 my $msgref = $git->cat_file($oid);
1200 my $mime = PublicInbox::Eml->new($msgref);
1201 my $mids = mids($mime->header_obj);
1202 $mime = $msgref = undef;
1203 my $over = $self->{over};
1204 foreach my $mid (@$mids) {
1207 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
1208 $gone{$smsg->{num}} = 1 if $oid eq $smsg->{blob};
1211 my $n = scalar keys %gone;
1214 warn "BUG: multiple articles linked to $oid\n",
1215 join(',',sort keys %gone), "\n";
1217 foreach my $num (keys %gone) {
1219 my $mid0 = $mm->mid_for($num);
1220 $unindexed->{$mid0} = $num;
1222 $mm->num_delete($num);
1224 unindex_oid_remote($self, $oid, $mid);
1228 my $x40 = qr/[a-f0-9]{40}/;
1229 sub unindex ($$$$) {
1230 my ($self, $sync, $git, $unindex_range) = @_;
1231 my $unindexed = $self->{unindexed} ||= {}; # $mid0 => $num
1232 my $before = scalar keys %$unindexed;
1233 # order does not matter, here:
1234 my @cmd = qw(log --raw -r
1235 --no-notes --no-color --no-abbrev --no-renames);
1236 my $fh = $self->{reindex_pipe} = $git->popen(@cmd, $unindex_range);
1238 /\A:\d{6} 100644 $x40 ($x40) [AM]\tm$/o or next;
1239 unindex_oid($self, $git, $1, $unindexed);
1241 delete $self->{reindex_pipe};
1242 close $fh or die "git log failed: \$?=$?";
1244 return unless $sync->{-opt}->{prune};
1245 my $after = scalar keys %$unindexed;
1246 return if $before == $after;
1248 # ensure any blob can not longer be accessed via dumb HTTP
1249 PublicInbox::Import::run_die(['git', "--git-dir=$git->{git_dir}",
1250 qw(-c gc.reflogExpire=now gc --prune=all --quiet)]);
1253 sub sync_ranges ($$$) {
1254 my ($self, $sync, $epoch_max) = @_;
1255 my $reindex = $sync->{reindex};
1257 return last_commits($self, $epoch_max) unless $reindex;
1258 return [] if ref($reindex) ne 'HASH';
1260 my $ranges = $reindex->{from}; # arrayref;
1261 if (ref($ranges) ne 'ARRAY') {
1262 die 'BUG: $reindex->{from} not an ARRAY';
1267 sub index_epoch ($$$) {
1268 my ($self, $sync, $i) = @_;
1270 my $git_dir = git_dir_n($self, $i);
1271 die 'BUG: already reindexing!' if $self->{reindex_pipe};
1272 -d $git_dir or return; # missing epochs are fine
1273 fill_alternates($self, $i);
1274 my $git = PublicInbox::Git->new($git_dir);
1275 if (my $unindex_range = delete $sync->{unindex_range}->{$i}) {
1276 unindex($self, $sync, $git, $unindex_range);
1278 defined(my $range = $sync->{ranges}->[$i]) or return;
1279 if (my $pr = $sync->{-opt}->{-progress}) {
1280 $pr->("$i.git indexing $range\n");
1283 my @cmd = qw(log --raw -r --pretty=tformat:%H.%at.%ct
1284 --no-notes --no-color --no-abbrev --no-renames);
1285 my $fh = $self->{reindex_pipe} = $git->popen(@cmd, $range);
1289 $self->{current_info} = "$i.git $_";
1290 if (/\A($x40)\.([0-9]+)\.([0-9]+)$/o) {
1292 $self->{autime} = $2;
1293 $self->{cotime} = $3;
1294 } elsif (/\A:\d{6} 100644 $x40 ($x40) [AM]\tm$/o) {
1295 reindex_oid($self, $sync, $git, $1);
1296 } elsif (/\A:\d{6} 100644 $x40 ($x40) [AM]\td$/o) {
1297 mark_deleted($self, $sync, $git, $1);
1300 close $fh or die "git log failed: \$?=$?";
1301 delete $self->{reindex_pipe};
1302 update_last_commit($self, $git, $i, $cmt) if defined $cmt;
1305 # public, called by public-inbox-index
1307 my ($self, $opt) = @_;
1309 my $pr = $opt->{-progress};
1311 my $latest = git_dir_latest($self, \$epoch_max);
1312 return unless defined $latest;
1313 $self->idx_init($opt); # acquire lock
1315 D => {}, # "$mid\0$chash" => $oid
1316 unindex_range => {}, # EPOCH => oid_old..oid_new
1317 reindex => $opt->{reindex},
1320 $sync->{ranges} = sync_ranges($self, $sync, $epoch_max);
1321 $sync->{regen} = sync_prepare($self, $sync, $epoch_max);
1323 if ($sync->{regen}) {
1324 # tmp_clone seems to fail if inside a transaction, so
1325 # we rollback here (because we opened {mm} for reading)
1326 # Note: we do NOT rely on DBI transactions for atomicity;
1327 # only for batch performance.
1328 $self->{mm}->{dbh}->rollback;
1329 $self->{mm}->{dbh}->begin_work;
1330 $sync->{mm_tmp} = $self->{mm}->tmp_clone;
1333 # work backwards through history
1334 for (my $i = $epoch_max; $i >= 0; $i--) {
1335 index_epoch($self, $sync, $i);
1338 # unindex is required for leftovers if "deletes" affect messages
1339 # in a previous fetch+index window:
1341 if (my @leftovers = values %{delete $sync->{D}}) {
1342 $git = $self->{-inbox}->git;
1343 for my $oid (@leftovers) {
1344 $self->{current_info} = "leftover $oid";
1345 unindex_oid($self, $git, $oid);
1348 if (my $multi_mid = delete $sync->{multi_mid}) {
1349 $git //= $self->{-inbox}->git;
1350 my $min = $multi_mid->{min};
1351 my $max = $multi_mid->{max};
1352 if ($sync->{reindex}) {
1353 # we may need to create new Message-IDs if mirrors
1354 # were initially indexed with old versions
1355 for (my $i = $max; $i >= $min; $i--) {
1357 $oid = $multi_mid->get_oid($i, $self) or next;
1358 next unless defined $oid;
1359 reindex_oid_m($self, $sync, $git, $oid);
1361 } else { # regen on initial index
1362 for my $num ($min..$max) {
1364 $oid = $multi_mid->get_oid($num, $self) or next;
1365 reindex_oid_m($self, $sync, $git, $oid, $num);
1369 $git->cleanup if $git;
1372 if (my $nr = $sync->{nr}) {
1373 my $pr = $sync->{-opt}->{-progress};
1374 $pr->('all.git '.sprintf($sync->{-regen_fmt}, $nr)) if $pr;
1377 # reindex does not pick up new changes, so we rerun w/o it:
1378 if ($opt->{reindex}) {
1381 delete @again{qw(reindex -skip_lock)};
1382 index_sync($self, \%again);