1 # Copyright (C) 2018-2019 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);
10 use PublicInbox::SearchIdxShard;
11 use PublicInbox::MIME;
13 use PublicInbox::Import;
14 use PublicInbox::MID qw(mids references);
15 use PublicInbox::ContentId qw(content_id content_digest);
16 use PublicInbox::Inbox;
17 use PublicInbox::OverIdx;
18 use PublicInbox::Msgmap;
19 use PublicInbox::Spawn qw(spawn);
20 use PublicInbox::SearchIdx;
22 use File::Temp qw(tempfile);
24 # an estimate of the post-packed size to the raw uncompressed size
25 my $PACKING_FACTOR = 0.4;
27 # SATA storage lags behind what CPUs are capable of, so relying on
28 # nproc(1) can be misleading and having extra Xapian shards is a
29 # waste of FDs and space. It can also lead to excessive IO latency
30 # and slow things down. Users on NVME or other fast storage can
31 # use the NPROC env or switches in our script/public-inbox-* programs
32 # to increase Xapian shards
33 our $NPROC_MAX_DEFAULT = 4;
35 sub nproc_shards ($) {
37 if (ref($creat_opt) eq 'HASH') {
38 if (defined(my $n = $creat_opt->{nproc})) {
45 chomp($n = `nproc 2>/dev/null`);
46 # assume 2 cores if GNU nproc(1) is not available
48 $n = $NPROC_MAX_DEFAULT if $n > $NPROC_MAX_DEFAULT;
51 # subtract for the main process and git-fast-import
56 sub count_shards ($) {
59 my $xpfx = $self->{xpfx};
61 # always load existing shards in case core count changes:
62 # Also, shard count may change while -watch is running
63 # due to "xcpdb --reshard"
65 foreach my $shard (<$xpfx/*>) {
66 -d $shard && $shard =~ m!/[0-9]+\z! or next;
68 Search::Xapian::Database->new($shard)->close;
77 # $creat may be any true value, or 0/undef. A hashref is true,
78 # and $creat->{nproc} may be set to an integer
79 my ($class, $v2ibx, $creat) = @_;
80 $v2ibx = PublicInbox::InboxWritable->new($v2ibx);
81 my $dir = $v2ibx->assert_usable_dir;
85 File::Path::mkpath($dir);
87 die "$dir does not exist\n";
90 $v2ibx->umask_prepare;
92 my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION;
95 im => undef, # PublicInbox::Import
100 over => PublicInbox::OverIdx->new("$xpfx/over.sqlite3", 1),
101 lock_path => "$dir/inbox.lock",
102 # limit each git repo (epoch) to 1GB or so
103 rotate_bytes => int((1024 * 1024 * 1024) / $PACKING_FACTOR),
104 last_commit => [], # git repo -> commit
106 $self->{shards} = count_shards($self) || nproc_shards($creat);
112 my ($self, $parallel, $skip_epoch) = @_;
113 $self->{parallel} = $parallel;
116 git_dir_latest($self, \$epoch_max);
117 if (defined $skip_epoch && $epoch_max == -1) {
118 $epoch_max = $skip_epoch;
120 $self->git_init($epoch_max >= 0 ? $epoch_max : 0);
124 # returns undef on duplicate or spam
125 # mimics Import::add and wraps it for v2
127 my ($self, $mime, $check_cb) = @_;
128 $self->{-inbox}->with_umask(sub {
129 _add($self, $mime, $check_cb)
133 # indexes a message, returns true if checkpointing is needed
134 sub do_idx ($$$$$$$) {
135 my ($self, $msgref, $mime, $len, $num, $oid, $mid0) = @_;
136 $self->{over}->add_overview($mime, $len, $num, $oid, $mid0);
137 my $idx = idx_shard($self, $num % $self->{shards});
138 $idx->index_raw($len, $msgref, $num, $oid, $mid0, $mime);
139 my $n = $self->{transact_bytes} += $len;
140 $n >= (PublicInbox::SearchIdx::BATCH_BYTES * $self->{shards});
144 my ($self, $mime, $check_cb) = @_;
148 $mime = $check_cb->($mime) or return;
151 # All pipes (> $^F) known to Perl 5.6+ have FD_CLOEXEC set,
152 # as does SQLite 3.4.1+ (released in 2007-07-20), and
153 # Xapian 1.3.2+ (released 2015-03-15).
154 # For the most part, we can spawn git-fast-import without
155 # leaking FDs to it...
158 my ($num, $mid0) = v2_num_for($self, $mime);
159 defined $num or return; # duplicate
160 defined $mid0 or die "BUG: $mid0 undefined\n";
161 my $im = $self->importer;
162 my $cmt = $im->add($mime);
163 $cmt = $im->get_mark($cmt);
164 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
166 my ($oid, $len, $msgref) = @{$im->{last_object}};
167 if (do_idx($self, $msgref, $mime, $len, $num, $oid, $mid0)) {
175 my ($self, $mime) = @_;
176 my $mids = mids($mime->header_obj);
178 my $mid = $mids->[0];
179 my $num = $self->{mm}->mid_insert($mid);
180 if (defined $num) { # common case
184 # crap, Message-ID is already known, hope somebody just resent:
185 foreach my $m (@$mids) {
186 # read-only lookup now safe to do after above barrier
187 my $existing = lookup_content($self, $mime, $m);
188 # easy, don't store duplicates
189 # note: do not add more diagnostic info here since
190 # it gets noisy on public-inbox-watch restarts
191 return () if $existing;
194 # AltId may pre-populate article numbers (e.g. X-Mail-Count
195 # or NNTP article number), use that article number if it's
197 my $altid = $self->{-inbox}->{altid};
198 if ($altid && grep(/:file=msgmap\.sqlite3\z/, @$altid)) {
199 my $num = $self->{mm}->num_for($mid);
201 if (defined $num && !$self->{over}->get_art($num)) {
207 warn "<$mid> reused for mismatched content\n";
209 # try the rest of the mids
210 for(my $i = $#$mids; $i >= 1; $i--) {
212 $num = $self->{mm}->mid_insert($m);
214 warn "alternative <$m> for <$mid> found\n";
219 # none of the existing Message-IDs are good, generate a new one:
220 v2_num_for_harder($self, $mime);
223 sub v2_num_for_harder {
224 my ($self, $mime) = @_;
226 my $hdr = $mime->header_obj;
227 my $dig = content_digest($mime);
228 my $mid0 = PublicInbox::Import::digest2mid($dig, $hdr);
229 my $num = $self->{mm}->mid_insert($mid0);
230 unless (defined $num) {
231 # it's hard to spoof the last Received: header
232 my @recvd = $hdr->header_raw('Received');
233 $dig->add("Received: $_") foreach (@recvd);
234 $mid0 = PublicInbox::Import::digest2mid($dig, $hdr);
235 $num = $self->{mm}->mid_insert($mid0);
237 # fall back to a random Message-ID and give up determinism:
238 until (defined($num)) {
240 $mid0 = PublicInbox::Import::digest2mid($dig, $hdr);
241 warn "using random Message-ID <$mid0> as fallback\n";
242 $num = $self->{mm}->mid_insert($mid0);
245 PublicInbox::Import::append_mid($hdr, $mid0);
250 my ($self, $shard_i) = @_;
251 $self->{idx_shards}->[$shard_i];
256 my ($self, $opt) = @_;
257 return if $self->{idx_shards};
258 my $ibx = $self->{-inbox};
260 # do not leak read-only FDs to child processes, we only have these
261 # FDs for duplicate detection so they should not be
262 # frequently activated.
263 delete $ibx->{$_} foreach (qw(git mm search));
265 my $indexlevel = $ibx->{indexlevel};
266 if ($indexlevel && $indexlevel eq 'basic') {
267 $self->{parallel} = 0;
270 if ($self->{parallel}) {
271 pipe(my ($r, $w)) or die "pipe failed: $!";
272 # pipe for barrier notifications doesn't need to be big,
274 fcntl($w, 1031, 4096) if $^O eq 'linux';
275 $self->{bnote} = [ $r, $w ];
279 my $over = $self->{over};
281 $ibx->with_umask(sub {
282 $self->lock_acquire unless ($opt && $opt->{-skip_lock});
285 # xcpdb can change shard count while -watch is idle
286 my $nshards = count_shards($self);
287 if ($nshards && $nshards != $self->{shards}) {
288 $self->{shards} = $nshards;
291 # need to create all shards before initializing msgmap FD
292 my $max = $self->{shards} - 1;
294 # idx_shards must be visible to all forked processes
295 my $idx = $self->{idx_shards} = [];
296 for my $i (0..$max) {
297 push @$idx, PublicInbox::SearchIdxShard->new($self, $i);
300 # Now that all subprocesses are up, we can open the FDs
302 my $mm = $self->{mm} = PublicInbox::Msgmap->new_file(
303 "$self->{-inbox}->{inboxdir}/msgmap.sqlite3", 1);
304 $mm->{dbh}->begin_work;
308 # returns an array mapping [ epoch => latest_commit ]
309 # latest_commit may be undef if nothing was done to that epoch
310 # $replace_map = { $object_id => $strref, ... }
311 sub _replace_oids ($$$) {
312 my ($self, $mime, $replace_map) = @_;
314 my $pfx = "$self->{-inbox}->{inboxdir}/git";
315 my $rewrites = []; # epoch => commit
316 my $max = $self->{epoch_max};
318 unless (defined($max)) {
319 defined(my $latest = git_dir_latest($self, \$max)) or return;
320 $self->{epoch_max} = $max;
323 foreach my $i (0..$max) {
324 my $git_dir = "$pfx/$i.git";
326 my $git = PublicInbox::Git->new($git_dir);
327 my $im = $self->import_init($git, 0, 1);
328 $rewrites->[$i] = $im->replace_oids($mime, $replace_map);
334 sub content_ids ($) {
336 my @cids = ( content_id($mime) );
338 # Email::MIME->as_string doesn't always round-trip, so we may
339 # use a second content_id
340 my $rt = content_id(PublicInbox::MIME->new(\($mime->as_string)));
341 push @cids, $rt if $cids[0] ne $rt;
345 sub content_matches ($$) {
346 my ($cids, $existing) = @_;
347 my $cid = content_id($existing);
349 return 1 if $_ eq $cid
354 # used for removing or replacing (purging)
355 sub rewrite_internal ($$;$$$) {
356 my ($self, $old_mime, $cmt_msg, $new_mime, $sref) = @_;
358 my ($im, $need_reindex, $replace_map);
360 $replace_map = {}; # oid => sref
361 $need_reindex = [] if $new_mime;
363 $im = $self->importer;
365 my $over = $self->{over};
366 my $cids = content_ids($old_mime);
368 my $mids = mids($old_mime->header_obj);
370 # We avoid introducing new blobs into git since the raw content
371 # can be slightly different, so we do not need the user-supplied
372 # message now that we have the mids and content_id
376 foreach my $mid (@$mids) {
377 my %gone; # num => [ smsg, raw ]
379 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
380 my $msg = get_blob($self, $smsg);
381 if (!defined($msg)) {
382 warn "broken smsg for $mid\n";
386 my $cur = PublicInbox::MIME->new($msg);
387 if (content_matches($cids, $cur)) {
388 $smsg->{mime} = $cur;
389 $gone{$smsg->{num}} = [ $smsg, \$orig ];
392 my $n = scalar keys %gone;
395 warn "BUG: multiple articles linked to <$mid>\n",
396 join(',', sort keys %gone), "\n";
398 foreach my $num (keys %gone) {
399 my ($smsg, $orig) = @{$gone{$num}};
400 # $removed should only be set once assuming
401 # no bugs in our deduplication code:
403 my $oid = $smsg->{blob};
405 $replace_map->{$oid} = $sref;
407 ($mark, undef) = $im->remove($orig, $cmt_msg);
410 if ($need_reindex) { # ->replace
411 push @$need_reindex, $smsg;
412 } else { # ->purge or ->remove
413 $self->{mm}->num_delete($num);
415 unindex_oid_remote($self, $oid, $mid);
420 my $cmt = $im->get_mark($mark);
421 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
423 if ($replace_map && scalar keys %$replace_map) {
424 my $rewrites = _replace_oids($self, $new_mime, $replace_map);
425 return { rewrites => $rewrites, need_reindex => $need_reindex };
432 my ($self, $mime, $cmt_msg) = @_;
433 $self->{-inbox}->with_umask(sub {
434 rewrite_internal($self, $mime, $cmt_msg);
438 sub _replace ($$;$$) {
439 my ($self, $old_mime, $new_mime, $sref) = @_;
440 my $rewritten = $self->{-inbox}->with_umask(sub {
441 rewrite_internal($self, $old_mime, undef, $new_mime, $sref);
444 my $rewrites = $rewritten->{rewrites};
445 # ->done is called if there are rewrites since we gc+prune from git
446 $self->idx_init if @$rewrites;
448 for my $i (0..$#$rewrites) {
449 defined(my $cmt = $rewrites->[$i]) or next;
450 $self->{last_commit}->[$i] = $cmt;
457 my ($self, $mime) = @_;
458 my $rewritten = _replace($self, $mime, undef, \'') or return;
459 $rewritten->{rewrites}
462 # returns the git object_id of $fh, does not write the object to FS
463 sub git_hash_raw ($$) {
464 my ($self, $raw) = @_;
465 # grab the expected OID we have to reindex:
466 open my $tmp_fh, '+>', undef or die "failed to open tmp: $!";
467 $tmp_fh->autoflush(1);
468 print $tmp_fh $$raw or die "print \$tmp_fh: $!";
469 sysseek($tmp_fh, 0, 0) or die "seek failed: $!";
472 pipe($r, $w) or die "failed to create pipe: $!";
473 my $rdr = { 0 => fileno($tmp_fh), 1 => fileno($w) };
474 my $git_dir = $self->{-inbox}->git->{git_dir};
475 my $cmd = ['git', "--git-dir=$git_dir", qw(hash-object --stdin)];
476 my $pid = spawn($cmd, undef, $rdr);
479 chomp(my $oid = <$r>);
480 waitpid($pid, 0) == $pid or die "git hash-object did not finish";
481 die "git hash-object failed: $?" if $?;
482 $oid =~ /\A[a-f0-9]{40}\z/ or die "OID not expected: $oid";
486 sub _check_mids_match ($$$) {
487 my ($old_list, $new_list, $hdrs) = @_;
488 my %old_mids = map { $_ => 1 } @$old_list;
489 my %new_mids = map { $_ => 1 } @$new_list;
490 my @old = keys %old_mids;
491 my @new = keys %new_mids;
492 my $err = "$hdrs may not be changed when replacing\n";
493 die $err if scalar(@old) != scalar(@new);
494 delete @new_mids{@old};
495 delete @old_mids{@new};
496 die $err if (scalar(keys %old_mids) || scalar(keys %new_mids));
499 # Changing Message-IDs or References with ->replace isn't supported.
500 # The rules for dealing with messages with multiple or conflicting
501 # Message-IDs are pretty complex and rethreading hasn't been fully
503 sub check_mids_match ($$) {
504 my ($old_mime, $new_mime) = @_;
505 my $old = $old_mime->header_obj;
506 my $new = $new_mime->header_obj;
507 _check_mids_match(mids($old), mids($new), 'Message-ID(s)');
508 _check_mids_match(references($old), references($new),
509 'References/In-Reply-To');
514 my ($self, $old_mime, $new_mime) = @_;
516 check_mids_match($old_mime, $new_mime);
518 # mutt will always add Content-Length:, Status:, Lines: when editing
519 PublicInbox::Import::drop_unwanted_headers($new_mime);
521 my $raw = $new_mime->as_string;
522 my $expect_oid = git_hash_raw($self, \$raw);
523 my $rewritten = _replace($self, $old_mime, $new_mime, \$raw) or return;
524 my $need_reindex = $rewritten->{need_reindex};
526 # just in case we have bugs in deduplication code:
527 my $n = scalar(@$need_reindex);
529 my $list = join(', ', map {
530 "$_->{num}: <$_->{mid}>"
533 W: rewritten $n messages matching content of original message (expected: 1).
534 W: possible bug in public-inbox, NNTP article IDs and Message-IDs follow:
539 # make sure we really got the OID:
540 my ($oid, $type, $len) = $self->{-inbox}->git->check($expect_oid);
541 $oid eq $expect_oid or die "BUG: $expect_oid not found after replace";
543 # don't leak FDs to Xapian:
544 $self->{-inbox}->git->cleanup;
546 # reindex modified messages:
547 for my $smsg (@$need_reindex) {
548 my $num = $smsg->{num};
549 my $mid0 = $smsg->{mid};
550 do_idx($self, \$raw, $new_mime, $len, $num, $oid, $mid0);
552 $rewritten->{rewrites};
555 sub last_epoch_commit ($$;$) {
556 my ($self, $i, $cmt) = @_;
557 my $v = PublicInbox::Search::SCHEMA_VERSION();
558 $self->{mm}->last_commit_xap($v, $i, $cmt);
561 sub set_last_commits ($) {
563 defined(my $epoch_max = $self->{epoch_max}) or return;
564 my $last_commit = $self->{last_commit};
565 foreach my $i (0..$epoch_max) {
566 defined(my $cmt = $last_commit->[$i]) or next;
567 $last_commit->[$i] = undef;
568 last_epoch_commit($self, $i, $cmt);
574 $self->{bnote} or return;
576 my $barrier = { map { $_ => 1 } (0..$n) };
580 my ($self, $barrier) = @_;
581 my $bnote = $self->{bnote} or return;
583 while (scalar keys %$barrier) {
584 defined(my $l = $r->getline) or die "EOF on barrier_wait: $!";
585 $l =~ /\Abarrier (\d+)/ or die "bad line on barrier_wait: $l";
586 delete $barrier->{$1} or die "bad shard[$1] on barrier wait";
591 sub checkpoint ($;$) {
592 my ($self, $wait) = @_;
594 if (my $im = $self->{im}) {
601 my $shards = $self->{idx_shards};
603 my $dbh = $self->{mm}->{dbh};
605 # SQLite msgmap data is second in importance
608 # SQLite overview is third
609 $self->{over}->commit_lazy;
611 # Now deal with Xapian
613 my $barrier = $self->barrier_init(scalar @$shards);
615 # each shard needs to issue a barrier command
616 $_->remote_barrier for @$shards;
618 # wait for each Xapian shard
619 $self->barrier_wait($barrier);
621 $_->remote_commit for @$shards;
624 # last_commit is special, don't commit these until
625 # remote shards are done:
627 set_last_commits($self);
632 $self->{transact_bytes} = 0;
635 # issue a write barrier to ensure all data is visible to other processes
636 # and read-only ops. Order of data importance is: git > SQLite > Xapian
638 sub barrier { checkpoint($_[0], 1) };
643 my $im = delete $self->{im};
644 $im->done if $im; # PublicInbox::Import::done
646 my $mm = delete $self->{mm};
647 $mm->{dbh}->commit if $mm;
648 my $shards = delete $self->{idx_shards};
650 $_->remote_close for @$shards;
652 $self->{over}->disconnect;
653 delete $self->{bnote};
654 $self->{transact_bytes} = 0;
655 $self->lock_release if $shards;
656 $self->{-inbox}->git->cleanup;
659 sub fill_alternates ($$) {
660 my ($self, $epoch) = @_;
662 my $pfx = "$self->{-inbox}->{inboxdir}/git";
663 my $all = "$self->{-inbox}->{inboxdir}/all.git";
666 PublicInbox::Import::init_bare($all);
668 my $alt = "$all/objects/info/alternates";
672 open(my $fh, '<', $alt) or die "open < $alt: $!\n";
673 %alts = map { chomp; $_ => 1 } (<$fh>);
675 foreach my $i (0..$epoch) {
676 my $dir = "../../git/$i.git/objects";
677 push @add, $dir if !$alts{$dir} && -d "$pfx/$i.git";
680 open my $fh, '>>', $alt or die "open >> $alt: $!\n";
681 foreach my $dir (@add) {
682 print $fh "$dir\n" or die "print >> $alt: $!\n";
684 close $fh or die "close $alt: $!\n";
688 my ($self, $epoch) = @_;
689 my $git_dir = "$self->{-inbox}->{inboxdir}/git/$epoch.git";
690 my @cmd = (qw(git init --bare -q), $git_dir);
691 PublicInbox::Import::run_die(\@cmd);
692 @cmd = (qw/git config/, "--file=$git_dir/config",
693 'include.path', '../../all.git/config');
694 PublicInbox::Import::run_die(\@cmd);
695 fill_alternates($self, $epoch);
700 my ($self, $max) = @_;
702 my $pfx = "$self->{-inbox}->{inboxdir}/git";
703 return unless -d $pfx;
705 opendir my $dh, $pfx or die "opendir $pfx: $!\n";
706 while (defined(my $git_dir = readdir($dh))) {
707 $git_dir =~ m!\A([0-9]+)\.git\z! or next;
710 $latest = "$pfx/$git_dir";
718 my $im = $self->{im};
720 if ($im->{bytes_added} < $self->{rotate_bytes}) {
727 my $git_dir = $self->git_init(++$self->{epoch_max});
728 my $git = PublicInbox::Git->new($git_dir);
729 return $self->import_init($git, 0);
734 my $latest = git_dir_latest($self, \$max);
735 if (defined $latest) {
736 my $git = PublicInbox::Git->new($latest);
737 my $packed_bytes = $git->packed_bytes;
738 my $unpacked_bytes = $packed_bytes / $PACKING_FACTOR;
740 if ($unpacked_bytes >= $self->{rotate_bytes}) {
743 $self->{epoch_max} = $max;
744 return $self->import_init($git, $packed_bytes);
747 $self->{epoch_max} = $epoch;
748 $latest = $self->git_init($epoch);
749 $self->import_init(PublicInbox::Git->new($latest), 0);
753 my ($self, $git, $packed_bytes, $tmp) = @_;
754 my $im = PublicInbox::Import->new($git, undef, undef, $self->{-inbox});
755 $im->{bytes_added} = int($packed_bytes / $PACKING_FACTOR);
756 $im->{want_object_info} = 1;
757 $im->{lock_path} = undef;
758 $im->{path_type} = 'v2';
759 $self->{im} = $im unless $tmp;
765 my ($mid, $cur, $new) = @_;
767 my ($ah, $an) = tempfile('email-cur-XXXXXXXX', TMPDIR => 1);
768 print $ah $cur->as_string or die "print: $!";
769 close $ah or die "close: $!";
770 my ($bh, $bn) = tempfile('email-new-XXXXXXXX', TMPDIR => 1);
771 PublicInbox::Import::drop_unwanted_headers($new);
772 print $bh $new->as_string or die "print: $!";
773 close $bh or die "close: $!";
774 my $cmd = [ qw(diff -u), $an, $bn ];
775 print STDERR "# MID conflict <$mid>\n";
776 my $pid = spawn($cmd, undef, { 1 => 2 });
777 defined $pid or die "diff failed to spawn $!";
778 waitpid($pid, 0) == $pid or die "diff did not finish";
783 my ($self, $smsg) = @_;
784 if (my $im = $self->{im}) {
785 my $msg = $im->cat_blob($smsg->{blob});
788 # older message, should be in alternates
789 my $ibx = $self->{-inbox};
790 $ibx->msg_by_smsg($smsg);
793 sub lookup_content ($$$) {
794 my ($self, $mime, $mid) = @_;
795 my $over = $self->{over};
796 my $cids = content_ids($mime);
798 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
799 my $msg = get_blob($self, $smsg);
800 if (!defined($msg)) {
801 warn "broken smsg for $mid\n";
804 my $cur = PublicInbox::MIME->new($msg);
805 if (content_matches($cids, $cur)) {
806 $smsg->{mime} = $cur;
811 # XXX DEBUG_DIFF is experimental and may be removed
812 diff($mid, $cur, $mime) if $ENV{DEBUG_DIFF};
819 my $fh = delete $self->{reindex_pipe};
821 if (my $shards = $self->{idx_shards}) {
822 $_->atfork_child foreach @$shards;
824 if (my $im = $self->{im}) {
827 die "unexpected mm" if $self->{mm};
828 close $self->{bnote}->[0] or die "close bnote[0]: $!\n";
832 sub mark_deleted ($$$$) {
833 my ($self, $sync, $git, $oid) = @_;
834 my $msgref = $git->cat_file($oid);
835 my $mime = PublicInbox::MIME->new($$msgref);
836 my $mids = mids($mime->header_obj);
837 my $cid = content_id($mime);
838 foreach my $mid (@$mids) {
839 $sync->{D}->{"$mid\0$cid"} = $oid;
843 sub reindex_checkpoint ($$$) {
844 my ($self, $sync, $git) = @_;
847 $sync->{mm_tmp}->atfork_prepare;
848 $self->done; # release lock
850 if (my $pr = $sync->{-opt}->{-progress}) {
851 my ($bn) = (split('/', $git->{git_dir}))[-1];
852 $pr->("$bn ".sprintf($sync->{-regen_fmt}, $sync->{nr}));
855 # allow -watch or -mda to write...
856 $self->idx_init; # reacquire lock
857 $sync->{mm_tmp}->atfork_parent;
860 # only for a few odd messages with multiple Message-IDs
861 sub reindex_oid_m ($$$$;$) {
862 my ($self, $sync, $git, $oid, $regen_num) = @_;
863 $self->{current_info} = "multi_mid $oid";
864 my ($num, $mid0, $len);
865 my $msgref = $git->cat_file($oid, \$len);
866 my $mime = PublicInbox::MIME->new($$msgref);
867 my $mids = mids($mime->header_obj);
868 my $cid = content_id($mime);
869 die "BUG: reindex_oid_m called for <=1 mids" if scalar(@$mids) <= 1;
871 for my $mid (reverse @$mids) {
872 delete($sync->{D}->{"$mid\0$cid"}) and
873 die "BUG: reindex_oid should handle <$mid> delete";
875 my $over = $self->{over};
876 for my $mid (reverse @$mids) {
877 ($num, $mid0) = $over->num_mid0_for_oid($oid, $mid);
878 next unless defined $num;
879 if (defined($regen_num) && $regen_num != $num) {
880 die "BUG: regen(#$regen_num) != over(#$num)";
883 unless (defined($num)) {
884 for my $mid (reverse @$mids) {
885 # is this a number we got before?
886 my $n = $sync->{mm_tmp}->num_for($mid);
887 next unless defined $n;
888 next if defined($regen_num) && $regen_num != $n;
889 ($num, $mid0) = ($n, $mid);
894 $sync->{mm_tmp}->num_delete($num);
895 } elsif (defined $regen_num) {
897 for my $mid (reverse @$mids) {
898 $self->{mm}->mid_set($num, $mid) == 1 or next;
902 unless (defined $mid0) {
903 warn "E: cannot regen #$num\n";
906 } else { # fixup bugs in old mirrors on reindex
907 for my $mid (reverse @$mids) {
908 $num = $self->{mm}->mid_insert($mid);
909 next unless defined $num;
914 if ($sync->{reindex}) {
915 warn "reindex added #$num <$mid0>\n";
918 warn "E: cannot find article #\n";
923 if (do_idx($self, $msgref, $mime, $len, $num, $oid, $mid0)) {
924 reindex_checkpoint($self, $sync, $git);
928 sub check_unindexed ($$$) {
929 my ($self, $num, $mid0) = @_;
930 my $unindexed = $self->{unindexed} // {};
931 my $n = delete($unindexed->{$mid0});
932 defined $n or return;
934 die "BUG: unindexed $n != $num <$mid0>\n";
936 $self->{mm}->mid_set($num, $mid0);
940 # reuse Msgmap to store num => oid mapping (rather than num => mid)
941 sub multi_mid_q_new () {
942 my ($fh, $fn) = tempfile('multi_mid-XXXXXXX', EXLOCK => 0, TMPDIR => 1);
943 my $multi_mid = PublicInbox::Msgmap->new_file($fn, 1);
944 $multi_mid->{dbh}->do('PRAGMA synchronous = OFF');
945 # for Msgmap->DESTROY:
946 $multi_mid->{tmp_name} = $fn;
947 $multi_mid->{pid} = $$;
948 close $fh or die "failed to close $fn: $!";
952 sub multi_mid_q_push ($$) {
953 my ($sync, $oid) = @_;
954 my $multi_mid = $sync->{multi_mid} //= multi_mid_q_new();
955 if ($sync->{reindex}) { # no regen on reindex
956 $multi_mid->mid_insert($oid);
958 my $num = $sync->{regen}--;
959 die "BUG: ran out of article numbers" if $num <= 0;
960 $multi_mid->mid_set($num, $oid);
964 sub reindex_oid ($$$$) {
965 my ($self, $sync, $git, $oid) = @_;
966 my ($num, $mid0, $len);
967 my $msgref = $git->cat_file($oid, \$len);
968 return if $len == 0; # purged
969 my $mime = PublicInbox::MIME->new($$msgref);
970 my $mids = mids($mime->header_obj);
971 my $cid = content_id($mime);
973 if (scalar(@$mids) == 0) {
974 warn "E: $oid has no Message-ID, skipping\n";
976 } elsif (scalar(@$mids) == 1) {
977 my $mid = $mids->[0];
979 # was the file previously marked as deleted?, skip if so
980 if (delete($sync->{D}->{"$mid\0$cid"})) {
981 if (!$sync->{reindex}) {
982 $num = $sync->{regen}--;
983 $self->{mm}->num_highwater($num);
988 # is this a number we got before?
989 $num = $sync->{mm_tmp}->num_for($mid);
992 check_unindexed($self, $num, $mid0);
994 $num = $sync->{regen}--;
995 die "BUG: ran out of article numbers" if $num <= 0;
996 if ($self->{mm}->mid_set($num, $mid) != 1) {
997 warn "E: unable to assign $num => <$mid>\n";
1002 } else { # multiple MIDs are a weird case:
1005 $del += delete($sync->{D}->{"$_\0$cid"}) // 0;
1008 unindex_oid_remote($self, $oid, $_) for @$mids;
1009 # do not delete from {mm_tmp}, since another
1010 # single-MID message may use it.
1011 } else { # handle them at the end:
1012 multi_mid_q_push($sync, $oid);
1016 $sync->{mm_tmp}->mid_delete($mid0) or
1017 die "failed to delete <$mid0> for article #$num\n";
1019 if (do_idx($self, $msgref, $mime, $len, $num, $oid, $mid0)) {
1020 reindex_checkpoint($self, $sync, $git);
1024 # only update last_commit for $i on reindex iff newer than current
1025 sub update_last_commit ($$$$) {
1026 my ($self, $git, $i, $cmt) = @_;
1027 my $last = last_epoch_commit($self, $i);
1028 if (defined $last && is_ancestor($git, $last, $cmt)) {
1029 my @cmd = (qw(rev-list --count), "$last..$cmt");
1030 chomp(my $n = $git->qx(@cmd));
1031 return if $n ne '' && $n == 0;
1033 last_epoch_commit($self, $i, $cmt);
1036 sub git_dir_n ($$) { "$_[0]->{-inbox}->{inboxdir}/git/$_[1].git" }
1038 sub last_commits ($$) {
1039 my ($self, $epoch_max) = @_;
1041 for (my $i = $epoch_max; $i >= 0; $i--) {
1042 $heads->[$i] = last_epoch_commit($self, $i);
1047 *is_ancestor = *PublicInbox::SearchIdx::is_ancestor;
1049 # returns a revision range for git-log(1)
1050 sub log_range ($$$$$) {
1051 my ($self, $sync, $git, $i, $tip) = @_;
1052 my $opt = $sync->{-opt};
1053 my $pr = $opt->{-progress} if (($opt->{verbose} || 0) > 1);
1054 my $cur = $sync->{ranges}->[$i] or do {
1055 $pr->("$i.git indexing all of $tip") if $pr;
1056 return $tip; # all of it
1059 # fast equality check to avoid (v)fork+execve overhead
1061 $sync->{ranges}->[$i] = undef;
1065 my $range = "$cur..$tip";
1066 $pr->("$i.git checking contiguity... ") if $pr;
1067 if (is_ancestor($git, $cur, $tip)) { # common case
1068 $pr->("OK\n") if $pr;
1069 my $n = $git->qx(qw(rev-list --count), $range);
1072 $sync->{ranges}->[$i] = undef;
1073 $pr->("$i.git has nothing new\n") if $pr;
1074 return; # nothing to do
1076 $pr->("$i.git has $n changes since $cur\n") if $pr;
1078 $pr->("FAIL\n") if $pr;
1080 discontiguous range: $range
1081 Rewritten history? (in $git->{git_dir})
1083 chomp(my $base = $git->qx('merge-base', $tip, $cur));
1085 $range = "$base..$tip";
1086 warn "found merge-base: $base\n"
1089 warn "discarding history at $cur\n";
1092 reindexing $git->{git_dir} starting at
1095 $sync->{unindex_range}->{$i} = "$base..$cur";
1100 sub sync_prepare ($$$) {
1101 my ($self, $sync, $epoch_max) = @_;
1102 my $pr = $sync->{-opt}->{-progress};
1104 my $head = $self->{-inbox}->{ref_head} || 'refs/heads/master';
1106 # reindex stops at the current heads and we later rerun index_sync
1108 my $reindex_heads = last_commits($self, $epoch_max) if $sync->{reindex};
1110 for (my $i = $epoch_max; $i >= 0; $i--) {
1111 die 'BUG: already indexing!' if $self->{reindex_pipe};
1112 my $git_dir = git_dir_n($self, $i);
1113 -d $git_dir or next; # missing epochs are fine
1114 my $git = PublicInbox::Git->new($git_dir);
1115 if ($reindex_heads) {
1116 $head = $reindex_heads->[$i] or next;
1118 chomp(my $tip = $git->qx(qw(rev-parse -q --verify), $head));
1120 next if $?; # new repo
1121 my $range = log_range($self, $sync, $git, $i, $tip) or next;
1122 $sync->{ranges}->[$i] = $range;
1124 # can't use 'rev-list --count' if we use --diff-filter
1125 $pr->("$i.git counting $range ... ") if $pr;
1127 my $fh = $git->popen(qw(log --pretty=tformat:%H
1128 --no-notes --no-color --no-renames
1129 --diff-filter=AM), $range, '--', 'm');
1131 close $fh or die "git log failed: \$?=$?";
1132 $pr->("$n\n") if $pr;
1136 return 0 if (!$regen_max && !keys(%{$self->{unindex_range}}));
1138 # reindex should NOT see new commits anymore, if we do,
1139 # it's a problem and we need to notice it via die()
1140 my $pad = length($regen_max) + 1;
1141 $sync->{-regen_fmt} = "% ${pad}u/$regen_max\n";
1143 return -1 if $sync->{reindex};
1144 $regen_max + $self->{mm}->num_highwater() || 0;
1147 sub unindex_oid_remote ($$$) {
1148 my ($self, $oid, $mid) = @_;
1149 $_->remote_remove($oid, $mid) foreach @{$self->{idx_shards}};
1150 $self->{over}->remove_oid($oid, $mid);
1153 sub unindex_oid ($$$;$) {
1154 my ($self, $git, $oid, $unindexed) = @_;
1155 my $mm = $self->{mm};
1156 my $msgref = $git->cat_file($oid);
1157 my $mime = PublicInbox::MIME->new($msgref);
1158 my $mids = mids($mime->header_obj);
1159 $mime = $msgref = undef;
1160 my $over = $self->{over};
1161 foreach my $mid (@$mids) {
1164 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
1165 $gone{$smsg->{num}} = 1 if $oid eq $smsg->{blob};
1168 my $n = scalar keys %gone;
1171 warn "BUG: multiple articles linked to $oid\n",
1172 join(',',sort keys %gone), "\n";
1174 foreach my $num (keys %gone) {
1176 my $mid0 = $mm->mid_for($num);
1177 $unindexed->{$mid0} = $num;
1179 $mm->num_delete($num);
1181 unindex_oid_remote($self, $oid, $mid);
1185 my $x40 = qr/[a-f0-9]{40}/;
1186 sub unindex ($$$$) {
1187 my ($self, $sync, $git, $unindex_range) = @_;
1188 my $unindexed = $self->{unindexed} ||= {}; # $mid0 => $num
1189 my $before = scalar keys %$unindexed;
1190 # order does not matter, here:
1191 my @cmd = qw(log --raw -r
1192 --no-notes --no-color --no-abbrev --no-renames);
1193 my $fh = $self->{reindex_pipe} = $git->popen(@cmd, $unindex_range);
1195 /\A:\d{6} 100644 $x40 ($x40) [AM]\tm$/o or next;
1196 unindex_oid($self, $git, $1, $unindexed);
1198 delete $self->{reindex_pipe};
1199 close $fh or die "git log failed: \$?=$?";
1201 return unless $sync->{-opt}->{prune};
1202 my $after = scalar keys %$unindexed;
1203 return if $before == $after;
1205 # ensure any blob can not longer be accessed via dumb HTTP
1206 PublicInbox::Import::run_die(['git', "--git-dir=$git->{git_dir}",
1207 qw(-c gc.reflogExpire=now gc --prune=all --quiet)]);
1210 sub sync_ranges ($$$) {
1211 my ($self, $sync, $epoch_max) = @_;
1212 my $reindex = $sync->{reindex};
1214 return last_commits($self, $epoch_max) unless $reindex;
1215 return [] if ref($reindex) ne 'HASH';
1217 my $ranges = $reindex->{from}; # arrayref;
1218 if (ref($ranges) ne 'ARRAY') {
1219 die 'BUG: $reindex->{from} not an ARRAY';
1224 sub index_epoch ($$$) {
1225 my ($self, $sync, $i) = @_;
1227 my $git_dir = git_dir_n($self, $i);
1228 die 'BUG: already reindexing!' if $self->{reindex_pipe};
1229 -d $git_dir or return; # missing epochs are fine
1230 fill_alternates($self, $i);
1231 my $git = PublicInbox::Git->new($git_dir);
1232 if (my $unindex_range = delete $sync->{unindex_range}->{$i}) {
1233 unindex($self, $sync, $git, $unindex_range);
1235 defined(my $range = $sync->{ranges}->[$i]) or return;
1236 if (my $pr = $sync->{-opt}->{-progress}) {
1237 $pr->("$i.git indexing $range\n");
1240 my @cmd = qw(log --raw -r --pretty=tformat:%H
1241 --no-notes --no-color --no-abbrev --no-renames);
1242 my $fh = $self->{reindex_pipe} = $git->popen(@cmd, $range);
1246 $self->{current_info} = "$i.git $_";
1247 if (/\A$x40$/o && !defined($cmt)) {
1249 } elsif (/\A:\d{6} 100644 $x40 ($x40) [AM]\tm$/o) {
1250 reindex_oid($self, $sync, $git, $1);
1251 } elsif (/\A:\d{6} 100644 $x40 ($x40) [AM]\td$/o) {
1252 mark_deleted($self, $sync, $git, $1);
1255 close $fh or die "git log failed: \$?=$?";
1256 delete $self->{reindex_pipe};
1257 update_last_commit($self, $git, $i, $cmt) if defined $cmt;
1260 # public, called by public-inbox-index
1262 my ($self, $opt) = @_;
1264 my $pr = $opt->{-progress};
1266 my $latest = git_dir_latest($self, \$epoch_max);
1267 return unless defined $latest;
1268 $self->idx_init($opt); # acquire lock
1270 D => {}, # "$mid\0$cid" => $oid
1271 unindex_range => {}, # EPOCH => oid_old..oid_new
1272 reindex => $opt->{reindex},
1275 $sync->{ranges} = sync_ranges($self, $sync, $epoch_max);
1276 $sync->{regen} = sync_prepare($self, $sync, $epoch_max);
1278 if ($sync->{regen}) {
1279 # tmp_clone seems to fail if inside a transaction, so
1280 # we rollback here (because we opened {mm} for reading)
1281 # Note: we do NOT rely on DBI transactions for atomicity;
1282 # only for batch performance.
1283 $self->{mm}->{dbh}->rollback;
1284 $self->{mm}->{dbh}->begin_work;
1285 $sync->{mm_tmp} = $self->{mm}->tmp_clone;
1288 # work backwards through history
1289 for (my $i = $epoch_max; $i >= 0; $i--) {
1290 index_epoch($self, $sync, $i);
1293 # unindex is required for leftovers if "deletes" affect messages
1294 # in a previous fetch+index window:
1296 if (my @leftovers = values %{delete $sync->{D}}) {
1297 $git = $self->{-inbox}->git;
1298 for my $oid (@leftovers) {
1299 $self->{current_info} = "leftover $oid";
1300 unindex_oid($self, $git, $oid);
1303 if (my $multi_mid = delete $sync->{multi_mid}) {
1304 $git //= $self->{-inbox}->git;
1305 my ($min, $max) = $multi_mid->minmax;
1306 if ($sync->{reindex}) {
1307 # we may need to create new Message-IDs if mirrors
1308 # were initially indexed with old versions
1309 for (my $i = $max; $i >= $min; $i--) {
1310 my $oid = $multi_mid->mid_for($i);
1311 next unless defined $oid;
1312 reindex_oid_m($self, $sync, $git, $oid);
1314 } else { # regen on initial index
1315 for my $num ($min..$max) {
1316 my $oid = $multi_mid->mid_for($num);
1317 next unless defined $oid;
1318 reindex_oid_m($self, $sync, $git, $oid, $num);
1322 $git->cleanup if $git;
1325 if (my $nr = $sync->{nr}) {
1326 my $pr = $sync->{-opt}->{-progress};
1327 $pr->('all.git '.sprintf($sync->{-regen_fmt}, $nr)) if $pr;
1330 # reindex does not pick up new changes, so we rerun w/o it:
1331 if ($opt->{reindex}) {
1334 delete @again{qw(reindex -skip_lock)};
1335 index_sync($self, \%again);