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;
11 use PublicInbox::IdxStack;
14 use PublicInbox::Import;
15 use PublicInbox::MID qw(mids references);
16 use PublicInbox::ContentHash qw(content_hash content_digest);
17 use PublicInbox::InboxWritable;
18 use PublicInbox::OverIdx;
19 use PublicInbox::Msgmap;
20 use PublicInbox::Spawn qw(spawn popen_rd);
21 use PublicInbox::SearchIdx;
22 use IO::Handle; # ->autoflush
23 use File::Temp qw(tempfile);
25 my $OID = qr/[a-f0-9]{40,}/;
26 # an estimate of the post-packed size to the raw uncompressed size
27 my $PACKING_FACTOR = 0.4;
29 # SATA storage lags behind what CPUs are capable of, so relying on
30 # nproc(1) can be misleading and having extra Xapian shards is a
31 # waste of FDs and space. It can also lead to excessive IO latency
32 # and slow things down. Users on NVME or other fast storage can
33 # use the NPROC env or switches in our script/public-inbox-* programs
34 # to increase Xapian shards
35 our $NPROC_MAX_DEFAULT = 4;
38 for my $nproc (qw(nproc gnproc)) { # GNU coreutils nproc
39 `$nproc 2>/dev/null` =~ /^(\d+)$/ and return $1;
42 # getconf(1) is POSIX, but *NPROCESSORS* vars are not
43 for (qw(_NPROCESSORS_ONLN NPROCESSORS_ONLN)) {
44 `getconf $_ 2>/dev/null` =~ /^(\d+)$/ and return $1;
47 # should we bother with `sysctl hw.ncpu`? Those only give
48 # us total processor count, not online processor count.
52 sub nproc_shards ($) {
54 my $n = $creat_opt->{nproc} if ref($creat_opt) eq 'HASH';
57 # assume 2 cores if not detectable or zero
58 state $NPROC_DETECTED = detect_nproc() || 2;
60 $n = $NPROC_MAX_DEFAULT if $n > $NPROC_MAX_DEFAULT;
63 # subtract for the main process and git-fast-import
68 sub count_shards ($) {
71 my $xpfx = $self->{xpfx};
73 # always load existing shards in case core count changes:
74 # Also, shard count may change while -watch is running
75 # due to "xcpdb --reshard"
78 foreach my $shard (<$xpfx/*>) {
79 -d $shard && $shard =~ m!/[0-9]+\z! or next;
80 $XapianDatabase //= do {
81 require PublicInbox::Search;
82 PublicInbox::Search::load_xapian();
83 $PublicInbox::Search::X{Database};
86 $XapianDatabase->new($shard)->close;
95 # $creat may be any true value, or 0/undef. A hashref is true,
96 # and $creat->{nproc} may be set to an integer
97 my ($class, $v2ibx, $creat) = @_;
98 $v2ibx = PublicInbox::InboxWritable->new($v2ibx);
99 my $dir = $v2ibx->assert_usable_dir;
103 File::Path::mkpath($dir);
105 die "$dir does not exist\n";
108 $v2ibx->umask_prepare;
110 my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION;
113 im => undef, # PublicInbox::Import
119 over => PublicInbox::OverIdx->new("$xpfx/over.sqlite3"),
120 lock_path => "$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 epoch -> commit
125 $self->{over}->{-no_sync} = 1 if $v2ibx->{-no_sync};
126 $self->{shards} = count_shards($self) || nproc_shards($creat);
127 $self->{index_max_size} = $v2ibx->{index_max_size};
133 my ($self, $shards, $skip_epoch, $skip_artnum) = @_;
134 if (defined $shards) {
135 $self->{parallel} = 0 if $shards == 0;
136 $self->{shards} = $shards if $shards > 0;
139 $self->{mm}->skip_artnum($skip_artnum) if defined $skip_artnum;
141 git_dir_latest($self, \$epoch_max);
142 if (defined $skip_epoch && $epoch_max == -1) {
143 $epoch_max = $skip_epoch;
145 $self->git_init($epoch_max >= 0 ? $epoch_max : 0);
149 # returns undef on duplicate or spam
150 # mimics Import::add and wraps it for v2
152 my ($self, $eml, $check_cb) = @_;
153 $self->{ibx}->with_umask(\&_add, $self, $eml, $check_cb);
156 # indexes a message, returns true if checkpointing is needed
158 my ($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->{ibx}) 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->{ibx}->{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];
279 sub _idx_init { # with_umask callback
280 my ($self, $opt) = @_;
281 $self->lock_acquire unless $opt && $opt->{-skip_lock};
282 $self->{over}->create;
284 # xcpdb can change shard count while -watch is idle
285 my $nshards = count_shards($self);
286 $self->{shards} = $nshards if $nshards && $nshards != $self->{shards};
288 # need to create all shards before initializing msgmap FD
289 # idx_shards must be visible to all forked processes
290 my $max = $self->{shards} - 1;
291 my $idx = $self->{idx_shards} = [];
292 push @$idx, PublicInbox::SearchIdxShard->new($self, $_) for (0..$max);
294 # Now that all subprocesses are up, we can open the FDs
296 my $mm = $self->{mm} = PublicInbox::Msgmap->new_file(
297 "$self->{ibx}->{inboxdir}/msgmap.sqlite3",
298 $self->{ibx}->{-no_sync} ? 2 : 1);
299 $mm->{dbh}->begin_work;
304 my ($self, $opt) = @_;
305 return if $self->{idx_shards};
306 my $ibx = $self->{ibx};
308 # do not leak read-only FDs to child processes, we only have these
309 # FDs for duplicate detection so they should not be
310 # frequently activated.
311 # delete @$ibx{qw(git mm search)};
312 delete $ibx->{$_} foreach (qw(git mm search));
314 $self->{parallel} = 0 if ($ibx->{indexlevel}//'') eq 'basic';
315 if ($self->{parallel}) {
316 pipe(my ($r, $w)) or die "pipe failed: $!";
317 # pipe for barrier notifications doesn't need to be big,
319 fcntl($w, 1031, 4096) if $^O eq 'linux';
320 $self->{bnote} = [ $r, $w ];
325 $ibx->with_umask(\&_idx_init, $self, $opt);
328 # returns an array mapping [ epoch => latest_commit ]
329 # latest_commit may be undef if nothing was done to that epoch
330 # $replace_map = { $object_id => $strref, ... }
331 sub _replace_oids ($$$) {
332 my ($self, $mime, $replace_map) = @_;
334 my $pfx = "$self->{ibx}->{inboxdir}/git";
335 my $rewrites = []; # epoch => commit
336 my $max = $self->{epoch_max};
338 unless (defined($max)) {
339 defined(my $latest = git_dir_latest($self, \$max)) or return;
340 $self->{epoch_max} = $max;
343 foreach my $i (0..$max) {
344 my $git_dir = "$pfx/$i.git";
346 my $git = PublicInbox::Git->new($git_dir);
347 my $im = $self->import_init($git, 0, 1);
348 $rewrites->[$i] = $im->replace_oids($mime, $replace_map);
354 sub content_hashes ($) {
356 my @chashes = ( content_hash($mime) );
358 # We still support Email::MIME, here, and
359 # Email::MIME->as_string doesn't always round-trip, so we may
360 # use a second content_hash
361 my $rt = content_hash(PublicInbox::Eml->new(\($mime->as_string)));
362 push @chashes, $rt if $chashes[0] ne $rt;
366 sub content_matches ($$) {
367 my ($chashes, $existing) = @_;
368 my $chash = content_hash($existing);
369 foreach (@$chashes) {
370 return 1 if $_ eq $chash
375 # used for removing or replacing (purging)
376 sub rewrite_internal ($$;$$$) {
377 my ($self, $old_eml, $cmt_msg, $new_eml, $sref) = @_;
379 my ($im, $need_reindex, $replace_map);
381 $replace_map = {}; # oid => sref
382 $need_reindex = [] if $new_eml;
384 $im = $self->importer;
386 my $over = $self->{over};
387 my $chashes = content_hashes($old_eml);
389 my $mids = mids($old_eml->header_obj);
391 # We avoid introducing new blobs into git since the raw content
392 # can be slightly different, so we do not need the user-supplied
393 # message now that we have the mids and content_hash
397 foreach my $mid (@$mids) {
398 my %gone; # num => [ smsg, $mime, raw ]
400 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
401 my $msg = get_blob($self, $smsg);
402 if (!defined($msg)) {
403 warn "broken smsg for $mid\n";
407 my $cur = PublicInbox::Eml->new($msg);
408 if (content_matches($chashes, $cur)) {
409 $gone{$smsg->{num}} = [ $smsg, $cur, \$orig ];
412 my $n = scalar keys %gone;
415 warn "BUG: multiple articles linked to <$mid>\n",
416 join(',', sort keys %gone), "\n";
418 foreach my $num (keys %gone) {
419 my ($smsg, $mime, $orig) = @{$gone{$num}};
420 # $removed should only be set once assuming
421 # no bugs in our deduplication code:
422 $removed = [ undef, $mime, $smsg ];
423 my $oid = $smsg->{blob};
425 $replace_map->{$oid} = $sref;
427 ($mark, undef) = $im->remove($orig, $cmt_msg);
428 $removed->[0] = $mark;
431 if ($need_reindex) { # ->replace
432 push @$need_reindex, $smsg;
433 } else { # ->purge or ->remove
434 $self->{mm}->num_delete($num);
436 unindex_oid_remote($self, $oid, $mid);
441 my $cmt = $im->get_mark($mark);
442 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
444 if ($replace_map && scalar keys %$replace_map) {
445 my $rewrites = _replace_oids($self, $new_eml, $replace_map);
446 return { rewrites => $rewrites, need_reindex => $need_reindex };
448 defined($mark) ? $removed : undef;
451 # public (see PublicInbox::Import->remove), but note the 3rd element
452 # (retval[2]) is not part of the stable API shared with Import->remove
454 my ($self, $eml, $cmt_msg) = @_;
455 my $r = $self->{ibx}->with_umask(\&rewrite_internal,
456 $self, $eml, $cmt_msg);
457 defined($r) && defined($r->[0]) ? @$r: undef;
460 sub _replace ($$;$$) {
461 my ($self, $old_eml, $new_eml, $sref) = @_;
462 my $arg = [ $self, $old_eml, undef, $new_eml, $sref ];
463 my $rewritten = $self->{ibx}->with_umask(\&rewrite_internal,
464 $self, $old_eml, undef, $new_eml, $sref) or return;
466 my $rewrites = $rewritten->{rewrites};
467 # ->done is called if there are rewrites since we gc+prune from git
468 $self->idx_init if @$rewrites;
470 for my $i (0..$#$rewrites) {
471 defined(my $cmt = $rewrites->[$i]) or next;
472 $self->{last_commit}->[$i] = $cmt;
479 my ($self, $mime) = @_;
480 my $rewritten = _replace($self, $mime, undef, \'') or return;
481 $rewritten->{rewrites}
484 # returns the git object_id of $fh, does not write the object to FS
485 sub git_hash_raw ($$) {
486 my ($self, $raw) = @_;
487 # grab the expected OID we have to reindex:
488 pipe(my($in, $w)) or die "pipe: $!";
489 my $git_dir = $self->{ibx}->git->{git_dir};
490 my $cmd = ['git', "--git-dir=$git_dir", qw(hash-object --stdin)];
491 my $r = popen_rd($cmd, undef, { 0 => $in });
492 print $w $$raw or die "print \$w: $!";
493 close $w or die "close \$w: $!";
495 chomp(my $oid = <$r>);
496 close $r or die "git hash-object failed: $?";
497 $oid =~ /\A$OID\z/ or die "OID not expected: $oid";
501 sub _check_mids_match ($$$) {
502 my ($old_list, $new_list, $hdrs) = @_;
503 my %old_mids = map { $_ => 1 } @$old_list;
504 my %new_mids = map { $_ => 1 } @$new_list;
505 my @old = keys %old_mids;
506 my @new = keys %new_mids;
507 my $err = "$hdrs may not be changed when replacing\n";
508 die $err if scalar(@old) != scalar(@new);
509 delete @new_mids{@old};
510 delete @old_mids{@new};
511 die $err if (scalar(keys %old_mids) || scalar(keys %new_mids));
514 # Changing Message-IDs or References with ->replace isn't supported.
515 # The rules for dealing with messages with multiple or conflicting
516 # Message-IDs are pretty complex and rethreading hasn't been fully
518 sub check_mids_match ($$) {
519 my ($old_mime, $new_mime) = @_;
520 my $old = $old_mime->header_obj;
521 my $new = $new_mime->header_obj;
522 _check_mids_match(mids($old), mids($new), 'Message-ID(s)');
523 _check_mids_match(references($old), references($new),
524 'References/In-Reply-To');
529 my ($self, $old_mime, $new_mime) = @_;
531 check_mids_match($old_mime, $new_mime);
533 # mutt will always add Content-Length:, Status:, Lines: when editing
534 PublicInbox::Import::drop_unwanted_headers($new_mime);
536 my $raw = $new_mime->as_string;
537 my $expect_oid = git_hash_raw($self, \$raw);
538 my $rewritten = _replace($self, $old_mime, $new_mime, \$raw) or return;
539 my $need_reindex = $rewritten->{need_reindex};
541 # just in case we have bugs in deduplication code:
542 my $n = scalar(@$need_reindex);
544 my $list = join(', ', map {
545 "$_->{num}: <$_->{mid}>"
548 W: rewritten $n messages matching content of original message (expected: 1).
549 W: possible bug in public-inbox, NNTP article IDs and Message-IDs follow:
554 # make sure we really got the OID:
555 my ($blob, $type, $bytes) = $self->{ibx}->git->check($expect_oid);
556 $blob eq $expect_oid or die "BUG: $expect_oid not found after replace";
558 # don't leak FDs to Xapian:
559 $self->{ibx}->git->cleanup;
561 # reindex modified messages:
562 for my $smsg (@$need_reindex) {
563 my $new_smsg = bless {
568 }, 'PublicInbox::Smsg';
569 my $sync = { autime => $smsg->{ds}, cotime => $smsg->{ts} };
570 $new_smsg->populate($new_mime, $sync);
571 do_idx($self, \$raw, $new_mime, $new_smsg);
573 $rewritten->{rewrites};
576 sub last_epoch_commit ($$;$) {
577 my ($self, $i, $cmt) = @_;
578 my $v = PublicInbox::Search::SCHEMA_VERSION();
579 $self->{mm}->last_commit_xap($v, $i, $cmt);
582 sub set_last_commits ($) {
584 defined(my $epoch_max = $self->{epoch_max}) or return;
585 my $last_commit = $self->{last_commit};
586 foreach my $i (0..$epoch_max) {
587 defined(my $cmt = $last_commit->[$i]) or next;
588 $last_commit->[$i] = undef;
589 last_epoch_commit($self, $i, $cmt);
595 $self->{bnote} or return;
597 my $barrier = { map { $_ => 1 } (0..$n) };
601 my ($self, $barrier) = @_;
602 my $bnote = $self->{bnote} or return;
604 while (scalar keys %$barrier) {
605 defined(my $l = readline($r)) or die "EOF on barrier_wait: $!";
606 $l =~ /\Abarrier (\d+)/ or die "bad line on barrier_wait: $l";
607 delete $barrier->{$1} or die "bad shard[$1] on barrier wait";
612 sub checkpoint ($;$) {
613 my ($self, $wait) = @_;
615 if (my $im = $self->{im}) {
622 my $shards = $self->{idx_shards};
624 my $dbh = $self->{mm}->{dbh};
626 # SQLite msgmap data is second in importance
629 # SQLite overview is third
630 $self->{over}->commit_lazy;
632 # Now deal with Xapian
634 my $barrier = $self->barrier_init(scalar @$shards);
636 # each shard needs to issue a barrier command
637 $_->remote_barrier for @$shards;
639 # wait for each Xapian shard
640 $self->barrier_wait($barrier);
642 $_->remote_commit for @$shards;
645 # last_commit is special, don't commit these until
646 # remote shards are done:
648 set_last_commits($self);
653 $self->{total_bytes} += $self->{transact_bytes};
654 $self->{transact_bytes} = 0;
657 # issue a write barrier to ensure all data is visible to other processes
658 # and read-only ops. Order of data importance is: git > SQLite > Xapian
660 sub barrier { checkpoint($_[0], 1) };
665 my $im = delete $self->{im};
666 $im->done if $im; # PublicInbox::Import::done
668 my $mm = delete $self->{mm};
669 $mm->{dbh}->commit if $mm;
670 my $shards = delete $self->{idx_shards};
672 $_->remote_close for @$shards;
674 $self->{over}->disconnect;
675 delete $self->{bnote};
676 my $nbytes = $self->{total_bytes};
677 $self->{total_bytes} = 0;
678 $self->lock_release(!!$nbytes) if $shards;
679 $self->{ibx}->git->cleanup;
682 sub fill_alternates ($$) {
683 my ($self, $epoch) = @_;
685 my $pfx = "$self->{ibx}->{inboxdir}/git";
686 my $all = "$self->{ibx}->{inboxdir}/all.git";
687 PublicInbox::Import::init_bare($all) unless -d $all;
688 my $info_dir = "$all/objects/info";
689 my $alt = "$info_dir/alternates";
693 open(my $fh, '<', $alt) or die "open < $alt: $!\n";
694 $mode = (stat($fh))[2] & 07777;
696 # we assign a sort score to every alternate and favor
697 # the newest (highest numbered) one because loose objects
698 # require scanning epochs and only the latest epoch is
699 # expected to see loose objects
701 my $other = 0; # in case admin adds non-epoch repos
703 if (m!\A\Q../../\E([0-9]+)\.git/objects\z!) {
709 } split(/\n+/, do { local $/; <$fh> });
712 foreach my $i (0..$epoch) {
713 my $dir = "../../git/$i.git/objects";
714 if (!exists($alt{$dir}) && -d "$pfx/$i.git") {
721 my ($fh, $tmp) = tempfile('alt-XXXXXXXX', DIR => $info_dir);
722 print $fh join("\n", sort { $alt{$b} <=> $alt{$a} } keys %alt), "\n"
723 or die "print $tmp: $!\n";
724 chmod($mode, $fh) or die "fchmod $tmp: $!\n";
725 close $fh or die "close $tmp $!\n";
726 rename($tmp, $alt) or die "rename $tmp => $alt: $!\n";
730 my ($self, $epoch) = @_;
731 my $git_dir = "$self->{ibx}->{inboxdir}/git/$epoch.git";
732 PublicInbox::Import::init_bare($git_dir);
733 my @cmd = (qw/git config/, "--file=$git_dir/config",
734 'include.path', '../../all.git/config');
735 PublicInbox::Import::run_die(\@cmd);
736 fill_alternates($self, $epoch);
741 my ($self, $max) = @_;
743 my $pfx = "$self->{ibx}->{inboxdir}/git";
744 return unless -d $pfx;
746 opendir my $dh, $pfx or die "opendir $pfx: $!\n";
747 while (defined(my $git_dir = readdir($dh))) {
748 $git_dir =~ m!\A([0-9]+)\.git\z! or next;
751 $latest = "$pfx/$git_dir";
759 my $im = $self->{im};
761 if ($im->{bytes_added} < $self->{rotate_bytes}) {
768 my $git_dir = $self->git_init(++$self->{epoch_max});
769 my $git = PublicInbox::Git->new($git_dir);
770 return $self->import_init($git, 0);
775 my $latest = git_dir_latest($self, \$max);
776 if (defined $latest) {
777 my $git = PublicInbox::Git->new($latest);
778 my $packed_bytes = $git->packed_bytes;
779 my $unpacked_bytes = $packed_bytes / $PACKING_FACTOR;
781 if ($unpacked_bytes >= $self->{rotate_bytes}) {
784 $self->{epoch_max} = $max;
785 return $self->import_init($git, $packed_bytes);
788 $self->{epoch_max} = $epoch;
789 $latest = $self->git_init($epoch);
790 $self->import_init(PublicInbox::Git->new($latest), 0);
794 my ($self, $git, $packed_bytes, $tmp) = @_;
795 my $im = PublicInbox::Import->new($git, undef, undef, $self->{ibx});
796 $im->{bytes_added} = int($packed_bytes / $PACKING_FACTOR);
797 $im->{lock_path} = undef;
798 $im->{path_type} = 'v2';
799 $self->{im} = $im unless $tmp;
805 my ($mid, $cur, $new) = @_;
807 my ($ah, $an) = tempfile('email-cur-XXXXXXXX', TMPDIR => 1);
808 print $ah $cur->as_string or die "print: $!";
809 close $ah or die "close: $!";
810 my ($bh, $bn) = tempfile('email-new-XXXXXXXX', TMPDIR => 1);
811 PublicInbox::Import::drop_unwanted_headers($new);
812 print $bh $new->as_string or die "print: $!";
813 close $bh or die "close: $!";
814 my $cmd = [ qw(diff -u), $an, $bn ];
815 print STDERR "# MID conflict <$mid>\n";
816 my $pid = spawn($cmd, undef, { 1 => 2 });
817 waitpid($pid, 0) == $pid or die "diff did not finish";
822 my ($self, $smsg) = @_;
823 if (my $im = $self->{im}) {
824 my $msg = $im->cat_blob($smsg->{blob});
827 # older message, should be in alternates
828 $self->{ibx}->msg_by_smsg($smsg);
831 sub content_exists ($$$) {
832 my ($self, $mime, $mid) = @_;
833 my $over = $self->{over};
834 my $chashes = content_hashes($mime);
836 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
837 my $msg = get_blob($self, $smsg);
838 if (!defined($msg)) {
839 warn "broken smsg for $mid\n";
842 my $cur = PublicInbox::Eml->new($msg);
843 return 1 if content_matches($chashes, $cur);
845 # XXX DEBUG_DIFF is experimental and may be removed
846 diff($mid, $cur, $mime) if $ENV{DEBUG_DIFF};
853 if (my $shards = $self->{idx_shards}) {
854 $_->atfork_child foreach @$shards;
856 if (my $im = $self->{im}) {
859 die "unexpected mm" if $self->{mm};
860 close $self->{bnote}->[0] or die "close bnote[0]: $!\n";
864 sub reindex_checkpoint ($$) {
865 my ($self, $sync) = @_;
867 $sync->{mm_tmp}->atfork_prepare;
868 $self->done; # release lock
870 if (my $pr = $sync->{-opt}->{-progress}) {
871 $pr->(sprintf($sync->{-regen_fmt}, $sync->{nr}));
874 # allow -watch or -mda to write...
875 $self->idx_init; # reacquire lock
876 $sync->{mm_tmp}->atfork_parent;
879 sub reindex_oid ($$$) {
880 my ($self, $sync, $oid) = @_;
881 return if PublicInbox::SearchIdx::too_big($self, $oid);
882 my ($num, $mid0, $len);
883 my $msgref = $self->{ibx}->git->cat_file($oid, \$len);
884 return if $len == 0; # purged
885 my $mime = PublicInbox::Eml->new($$msgref);
886 my $mids = mids($mime->header_obj);
887 my $chash = content_hash($mime);
889 if (scalar(@$mids) == 0) {
890 warn "E: $oid has no Message-ID, skipping\n";
894 # {unindexed} is unlikely
895 if ((my $unindexed = $self->{unindexed}) && scalar(@$mids) == 1) {
896 $num = delete($unindexed->{$mids->[0]});
899 $self->{mm}->mid_set($num, $mid0);
900 delete($self->{unindexed}) if !keys(%$unindexed);
903 if (!defined($num)) { # reuse if reindexing (or duplicates)
904 my $over = $self->{over};
905 for my $mid (@$mids) {
906 ($num, $mid0) = $over->num_mid0_for_oid($oid, $mid);
907 last if defined $num;
910 $mid0 //= do { # is this a number we got before?
911 $num = $sync->{mm_tmp}->num_for($mids->[0]);
912 defined($num) ? $mids->[0] : undef;
914 if (!defined($num)) {
915 for (my $i = $#$mids; $i >= 1; $i--) {
916 $num = $sync->{mm_tmp}->num_for($mids->[$i]);
924 $sync->{mm_tmp}->num_delete($num);
925 } else { # never seen
926 $num = $self->{mm}->mid_insert($mids->[0]);
929 } else { # rare, try the rest of them, backwards
930 for (my $i = $#$mids; $i >= 1; $i--) {
931 $num = $self->{mm}->mid_insert($mids->[$i]);
939 if (!defined($num)) {
940 warn "E: $oid <", join('> <', @$mids), "> is a duplicate\n";
949 }, 'PublicInbox::Smsg';
950 $smsg->populate($mime, $sync);
951 if (do_idx($self, $msgref, $mime, $smsg)) {
952 reindex_checkpoint($self, $sync);
956 # only update last_commit for $i on reindex iff newer than current
957 sub update_last_commit ($$$$) {
958 my ($self, $git, $i, $cmt) = @_;
959 my $last = last_epoch_commit($self, $i);
960 if (defined $last && is_ancestor($git, $last, $cmt)) {
961 my @cmd = (qw(rev-list --count), "$last..$cmt");
962 chomp(my $n = $git->qx(@cmd));
963 return if $n ne '' && $n == 0;
965 last_epoch_commit($self, $i, $cmt);
968 sub git_dir_n ($$) { "$_[0]->{ibx}->{inboxdir}/git/$_[1].git" }
970 sub last_commits ($$) {
971 my ($self, $epoch_max) = @_;
973 for (my $i = $epoch_max; $i >= 0; $i--) {
974 $heads->[$i] = last_epoch_commit($self, $i);
979 *is_ancestor = *PublicInbox::SearchIdx::is_ancestor;
981 # returns a revision range for git-log(1)
982 sub log_range ($$$$$) {
983 my ($self, $sync, $git, $i, $tip) = @_;
984 my $opt = $sync->{-opt};
985 my $pr = $opt->{-progress} if (($opt->{verbose} || 0) > 1);
986 my $cur = $sync->{ranges}->[$i] or do {
987 $pr->("$i.git indexing all of $tip") if $pr;
988 return $tip; # all of it
991 # fast equality check to avoid (v)fork+execve overhead
993 $sync->{ranges}->[$i] = undef;
997 my $range = "$cur..$tip";
998 $pr->("$i.git checking contiguity... ") if $pr;
999 if (is_ancestor($git, $cur, $tip)) { # common case
1000 $pr->("OK\n") if $pr;
1001 my $n = $git->qx(qw(rev-list --count), $range);
1004 $sync->{ranges}->[$i] = undef;
1005 $pr->("$i.git has nothing new\n") if $pr;
1006 return; # nothing to do
1008 $pr->("$i.git has $n changes since $cur\n") if $pr;
1010 $pr->("FAIL\n") if $pr;
1012 discontiguous range: $range
1013 Rewritten history? (in $git->{git_dir})
1015 chomp(my $base = $git->qx('merge-base', $tip, $cur));
1017 $range = "$base..$tip";
1018 warn "found merge-base: $base\n"
1021 warn "discarding history at $cur\n";
1024 reindexing $git->{git_dir} starting at
1027 $sync->{unindex_range}->{$i} = "$base..$cur";
1032 sub prepare_range_stack {
1033 my ($git, $sync, $range) = @_;
1034 # Don't bump num_highwater on --reindex by using {D}.
1035 # We intentionally do NOT use {D} in the non-reindex case because
1036 # we want NNTP article number gaps from unindexed messages to
1037 # show up in mirrors, too.
1038 my $D = $sync->{D} //= $sync->{reindex} ? {} : undef; # OID_BIN => NR
1040 my $fh = $git->popen(qw(log --raw -r --pretty=tformat:%at-%ct-%H
1041 --no-notes --no-color --no-renames --no-abbrev),
1043 my ($at, $ct, $stk);
1045 if (/\A([0-9]+)-([0-9]+)-($OID)$/o) {
1046 ($at, $ct) = ($1 + 0, $2 + 0);
1047 $stk //= PublicInbox::IdxStack->new($3);
1048 } elsif (/\A:\d{6} 100644 $OID ($OID) [AM]\td$/o) {
1050 if ($D) { # reindex case
1051 $D->{pack('H*', $oid)}++;
1052 } else { # non-reindex case:
1053 $stk->push_rec('d', $at, $ct, $oid);
1055 } elsif (/\A:\d{6} 100644 $OID ($OID) [AM]\tm$/o) {
1058 my $oid_bin = pack('H*', $oid);
1059 my $nr = --$D->{$oid_bin};
1060 delete($D->{$oid_bin}) if $nr <= 0;
1062 # nr < 0 (-1) means it never existed
1063 $stk->push_rec('m', $at, $ct, $oid) if $nr < 0;
1065 $stk->push_rec('m', $at, $ct, $oid);
1069 close $fh or die "git log failed: \$?=$?";
1070 $stk ? $stk->read_prepare : undef;
1073 sub sync_prepare ($$$) {
1074 my ($self, $sync, $epoch_max) = @_;
1075 my $pr = $sync->{-opt}->{-progress};
1077 my $head = $self->{ibx}->{ref_head} || 'refs/heads/master';
1079 # reindex stops at the current heads and we later rerun index_sync
1081 my $reindex_heads = last_commits($self, $epoch_max) if $sync->{reindex};
1083 for (my $i = $epoch_max; $i >= 0; $i--) {
1084 my $git_dir = git_dir_n($self, $i);
1085 -d $git_dir or next; # missing epochs are fine
1086 my $git = PublicInbox::Git->new($git_dir);
1087 if ($reindex_heads) {
1088 $head = $reindex_heads->[$i] or next;
1090 chomp(my $tip = $git->qx(qw(rev-parse -q --verify), $head));
1092 next if $?; # new repo
1093 my $range = log_range($self, $sync, $git, $i, $tip) or next;
1094 # can't use 'rev-list --count' if we use --diff-filter
1095 $pr->("$i.git counting $range ... ") if $pr;
1096 my $stk = prepare_range_stack($git, $sync, $range);
1097 my $nr = $stk ? $stk->num_records : 0;
1098 $pr->("$nr\n") if $pr;
1099 $sync->{stacks}->[$i] = $stk if $stk;
1103 # XXX this should not happen unless somebody bypasses checks in
1104 # our code and blindly injects "d" file history into git repos
1105 if (my @leftovers = keys %{delete($sync->{D}) // {}}) {
1106 warn('W: unindexing '.scalar(@leftovers)." leftovers\n");
1107 for my $oid (@leftovers) {
1108 $oid = unpack('H*', $oid);
1109 $self->{current_info} = "leftover $oid";
1110 unindex_oid($self, $oid);
1113 return 0 if (!$regen_max && !keys(%{$self->{unindex_range}}));
1115 # reindex should NOT see new commits anymore, if we do,
1116 # it's a problem and we need to notice it via die()
1117 my $pad = length($regen_max) + 1;
1118 $sync->{-regen_fmt} = "% ${pad}u/$regen_max\n";
1120 return -1 if $sync->{reindex};
1121 $regen_max + $self->{mm}->num_highwater() || 0;
1124 sub unindex_oid_remote ($$$) {
1125 my ($self, $oid, $mid) = @_;
1126 my @removed = $self->{over}->remove_oid($oid, $mid);
1127 for my $num (@removed) {
1128 my $idx = idx_shard($self, $num % $self->{shards});
1129 $idx->remote_remove($oid, $num);
1133 sub unindex_oid ($$;$) {
1134 my ($self, $oid, $unindexed) = @_;
1135 my $mm = $self->{mm};
1136 my $msgref = $self->{ibx}->git->cat_file($oid);
1137 my $mime = PublicInbox::Eml->new($msgref);
1138 my $mids = mids($mime->header_obj);
1139 $mime = $msgref = undef;
1140 my $over = $self->{over};
1141 foreach my $mid (@$mids) {
1144 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
1145 $gone{$smsg->{num}} = 1 if $oid eq $smsg->{blob};
1147 my $n = scalar(keys(%gone)) or next;
1149 warn "BUG: multiple articles linked to $oid\n",
1150 join(',',sort keys %gone), "\n";
1152 foreach my $num (keys %gone) {
1154 my $mid0 = $mm->mid_for($num);
1155 $unindexed->{$mid0} = $num;
1157 $mm->num_delete($num);
1159 unindex_oid_remote($self, $oid, $mid);
1163 # this is rare, it only happens when we get discontiguous history in
1164 # a mirror because the source used -purge or -edit
1165 sub unindex ($$$$) {
1166 my ($self, $sync, $git, $unindex_range) = @_;
1167 my $unindexed = $self->{unindexed} ||= {}; # $mid0 => $num
1168 my $before = scalar keys %$unindexed;
1169 # order does not matter, here:
1170 my @cmd = qw(log --raw -r
1171 --no-notes --no-color --no-abbrev --no-renames);
1172 my $fh = $git->popen(@cmd, $unindex_range);
1174 /\A:\d{6} 100644 $OID ($OID) [AM]\tm$/o or next;
1175 unindex_oid($self, $1, $unindexed);
1177 close $fh or die "git log failed: \$?=$?";
1179 return unless $sync->{-opt}->{prune};
1180 my $after = scalar keys %$unindexed;
1181 return if $before == $after;
1183 # ensure any blob can not longer be accessed via dumb HTTP
1184 PublicInbox::Import::run_die(['git', "--git-dir=$git->{git_dir}",
1185 qw(-c gc.reflogExpire=now gc --prune=all --quiet)]);
1188 sub sync_ranges ($$$) {
1189 my ($self, $sync, $epoch_max) = @_;
1190 my $reindex = $sync->{reindex};
1192 return last_commits($self, $epoch_max) unless $reindex;
1193 return [] if ref($reindex) ne 'HASH';
1195 my $ranges = $reindex->{from}; # arrayref;
1196 if (ref($ranges) ne 'ARRAY') {
1197 die 'BUG: $reindex->{from} not an ARRAY';
1202 sub index_epoch ($$$) {
1203 my ($self, $sync, $i) = @_;
1205 my $git_dir = git_dir_n($self, $i);
1206 -d $git_dir or return; # missing epochs are fine
1207 my $git = PublicInbox::Git->new($git_dir);
1208 if (my $unindex_range = delete $sync->{unindex_range}->{$i}) { # rare
1209 unindex($self, $sync, $git, $unindex_range);
1211 defined(my $stk = $sync->{stacks}->[$i]) or return;
1212 $sync->{stacks}->[$i] = undef;
1213 while (my ($f, $at, $ct, $oid) = $stk->pop_rec) {
1214 $self->{current_info} = "$i.git $oid";
1216 $sync->{autime} = $at;
1217 $sync->{cotime} = $ct;
1218 reindex_oid($self, $sync, $oid);
1219 } elsif ($f eq 'd') {
1220 unindex_oid($self, $oid);
1223 delete @$sync{qw(autime cotime)};
1224 update_last_commit($self, $git, $i, $stk->{latest_cmt});
1227 # public, called by public-inbox-index
1229 my ($self, $opt) = @_;
1231 my $pr = $opt->{-progress};
1233 my $latest = git_dir_latest($self, \$epoch_max);
1234 return unless defined $latest;
1235 $self->idx_init($opt); # acquire lock
1236 fill_alternates($self, $epoch_max);
1237 $self->{over}->rethread_prepare($opt);
1239 unindex_range => {}, # EPOCH => oid_old..oid_new
1240 reindex => $opt->{reindex},
1243 $sync->{ranges} = sync_ranges($self, $sync, $epoch_max);
1244 if (sync_prepare($self, $sync, $epoch_max)) {
1245 # tmp_clone seems to fail if inside a transaction, so
1246 # we rollback here (because we opened {mm} for reading)
1247 # Note: we do NOT rely on DBI transactions for atomicity;
1248 # only for batch performance.
1249 $self->{mm}->{dbh}->rollback;
1250 $self->{mm}->{dbh}->begin_work;
1251 $sync->{mm_tmp} = $self->{mm}->tmp_clone;
1254 # work forwards through history
1255 index_epoch($self, $sync, $_) for (0..$epoch_max);
1258 if (my $nr = $sync->{nr}) {
1259 my $pr = $sync->{-opt}->{-progress};
1260 $pr->('all.git '.sprintf($sync->{-regen_fmt}, $nr)) if $pr;
1262 $self->{over}->rethread_done($opt);
1264 # reindex does not pick up new changes, so we rerun w/o it:
1265 if ($opt->{reindex}) {
1268 delete @again{qw(rethread reindex -skip_lock)};
1269 index_sync($self, \%again);