1 # Copyright (C) 2018-2021 all contributors <meta@public-inbox.org>
2 # License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
4 # This interface wraps and mimics PublicInbox::Import
5 # Used to write to V2 inboxes (see L<public-inbox-v2-format(5)>).
6 package PublicInbox::V2Writable;
9 use parent qw(PublicInbox::Lock);
10 use PublicInbox::SearchIdxShard;
13 use PublicInbox::Import;
14 use PublicInbox::MID qw(mids references);
15 use PublicInbox::ContentHash qw(content_hash content_digest);
16 use PublicInbox::InboxWritable;
17 use PublicInbox::OverIdx;
18 use PublicInbox::Msgmap;
19 use PublicInbox::Spawn qw(spawn popen_rd run_die);
20 use PublicInbox::SearchIdx qw(log2stack is_ancestor check_size is_bad_blob);
21 use IO::Handle; # ->autoflush
24 my $OID = qr/[a-f0-9]{40,}/;
25 # an estimate of the post-packed size to the raw uncompressed size
26 our $PACKING_FACTOR = 0.4;
28 # SATA storage lags behind what CPUs are capable of, so relying on
29 # nproc(1) can be misleading and having extra Xapian shards is a
30 # waste of FDs and space. It can also lead to excessive IO latency
31 # and slow things down. Users on NVME or other fast storage can
32 # use the NPROC env or switches in our script/public-inbox-* programs
33 # to increase Xapian shards
34 our $NPROC_MAX_DEFAULT = 4;
37 # getconf(1) is POSIX, but *NPROCESSORS* vars are not
38 for (qw(_NPROCESSORS_ONLN NPROCESSORS_ONLN)) {
39 `getconf $_ 2>/dev/null` =~ /^(\d+)$/ and return $1;
41 for my $nproc (qw(nproc gnproc)) { # GNU coreutils nproc
42 `$nproc 2>/dev/null` =~ /^(\d+)$/ and return $1;
45 # should we bother with `sysctl hw.ncpu`? Those only give
46 # us total processor count, not online processor count.
50 sub nproc_shards ($) {
52 my $n = $creat_opt->{nproc} if ref($creat_opt) eq 'HASH';
55 # assume 2 cores if not detectable or zero
56 state $NPROC_DETECTED = detect_nproc() || 2;
58 $n = $NPROC_MAX_DEFAULT if $n > $NPROC_MAX_DEFAULT;
61 # subtract for the main process and git-fast-import
66 sub count_shards ($) {
68 if (my $ibx = $self->{ibx}) {
69 # always load existing shards in case core count changes:
70 # Also, shard count may change while -watch is running
71 my $srch = $ibx->search or return 0;
72 delete $ibx->{search};
74 } else { # ExtSearchIdx
75 $self->{nshard} ||= scalar($self->xdb_shards_flat);
80 # $creat may be any true value, or 0/undef. A hashref is true,
81 # and $creat->{nproc} may be set to an integer
82 my ($class, $v2ibx, $creat) = @_;
83 $v2ibx = PublicInbox::InboxWritable->new($v2ibx);
84 my $dir = $v2ibx->assert_usable_dir;
88 File::Path::mkpath($dir);
90 die "$dir does not exist\n";
93 my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION;
96 im => undef, # PublicInbox::Import
102 oidx => PublicInbox::OverIdx->new("$xpfx/over.sqlite3"),
103 lock_path => "$dir/inbox.lock",
104 # limit each git repo (epoch) to 1GB or so
105 rotate_bytes => int((1024 * 1024 * 1024) / $PACKING_FACTOR),
106 last_commit => [], # git epoch -> commit
108 $self->{oidx}->{-no_fsync} = 1 if $v2ibx->{-no_fsync};
109 $self->{shards} = count_shards($self) || nproc_shards($creat);
115 my ($self, $shards, $skip_epoch, $skip_artnum) = @_;
116 if (defined $shards) {
117 $self->{parallel} = 0 if $shards == 0;
118 $self->{shards} = $shards if $shards > 0;
121 $self->{mm}->skip_artnum($skip_artnum) if defined $skip_artnum;
122 my $max = $self->{ibx}->max_git_epoch;
123 $max = $skip_epoch if (defined($skip_epoch) && !defined($max));
124 $self->git_init($max // 0);
128 # returns undef on duplicate or spam
129 # mimics Import::add and wraps it for v2
131 my ($self, $eml, $check_cb) = @_;
132 $self->{ibx}->with_umask(\&_add, $self, $eml, $check_cb);
136 my ($self, $num) = @_;
137 $self->{idx_shards}->[$num % scalar(@{$self->{idx_shards}})];
140 # indexes a message, returns true if checkpointing is needed
142 my ($self, $eml, $smsg) = @_;
143 $self->{oidx}->add_overview($eml, $smsg);
144 if ($self->{-need_xapian}) {
145 my $idx = idx_shard($self, $smsg->{num});
146 $idx->index_eml($eml, $smsg);
148 my $n = $self->{transact_bytes} += $smsg->{bytes};
149 $n >= $self->{batch_bytes};
153 my ($self, $mime, $check_cb) = @_;
157 $mime = $check_cb->($mime, $self->{ibx}) or return;
160 # All pipes (> $^F) known to Perl 5.6+ have FD_CLOEXEC set,
161 # as does SQLite 3.4.1+ (released in 2007-07-20), and
162 # Xapian 1.3.2+ (released 2015-03-15).
163 # For the most part, we can spawn git-fast-import without
164 # leaking FDs to it...
167 my ($num, $mid0) = v2_num_for($self, $mime);
168 defined $num or return; # duplicate
169 defined $mid0 or die "BUG: \$mid0 undefined\n";
170 my $im = $self->importer;
171 my $smsg = bless { mid => $mid0, num => $num }, 'PublicInbox::Smsg';
172 my $cmt = $im->add($mime, undef, $smsg); # sets $smsg->{ds|ts|blob}
173 $cmt = $im->get_mark($cmt);
174 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
176 if (do_idx($self, $mime, $smsg)) {
184 my ($self, $mime) = @_;
185 my $mids = mids($mime);
187 my $mid = $mids->[0];
188 my $num = $self->{mm}->mid_insert($mid);
189 if (defined $num) { # common case
193 # crap, Message-ID is already known, hope somebody just resent:
194 foreach my $m (@$mids) {
195 # read-only lookup now safe to do after above barrier
196 # easy, don't store duplicates
197 # note: do not add more diagnostic info here since
198 # it gets noisy on public-inbox-watch restarts
199 return () if content_exists($self, $mime, $m);
202 # AltId may pre-populate article numbers (e.g. X-Mail-Count
203 # or NNTP article number), use that article number if it's
205 my $altid = $self->{ibx}->{altid};
206 if ($altid && grep(/:file=msgmap\.sqlite3\z/, @$altid)) {
207 my $num = $self->{mm}->num_for($mid);
209 if (defined $num && !$self->{oidx}->get_art($num)) {
215 warn "<$mid> reused for mismatched content\n";
217 # try the rest of the mids
218 for(my $i = $#$mids; $i >= 1; $i--) {
220 $num = $self->{mm}->mid_insert($m);
222 warn "alternative <$m> for <$mid> found\n";
227 # none of the existing Message-IDs are good, generate a new one:
228 v2_num_for_harder($self, $mime);
231 sub v2_num_for_harder {
232 my ($self, $eml) = @_;
234 my $dig = content_digest($eml);
235 my $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
236 my $num = $self->{mm}->mid_insert($mid0);
237 unless (defined $num) {
238 # it's hard to spoof the last Received: header
239 my @recvd = $eml->header_raw('Received');
240 $dig->add("Received: $_") foreach (@recvd);
241 $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
242 $num = $self->{mm}->mid_insert($mid0);
244 # fall back to a random Message-ID and give up determinism:
245 until (defined($num)) {
247 $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
248 warn "using random Message-ID <$mid0> as fallback\n";
249 $num = $self->{mm}->mid_insert($mid0);
252 PublicInbox::Import::append_mid($eml, $mid0);
256 sub _idx_init { # with_umask callback
257 my ($self, $opt) = @_;
258 $self->lock_acquire unless $opt && $opt->{-skip_lock};
259 $self->{oidx}->create;
261 # xcpdb can change shard count while -watch is idle
262 my $nshards = count_shards($self);
263 $self->{shards} = $nshards if $nshards && $nshards != $self->{shards};
264 $self->{batch_bytes} = $opt->{batch_size} //
265 $PublicInbox::SearchIdx::BATCH_BYTES;
267 # need to create all shards before initializing msgmap FD
268 # idx_shards must be visible to all forked processes
269 my $max = $self->{shards} - 1;
270 my $idx = $self->{idx_shards} = [];
271 push @$idx, PublicInbox::SearchIdxShard->new($self, $_) for (0..$max);
272 $self->{-need_xapian} = $idx->[0]->need_xapian;
274 # SearchIdxShard may do their own flushing, so don't scale
275 # until after forking
276 $self->{batch_bytes} *= $self->{shards} if $self->{parallel};
278 my $ibx = $self->{ibx} or return; # ExtIdxSearch
280 # Now that all subprocesses are up, we can open the FDs
282 my $mm = $self->{mm} = PublicInbox::Msgmap->new_file(
283 "$ibx->{inboxdir}/msgmap.sqlite3",
284 $ibx->{-no_fsync} ? 2 : 1);
285 $mm->{dbh}->begin_work;
288 sub parallel_init ($$) {
289 my ($self, $indexlevel) = @_;
290 $self->{parallel} = 0 if ($indexlevel // 'full') eq 'basic';
295 my ($self, $opt) = @_;
296 return if $self->{idx_shards};
297 my $ibx = $self->{ibx};
299 # do not leak read-only FDs to child processes, we only have these
300 # FDs for duplicate detection so they should not be
301 # frequently activated.
302 delete @$ibx{qw(mm search)};
305 parallel_init($self, $ibx->{indexlevel});
306 $ibx->with_umask(\&_idx_init, $self, $opt);
309 # returns an array mapping [ epoch => latest_commit ]
310 # latest_commit may be undef if nothing was done to that epoch
311 # $replace_map = { $object_id => $strref, ... }
312 sub _replace_oids ($$$) {
313 my ($self, $mime, $replace_map) = @_;
315 my $ibx = $self->{ibx};
316 my $pfx = "$ibx->{inboxdir}/git";
317 my $rewrites = []; # epoch => commit
318 my $max = $self->{epoch_max} //= $ibx->max_git_epoch // return;
320 foreach my $i (0..$max) {
321 my $git_dir = "$pfx/$i.git";
323 my $git = PublicInbox::Git->new($git_dir);
324 my $im = $self->import_init($git, 0, 1);
325 $rewrites->[$i] = $im->replace_oids($mime, $replace_map);
331 sub content_hashes ($) {
333 my @chashes = ( content_hash($mime) );
335 # We still support Email::MIME, here, and
336 # Email::MIME->as_string doesn't always round-trip, so we may
337 # use a second content_hash
338 my $rt = content_hash(PublicInbox::Eml->new(\($mime->as_string)));
339 push @chashes, $rt if $chashes[0] ne $rt;
343 sub content_matches ($$) {
344 my ($chashes, $existing) = @_;
345 my $chash = content_hash($existing);
346 foreach (@$chashes) {
347 return 1 if $_ eq $chash
352 # used for removing or replacing (purging)
353 sub rewrite_internal ($$;$$$) {
354 my ($self, $old_eml, $cmt_msg, $new_eml, $sref) = @_;
356 my ($im, $need_reindex, $replace_map);
358 $replace_map = {}; # oid => sref
359 $need_reindex = [] if $new_eml;
361 $im = $self->importer;
363 my $oidx = $self->{oidx};
364 my $chashes = content_hashes($old_eml);
366 my $mids = mids($old_eml);
368 # We avoid introducing new blobs into git since the raw content
369 # can be slightly different, so we do not need the user-supplied
370 # message now that we have the mids and content_hash
374 foreach my $mid (@$mids) {
375 my %gone; # num => [ smsg, $mime, raw ]
377 while (my $smsg = $oidx->next_by_mid($mid, \$id, \$prev)) {
378 my $msg = get_blob($self, $smsg);
379 if (!defined($msg)) {
380 warn "broken smsg for $mid\n";
384 my $cur = PublicInbox::Eml->new($msg);
385 if (content_matches($chashes, $cur)) {
386 $gone{$smsg->{num}} = [ $smsg, $cur, \$orig ];
389 my $n = scalar keys %gone;
392 warn "BUG: multiple articles linked to <$mid>\n",
393 join(',', sort keys %gone), "\n";
395 foreach my $num (keys %gone) {
396 my ($smsg, $mime, $orig) = @{$gone{$num}};
397 # $removed should only be set once assuming
398 # no bugs in our deduplication code:
399 $removed = [ undef, $mime, $smsg ];
400 my $oid = $smsg->{blob};
402 $replace_map->{$oid} = $sref;
404 ($mark, undef) = $im->remove($orig, $cmt_msg);
405 $removed->[0] = $mark;
408 if ($need_reindex) { # ->replace
409 push @$need_reindex, $smsg;
410 } else { # ->purge or ->remove
411 $self->{mm}->num_delete($num);
413 unindex_oid_aux($self, $oid, $mid);
418 my $cmt = $im->get_mark($mark);
419 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
421 if ($replace_map && scalar keys %$replace_map) {
422 my $rewrites = _replace_oids($self, $new_eml, $replace_map);
423 return { rewrites => $rewrites, need_reindex => $need_reindex };
425 defined($mark) ? $removed : undef;
428 # public (see PublicInbox::Import->remove), but note the 3rd element
429 # (retval[2]) is not part of the stable API shared with Import->remove
431 my ($self, $eml, $cmt_msg) = @_;
432 my $r = $self->{ibx}->with_umask(\&rewrite_internal,
433 $self, $eml, $cmt_msg);
434 defined($r) && defined($r->[0]) ? @$r: undef;
437 sub _replace ($$;$$) {
438 my ($self, $old_eml, $new_eml, $sref) = @_;
439 my $arg = [ $self, $old_eml, undef, $new_eml, $sref ];
440 my $rewritten = $self->{ibx}->with_umask(\&rewrite_internal,
441 $self, $old_eml, undef, $new_eml, $sref) or return;
443 my $rewrites = $rewritten->{rewrites};
444 # ->done is called if there are rewrites since we gc+prune from git
445 $self->idx_init if @$rewrites;
447 for my $i (0..$#$rewrites) {
448 defined(my $cmt = $rewrites->[$i]) or next;
449 $self->{last_commit}->[$i] = $cmt;
456 my ($self, $mime) = @_;
457 my $rewritten = _replace($self, $mime, undef, \'') or return;
458 $rewritten->{rewrites}
461 # returns the git object_id of $fh, does not write the object to FS
462 sub git_hash_raw ($$) {
463 my ($self, $raw) = @_;
464 # grab the expected OID we have to reindex:
465 pipe(my($in, $w)) or die "pipe: $!";
466 my $git_dir = $self->git->{git_dir};
467 my $cmd = ['git', "--git-dir=$git_dir", qw(hash-object --stdin)];
468 my $r = popen_rd($cmd, undef, { 0 => $in });
469 print $w $$raw or die "print \$w: $!";
470 close $w or die "close \$w: $!";
472 chomp(my $oid = <$r>);
473 close $r or die "git hash-object failed: $?";
474 $oid =~ /\A$OID\z/ or die "OID not expected: $oid";
478 sub _check_mids_match ($$$) {
479 my ($old_list, $new_list, $hdrs) = @_;
480 my %old_mids = map { $_ => 1 } @$old_list;
481 my %new_mids = map { $_ => 1 } @$new_list;
482 my @old = keys %old_mids;
483 my @new = keys %new_mids;
484 my $err = "$hdrs may not be changed when replacing\n";
485 die $err if scalar(@old) != scalar(@new);
486 delete @new_mids{@old};
487 delete @old_mids{@new};
488 die $err if (scalar(keys %old_mids) || scalar(keys %new_mids));
491 # Changing Message-IDs or References with ->replace isn't supported.
492 # The rules for dealing with messages with multiple or conflicting
493 # Message-IDs are pretty complex and rethreading hasn't been fully
495 sub check_mids_match ($$) {
496 my ($old, $new) = @_;
497 _check_mids_match(mids($old), mids($new), 'Message-ID(s)');
498 _check_mids_match(references($old), references($new),
499 'References/In-Reply-To');
504 my ($self, $old_mime, $new_mime) = @_;
506 check_mids_match($old_mime, $new_mime);
508 # mutt will always add Content-Length:, Status:, Lines: when editing
509 PublicInbox::Import::drop_unwanted_headers($new_mime);
511 my $raw = $new_mime->as_string;
512 my $expect_oid = git_hash_raw($self, \$raw);
513 my $rewritten = _replace($self, $old_mime, $new_mime, \$raw) or return;
514 my $need_reindex = $rewritten->{need_reindex};
516 # just in case we have bugs in deduplication code:
517 my $n = scalar(@$need_reindex);
519 my $list = join(', ', map {
520 "$_->{num}: <$_->{mid}>"
523 W: rewritten $n messages matching content of original message (expected: 1).
524 W: possible bug in public-inbox, NNTP article IDs and Message-IDs follow:
529 # make sure we really got the OID:
530 my ($blob, $type, $bytes) = $self->git->check($expect_oid);
531 $blob eq $expect_oid or die "BUG: $expect_oid not found after replace";
533 # don't leak FDs to Xapian:
536 # reindex modified messages:
537 for my $smsg (@$need_reindex) {
538 my $new_smsg = bless {
542 }, 'PublicInbox::Smsg';
543 my $sync = { autime => $smsg->{ds}, cotime => $smsg->{ts} };
544 $new_smsg->populate($new_mime, $sync);
545 $new_smsg->set_bytes($raw, $bytes);
546 do_idx($self, $new_mime, $new_smsg);
548 $rewritten->{rewrites};
551 sub last_epoch_commit ($$;$) {
552 my ($self, $i, $cmt) = @_;
553 my $v = PublicInbox::Search::SCHEMA_VERSION();
554 $self->{mm}->last_commit_xap($v, $i, $cmt);
557 sub set_last_commits ($) { # this is NOT for ExtSearchIdx
559 defined(my $epoch_max = $self->{epoch_max}) or return;
560 my $last_commit = $self->{last_commit};
561 foreach my $i (0..$epoch_max) {
562 defined(my $cmt = $last_commit->[$i]) or next;
563 $last_commit->[$i] = undef;
564 last_epoch_commit($self, $i, $cmt);
569 sub checkpoint ($;$) {
570 my ($self, $wait) = @_;
572 if (my $im = $self->{im}) {
579 my $shards = $self->{idx_shards};
581 my $mm = $self->{mm};
582 my $dbh = $mm->{dbh} if $mm;
584 # SQLite msgmap data is second in importance
585 $dbh->commit if $dbh;
587 # SQLite overview is third
588 $self->{oidx}->commit_lazy;
590 # Now deal with Xapian
592 # start commit_txn_lazy asynchronously on all parallel shards
593 # (non-parallel waits here)
594 $_->ipc_do('commit_txn_lazy') for @$shards;
596 # transactions started on parallel shards,
597 # wait for them by issuing an echo command (echo can only
598 # run after commit_txn_lazy is done)
599 if ($wait && $self->{parallel}) {
601 for my $shard (@$shards) {
602 my $echo = $shard->ipc_do('echo', $i);
603 $echo == $i or die <<"";
604 shard[$i] bad echo:$echo != $i waiting for txn commit
610 my $midx = $self->{midx}; # misc index
611 $midx->commit_txn if $midx;
613 # last_commit is special, don't commit these until
614 # Xapian shards are done:
615 $dbh->begin_work if $dbh;
616 set_last_commits($self);
621 $midx->begin_txn if $midx;
623 $self->{total_bytes} += $self->{transact_bytes};
624 $self->{transact_bytes} = 0;
627 # issue a write barrier to ensure all data is visible to other processes
628 # and read-only ops. Order of data importance is: git > SQLite > Xapian
630 sub barrier { checkpoint($_[0], 1) };
632 # true if locked and active
633 sub active { !!$_[0]->{im} }
639 if (my $im = delete $self->{im}) {
640 eval { $im->done }; # PublicInbox::Import::done
641 $err .= "import done: $@\n" if $@;
644 eval { checkpoint($self) };
645 $err .= "checkpoint: $@\n" if $@;
647 if (my $mm = delete $self->{mm}) {
648 my $m = $err ? 'rollback' : 'commit';
649 eval { $mm->{dbh}->$m };
650 $err .= "msgmap $m: $@\n" if $@;
652 my $shards = delete $self->{idx_shards};
655 eval { $_->shard_close };
656 $err .= "shard close: $@\n" if $@;
659 eval { $self->{oidx}->dbh_close };
660 $err .= "over close: $@\n" if $@;
661 delete $self->{midx};
662 my $nbytes = $self->{total_bytes};
663 $self->{total_bytes} = 0;
664 $self->lock_release(!!$nbytes) if $shards;
669 sub write_alternates ($$$) {
670 my ($info_dir, $mode, $out) = @_;
671 my $fh = File::Temp->new(TEMPLATE => 'alt-XXXXXXXX', DIR => $info_dir);
672 my $tmp = $fh->filename;
673 print $fh @$out or die "print $tmp: $!\n";
674 chmod($mode, $fh) or die "fchmod $tmp: $!\n";
675 close $fh or die "close $tmp $!\n";
676 my $alt = "$info_dir/alternates";
677 rename($tmp, $alt) or die "rename $tmp => $alt: $!\n";
678 $fh->unlink_on_destroy(0);
681 sub fill_alternates ($$) {
682 my ($self, $epoch) = @_;
684 my $pfx = "$self->{ibx}->{inboxdir}/git";
685 my $all = "$self->{ibx}->{inboxdir}/all.git";
686 PublicInbox::Import::init_bare($all) unless -d $all;
687 my $info_dir = "$all/objects/info";
688 my $alt = "$info_dir/alternates";
692 open(my $fh, '<', $alt) or die "open < $alt: $!\n";
693 $mode = (stat($fh))[2] & 07777;
695 # we assign a sort score to every alternate and favor
696 # the newest (highest numbered) one because loose objects
697 # require scanning epochs and only the latest epoch is
698 # expected to see loose objects
700 my $other = 0; # in case admin adds non-epoch repos
702 if (m!\A\Q../../\E([0-9]+)\.git/objects\z!) {
708 } split(/\n+/, do { local $/; <$fh> });
711 foreach my $i (0..$epoch) {
712 my $dir = "../../git/$i.git/objects";
713 if (!exists($alt{$dir}) && -d "$pfx/$i.git") {
719 write_alternates($info_dir, $mode,
720 [join("\n", sort { $alt{$b} <=> $alt{$a} } keys %alt), "\n"]);
724 my ($self, $epoch) = @_;
725 my $git_dir = "$self->{ibx}->{inboxdir}/git/$epoch.git";
726 PublicInbox::Import::init_bare($git_dir);
727 run_die([qw(git config), "--file=$git_dir/config",
728 qw(include.path ../../all.git/config)]);
729 fill_alternates($self, $epoch);
735 my $im = $self->{im};
737 if ($im->{bytes_added} < $self->{rotate_bytes}) {
744 my $git_dir = $self->git_init(++$self->{epoch_max});
745 my $git = PublicInbox::Git->new($git_dir);
746 return $self->import_init($git, 0);
751 my $latest = $self->{ibx}->git_dir_latest(\$max);
752 if (defined $latest) {
753 my $git = PublicInbox::Git->new($latest);
754 my $packed_bytes = $git->packed_bytes;
755 my $unpacked_bytes = $packed_bytes / $PACKING_FACTOR;
757 if ($unpacked_bytes >= $self->{rotate_bytes}) {
760 $self->{epoch_max} = $max;
761 return $self->import_init($git, $packed_bytes);
764 $self->{epoch_max} = $epoch;
765 $latest = $self->git_init($epoch);
766 $self->import_init(PublicInbox::Git->new($latest), 0);
770 my ($self, $git, $packed_bytes, $tmp) = @_;
771 my $im = PublicInbox::Import->new($git, undef, undef, $self->{ibx});
772 $im->{bytes_added} = int($packed_bytes / $PACKING_FACTOR);
773 $im->{lock_path} = undef;
774 $im->{path_type} = 'v2';
775 $self->{im} = $im unless $tmp;
781 my ($mid, $cur, $new) = @_;
783 my $ah = File::Temp->new(TEMPLATE => 'email-cur-XXXXXXXX', TMPDIR => 1);
784 print $ah $cur->as_string or die "print: $!";
785 $ah->flush or die "flush: $!";
786 PublicInbox::Import::drop_unwanted_headers($new);
787 my $bh = File::Temp->new(TEMPLATE => 'email-new-XXXXXXXX', TMPDIR => 1);
788 print $bh $new->as_string or die "print: $!";
789 $bh->flush or die "flush: $!";
790 my $cmd = [ qw(diff -u), $ah->filename, $bh->filename ];
791 print STDERR "# MID conflict <$mid>\n";
792 my $pid = spawn($cmd, undef, { 1 => 2 });
793 waitpid($pid, 0) == $pid or die "diff did not finish";
797 my ($self, $smsg) = @_;
798 if (my $im = $self->{im}) {
799 my $msg = $im->cat_blob($smsg->{blob});
802 # older message, should be in alternates
803 $self->{ibx}->msg_by_smsg($smsg);
806 sub content_exists ($$$) {
807 my ($self, $mime, $mid) = @_;
808 my $oidx = $self->{oidx};
809 my $chashes = content_hashes($mime);
811 while (my $smsg = $oidx->next_by_mid($mid, \$id, \$prev)) {
812 my $msg = get_blob($self, $smsg);
813 if (!defined($msg)) {
814 warn "broken smsg for $mid\n";
817 my $cur = PublicInbox::Eml->new($msg);
818 return 1 if content_matches($chashes, $cur);
820 # XXX DEBUG_DIFF is experimental and may be removed
821 diff($mid, $cur, $mime) if $ENV{DEBUG_DIFF};
828 if (my $older_siblings = $self->{idx_shards}) {
829 $_->ipc_sibling_atfork_child for @$older_siblings;
831 if (my $im = $self->{im}) {
834 die "BUG: unexpected mm" if $self->{mm};
837 sub reindex_checkpoint ($$) {
838 my ($self, $sync) = @_;
840 $self->git->async_wait_all;
841 $self->update_last_commit($sync);
842 ${$sync->{need_checkpoint}} = 0;
843 my $mm_tmp = $sync->{mm_tmp};
844 $mm_tmp->atfork_prepare if $mm_tmp;
845 die 'BUG: {im} during reindex' if $self->{im};
846 if ($self->{ibx_map} && !$sync->{checkpoint_unlocks}) {
847 checkpoint($self, 1); # no need to release lock on pure index
849 $self->done; # release lock
852 if (my $pr = $sync->{-regen_fmt} ? $sync->{-opt}->{-progress} : undef) {
853 $pr->(sprintf($sync->{-regen_fmt}, ${$sync->{nr}}));
856 # allow -watch or -mda to write...
857 $self->idx_init($sync->{-opt}); # reacquire lock
858 if (my $intvl = $sync->{check_intvl}) { # eidx
859 $sync->{next_check} = PublicInbox::DS::now() + $intvl;
861 $mm_tmp->atfork_parent if $mm_tmp;
864 sub index_finalize ($$) {
865 my ($arg, $index) = @_;
866 ++$arg->{self}->{nidx};
867 if (defined(my $cur = $arg->{cur_cmt})) {
868 ${$arg->{latest_cmt}} = $cur;
870 die 'BUG: {cur_cmt} missing';
871 } # else { unindexing @leftovers doesn't set {cur_cmt}
874 sub index_oid { # cat_async callback
875 my ($bref, $oid, $type, $size, $arg) = @_;
876 is_bad_blob($oid, $type, $size, $arg->{oid}) and
877 return index_finalize($arg, 1); # size == 0 purged returns here
878 my $self = $arg->{self};
879 local $self->{current_info} = "$self->{current_info} $oid";
881 my $eml = PublicInbox::Eml->new($$bref);
882 my $mids = mids($eml);
883 my $chash = content_hash($eml);
885 if (scalar(@$mids) == 0) {
886 warn "E: $oid has no Message-ID, skipping\n";
890 # {unindexed} is unlikely
891 if ((my $unindexed = $arg->{unindexed}) && scalar(@$mids) == 1) {
892 $num = delete($unindexed->{$mids->[0]});
895 $self->{mm}->mid_set($num, $mid0);
896 delete($arg->{unindexed}) if !keys(%$unindexed);
899 if (!defined($num)) { # reuse if reindexing (or duplicates)
900 my $oidx = $self->{oidx};
901 for my $mid (@$mids) {
902 ($num, $mid0) = $oidx->num_mid0_for_oid($oid, $mid);
903 last if defined $num;
906 $mid0 //= do { # is this a number we got before?
907 $num = $arg->{mm_tmp}->num_for($mids->[0]);
908 defined($num) ? $mids->[0] : undef;
910 if (!defined($num)) {
911 for (my $i = $#$mids; $i >= 1; $i--) {
912 $num = $arg->{mm_tmp}->num_for($mids->[$i]);
920 $arg->{mm_tmp}->num_delete($num);
921 } else { # never seen
922 $num = $self->{mm}->mid_insert($mids->[0]);
925 } else { # rare, try the rest of them, backwards
926 for (my $i = $#$mids; $i >= 1; $i--) {
927 $num = $self->{mm}->mid_insert($mids->[$i]);
935 if (!defined($num)) {
936 warn "E: $oid <", join('> <', @$mids), "> is a duplicate\n";
944 }, 'PublicInbox::Smsg';
945 $smsg->populate($eml, $arg);
946 $smsg->set_bytes($$bref, $size);
947 if (do_idx($self, $eml, $smsg)) {
948 ${$arg->{need_checkpoint}} = 1;
950 index_finalize($arg, 1);
953 # only update last_commit for $i on reindex iff newer than current
954 sub update_last_commit {
955 my ($self, $sync, $stk) = @_;
956 my $unit = $sync->{unit} // return;
957 my $latest_cmt = $stk ? $stk->{latest_cmt} : ${$sync->{latest_cmt}};
958 defined($latest_cmt) or return;
959 my $last = last_epoch_commit($self, $unit->{epoch});
960 if (defined $last && is_ancestor($self->git, $last, $latest_cmt)) {
961 my @cmd = (qw(rev-list --count), "$last..$latest_cmt");
962 chomp(my $n = $unit->{git}->qx(@cmd));
963 return if $n ne '' && $n == 0;
965 last_epoch_commit($self, $unit->{epoch}, $latest_cmt);
969 my ($self, $sync) = @_;
971 for (my $i = $sync->{epoch_max}; $i >= 0; $i--) {
972 $heads->[$i] = last_epoch_commit($self, $i);
977 # returns a revision range for git-log(1)
978 sub log_range ($$$) {
979 my ($sync, $unit, $tip) = @_;
980 my $opt = $sync->{-opt};
981 my $pr = $opt->{-progress} if (($opt->{verbose} || 0) > 1);
982 my $i = $unit->{epoch};
983 my $cur = $sync->{ranges}->[$i] or do {
984 $pr->("$i.git indexing all of $tip\n") if $pr;
985 return $tip; # all of it
988 # fast equality check to avoid (v)fork+execve overhead
990 $sync->{ranges}->[$i] = undef;
994 my $range = "$cur..$tip";
995 $pr->("$i.git checking contiguity... ") if $pr;
996 my $git = $unit->{git};
997 if (is_ancestor($sync->{self}->git, $cur, $tip)) { # common case
998 $pr->("OK\n") if $pr;
999 my $n = $git->qx(qw(rev-list --count), $range);
1002 $sync->{ranges}->[$i] = undef;
1003 $pr->("$i.git has nothing new\n") if $pr;
1004 return; # nothing to do
1006 $pr->("$i.git has $n changes since $cur\n") if $pr;
1008 $pr->("FAIL\n") if $pr;
1010 discontiguous range: $range
1011 Rewritten history? (in $git->{git_dir})
1013 chomp(my $base = $git->qx('merge-base', $tip, $cur));
1015 $range = "$base..$tip";
1016 warn "found merge-base: $base\n"
1019 warn "discarding history at $cur\n";
1022 reindexing $git->{git_dir}
1025 # $cur^0 may no longer exist if pruned by git
1026 if ($git->qx(qw(rev-parse -q --verify), "$cur^0")) {
1027 $unit->{unindex_range} = "$base..$cur";
1028 } elsif ($base && $git->qx(qw(rev-parse -q --verify), $base)) {
1029 $unit->{unindex_range} = "$base..";
1031 warn "W: unable to unindex before $range\n";
1037 # overridden by ExtSearchIdx
1038 sub artnum_max { $_[0]->{mm}->num_highwater }
1040 sub sync_prepare ($$) {
1041 my ($self, $sync) = @_;
1042 $sync->{ranges} = sync_ranges($self, $sync);
1043 my $pr = $sync->{-opt}->{-progress};
1045 my $head = $sync->{ibx}->{ref_head} || 'HEAD';
1048 ($pfx) = ($sync->{ibx}->{inboxdir} =~ m!([^/]+)\z!g);
1049 $pfx //= $sync->{ibx}->{inboxdir};
1053 if ($self->{ibx_map}) {
1054 # ExtSearchIdx won't index messages unless they're in
1055 # over.sqlite3 for a given inbox, so don't read beyond
1056 # what's in the per-inbox index.
1057 $reindex_heads = [];
1058 my $v = PublicInbox::Search::SCHEMA_VERSION;
1059 my $mm = $sync->{ibx}->mm;
1060 for my $i (0..$sync->{epoch_max}) {
1061 $reindex_heads->[$i] = $mm->last_commit_xap($v, $i);
1063 } elsif ($sync->{reindex}) { # V2 inbox
1064 # reindex stops at the current heads and we later
1065 # rerun index_sync without {reindex}
1066 $reindex_heads = $self->last_commits($sync);
1068 if ($sync->{max_size} = $sync->{-opt}->{max_size}) {
1069 $sync->{index_oid} = $self->can('index_oid');
1071 my $git_pfx = "$sync->{ibx}->{inboxdir}/git";
1072 for (my $i = $sync->{epoch_max}; $i >= 0; $i--) {
1073 my $git_dir = "$git_pfx/$i.git";
1074 -d $git_dir or next; # missing epochs are fine
1075 my $git = PublicInbox::Git->new($git_dir);
1076 my $unit = { git => $git, epoch => $i };
1078 if ($reindex_heads) {
1079 $tip = $head = $reindex_heads->[$i] or next;
1081 $tip = $git->qx(qw(rev-parse -q --verify), $head);
1082 next if $?; # new repo
1085 my $range = log_range($sync, $unit, $tip) or next;
1086 # can't use 'rev-list --count' if we use --diff-filter
1087 $pr->("$pfx $i.git counting $range ... ") if $pr;
1088 # Don't bump num_highwater on --reindex by using {D}.
1089 # We intentionally do NOT use {D} in the non-reindex case
1090 # because we want NNTP article number gaps from unindexed
1091 # messages to show up in mirrors, too.
1092 $sync->{D} //= $sync->{reindex} ? {} : undef; # OID_BIN => NR
1093 my $stk = log2stack($sync, $git, $range);
1094 return 0 if $sync->{quit};
1095 my $nr = $stk ? $stk->num_records : 0;
1096 $pr->("$nr\n") if $pr;
1097 $unit->{stack} = $stk; # may be undef
1098 unshift @{$sync->{todo}}, $unit;
1101 return 0 if $sync->{quit};
1103 # XXX this should not happen unless somebody bypasses checks in
1104 # our code and blindly injects "d" file history into git repos
1105 if (my @leftovers = keys %{delete($sync->{D}) // {}}) {
1106 warn('W: unindexing '.scalar(@leftovers)." leftovers\n");
1107 local $self->{current_info} = 'leftover ';
1108 my $unindex_oid = $self->can('unindex_oid');
1109 for my $oid (@leftovers) {
1110 last if $sync->{quit};
1111 $oid = unpack('H*', $oid);
1112 my $req = { %$sync, oid => $oid };
1113 $self->git->cat_async($oid, $unindex_oid, $req);
1115 $self->git->cat_async_wait;
1117 return 0 if $sync->{quit};
1119 $sync->{-regen_fmt} = "%u/?\n";
1123 # reindex should NOT see new commits anymore, if we do,
1124 # it's a problem and we need to notice it via die()
1125 my $pad = length($regen_max) + 1;
1126 $sync->{-regen_fmt} = "% ${pad}u/$regen_max\n";
1127 $sync->{nr} = \(my $nr = 0);
1128 return -1 if $sync->{reindex};
1129 $regen_max + $self->artnum_max || 0;
1132 sub unindex_oid_aux ($$$) {
1133 my ($self, $oid, $mid) = @_;
1134 my @removed = $self->{oidx}->remove_oid($oid, $mid);
1135 return unless $self->{-need_xapian};
1136 for my $num (@removed) {
1137 idx_shard($self, $num)->ipc_do('xdb_remove', $num);
1141 sub unindex_oid ($$;$) { # git->cat_async callback
1142 my ($bref, $oid, $type, $size, $arg) = @_;
1143 is_bad_blob($oid, $type, $size, $arg->{oid}) and
1144 return index_finalize($arg, 0);
1145 my $self = $arg->{self};
1146 local $self->{current_info} = "$self->{current_info} $oid";
1147 my $unindexed = $arg->{in_unindex} ? $arg->{unindexed} : undef;
1148 my $mm = $self->{mm};
1149 my $mids = mids(PublicInbox::Eml->new($bref));
1151 my $oidx = $self->{oidx};
1152 foreach my $mid (@$mids) {
1155 while (my $smsg = $oidx->next_by_mid($mid, \$id, \$prev)) {
1156 $gone{$smsg->{num}} = 1 if $oid eq $smsg->{blob};
1158 my $n = scalar(keys(%gone)) or next;
1160 warn "BUG: multiple articles linked to $oid\n",
1161 join(',',sort keys %gone), "\n";
1163 foreach my $num (keys %gone) {
1165 my $mid0 = $mm->mid_for($num);
1166 $unindexed->{$mid0} = $num;
1168 $mm->num_delete($num);
1170 unindex_oid_aux($self, $oid, $mid);
1172 index_finalize($arg, 0);
1175 sub git { $_[0]->{ibx}->git }
1177 # this is rare, it only happens when we get discontiguous history in
1178 # a mirror because the source used -purge or -edit
1179 sub unindex_todo ($$$) {
1180 my ($self, $sync, $unit) = @_;
1181 my $unindex_range = delete($unit->{unindex_range}) // return;
1182 my $unindexed = $sync->{unindexed} //= {}; # $mid0 => $num
1183 my $before = scalar keys %$unindexed;
1184 # order does not matter, here:
1185 my $fh = $unit->{git}->popen(qw(log --raw -r --no-notes --no-color
1186 --no-abbrev --no-renames), $unindex_range);
1187 local $sync->{in_unindex} = 1;
1188 my $unindex_oid = $self->can('unindex_oid');
1190 /\A:\d{6} 100644 $OID ($OID) [AM]\tm$/o or next;
1191 $self->git->cat_async($1, $unindex_oid, { %$sync, oid => $1 });
1193 close $fh or die "git log failed: \$?=$?";
1194 $self->git->cat_async_wait;
1196 return unless $sync->{-opt}->{prune};
1197 my $after = scalar keys %$unindexed;
1198 return if $before == $after;
1200 # ensure any blob can not longer be accessed via dumb HTTP
1201 run_die(['git', "--git-dir=$unit->{git}->{git_dir}",
1202 qw(-c gc.reflogExpire=now gc --prune=all --quiet)]);
1205 sub sync_ranges ($$) {
1206 my ($self, $sync) = @_;
1207 my $reindex = $sync->{reindex};
1208 return $self->last_commits($sync) unless $reindex;
1209 return [] if ref($reindex) ne 'HASH';
1211 my $ranges = $reindex->{from}; # arrayref;
1212 if (ref($ranges) ne 'ARRAY') {
1213 die 'BUG: $reindex->{from} not an ARRAY';
1218 sub index_xap_only { # git->cat_async callback
1219 my ($bref, $oid, $type, $size, $smsg) = @_;
1220 my $self = $smsg->{self};
1221 my $idx = idx_shard($self, $smsg->{num});
1222 $idx->index_eml(PublicInbox::Eml->new($bref), $smsg);
1223 $self->{transact_bytes} += $smsg->{bytes};
1226 sub index_xap_step ($$$;$) {
1227 my ($self, $sync, $beg, $step) = @_;
1228 my $end = $sync->{art_end};
1229 return if $beg > $end; # nothing to do
1231 $step //= $self->{shards};
1232 my $ibx = $self->{ibx};
1233 if (my $pr = $sync->{-opt}->{-progress}) {
1234 $pr->("Xapian indexlevel=$ibx->{indexlevel} ".
1235 "$beg..$end (% $step)\n");
1237 for (my $num = $beg; $num <= $end; $num += $step) {
1238 last if $sync->{quit};
1239 my $smsg = $ibx->over->get_art($num) or next;
1240 $smsg->{self} = $self;
1241 $ibx->git->cat_async($smsg->{blob}, \&index_xap_only, $smsg);
1242 if ($self->{transact_bytes} >= $self->{batch_bytes}) {
1243 ${$sync->{nr}} = $num;
1244 reindex_checkpoint($self, $sync);
1249 sub index_todo ($$$) {
1250 my ($self, $sync, $unit) = @_;
1251 return if $sync->{quit};
1252 unindex_todo($self, $sync, $unit);
1253 my $stk = delete($unit->{stack}) or return;
1254 my $all = $self->git;
1255 my $index_oid = $self->can('index_oid');
1256 my $unindex_oid = $self->can('unindex_oid');
1258 if ($unit->{git}->{git_dir} =~ m!/([^/]+)/git/([0-9]+\.git)\z!) {
1259 $pfx = "$1 $2"; # v2
1261 ($pfx) = ($unit->{git}->{git_dir} =~ m!/([^/]+)\z!g);
1262 $pfx //= $unit->{git}->{git_dir};
1264 local $self->{current_info} = "$pfx ";
1265 local $sync->{latest_cmt} = \(my $latest_cmt);
1266 local $sync->{unit} = $unit;
1267 while (my ($f, $at, $ct, $oid, $cmt) = $stk->pop_rec) {
1268 if ($sync->{quit}) {
1269 warn "waiting to quit...\n";
1270 $all->async_wait_all;
1271 $self->update_last_commit($sync);
1282 if ($sync->{max_size}) {
1283 $all->check_async($oid, \&check_size, $req);
1285 $all->cat_async($oid, $index_oid, $req);
1287 } elsif ($f eq 'd') {
1288 $all->cat_async($oid, $unindex_oid, $req);
1290 if (${$sync->{need_checkpoint}}) {
1291 reindex_checkpoint($self, $sync);
1294 $all->async_wait_all;
1295 $self->update_last_commit($sync, $stk);
1299 my ($self, $opt, $sync, $art_beg) = @_;
1300 my $seq = $opt->{sequential_shard};
1302 local $self->{parallel} = 0 if $seq;
1303 $self->idx_init($opt); # acquire lock
1304 if (my $art_end = $self->{ibx}->mm->max) {
1306 need_checkpoint => \(my $bool = 0),
1309 nr => \(my $nr = 0),
1310 -regen_fmt => "%u/?\n",
1312 $sync->{art_end} = $art_end;
1313 if ($seq || !$self->{parallel}) {
1314 my $shard_end = $self->{shards} - 1;
1315 for my $i (0..$shard_end) {
1316 last if $sync->{quit};
1317 index_xap_step($self, $sync, $art_beg + $i);
1318 if ($i != $shard_end) {
1319 reindex_checkpoint($self, $sync);
1322 } else { # parallel (maybe)
1323 index_xap_step($self, $sync, $art_beg, 1);
1326 $self->git->cat_async_wait;
1330 # public, called by public-inbox-index
1332 my ($self, $opt) = @_;
1334 return xapian_only($self, $opt) if $opt->{xapian_only};
1337 my $latest = $self->{ibx}->git_dir_latest(\$epoch_max) // return;
1338 if ($opt->{'fast-noop'}) { # nanosecond (st_ctim) comparison
1339 use Time::HiRes qw(stat);
1340 if (my @mm = stat("$self->{ibx}->{inboxdir}/msgmap.sqlite3")) {
1341 my $c = $mm[10]; # 10 = ctime (nsec NV)
1342 my @hd = stat("$latest/refs/heads");
1343 my @pr = stat("$latest/packed-refs");
1344 return if $c > ($hd[10] // 0) && $c > ($pr[10] // 0);
1348 my $pr = $opt->{-progress};
1349 my $seq = $opt->{sequential_shard};
1350 my $art_beg; # the NNTP article number we start xapian_only at
1351 my $idxlevel = $self->{ibx}->{indexlevel};
1352 local $self->{ibx}->{indexlevel} = 'basic' if $seq;
1354 $self->idx_init($opt); # acquire lock
1355 fill_alternates($self, $epoch_max);
1356 $self->{oidx}->rethread_prepare($opt);
1358 need_checkpoint => \(my $bool = 0),
1359 reindex => $opt->{reindex},
1362 ibx => $self->{ibx},
1363 epoch_max => $epoch_max,
1365 my $quit = PublicInbox::SearchIdx::quit_cb($sync);
1366 local $SIG{QUIT} = $quit;
1367 local $SIG{INT} = $quit;
1368 local $SIG{TERM} = $quit;
1370 if (sync_prepare($self, $sync)) {
1371 # tmp_clone seems to fail if inside a transaction, so
1372 # we rollback here (because we opened {mm} for reading)
1373 # Note: we do NOT rely on DBI transactions for atomicity;
1374 # only for batch performance.
1375 $self->{mm}->{dbh}->rollback;
1376 $self->{mm}->{dbh}->begin_work;
1378 $self->{mm}->tmp_clone($self->{ibx}->{inboxdir});
1380 # xapian_only works incrementally w/o --reindex
1381 if ($seq && !$opt->{reindex}) {
1382 $art_beg = $sync->{mm_tmp}->max || -1;
1386 # work forwards through history
1387 index_todo($self, $sync, $_) for @{delete($sync->{todo}) // []};
1388 $self->{oidx}->rethread_done($opt) unless $sync->{quit};
1391 if (my $nr = $sync->{nr}) {
1392 my $pr = $sync->{-opt}->{-progress};
1393 $pr->('all.git '.sprintf($sync->{-regen_fmt}, $$nr)) if $pr;
1397 # deal with Xapian shards sequentially
1398 if ($seq && delete($sync->{mm_tmp})) {
1399 if ($sync->{quit}) {
1402 $self->{ibx}->{indexlevel} = $idxlevel;
1403 xapian_only($self, $opt, $sync, $art_beg);
1404 $quit_warn = 1 if $sync->{quit};
1408 # --reindex on the command-line
1409 if (!$sync->{quit} && $opt->{reindex} &&
1410 !ref($opt->{reindex}) && $idxlevel ne 'basic') {
1411 $self->lock_acquire;
1412 my $s0 = PublicInbox::SearchIdx->new($self->{ibx}, 0, 0);
1413 if (my $xdb = $s0->idx_acquire) {
1414 my $n = $xdb->get_metadata('has_threadid');
1415 $xdb->set_metadata('has_threadid', '1') if $n ne '1';
1418 $self->lock_release;
1421 # reindex does not pick up new changes, so we rerun w/o it:
1422 if ($opt->{reindex} && !$sync->{quit}) {
1425 delete @again{qw(rethread reindex -skip_lock)};
1426 index_sync($self, \%again);
1427 $opt->{quit} = $again{quit}; # propagate to caller
1429 warn <<EOF if $quit_warn;
1430 W: interrupted, --xapian-only --reindex required upon restart