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 qw(log2stack crlf_adjust is_ancestor check_size);
22 use IO::Handle; # ->autoflush
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 # getconf(1) is POSIX, but *NPROCESSORS* vars are not
39 for (qw(_NPROCESSORS_ONLN NPROCESSORS_ONLN)) {
40 `getconf $_ 2>/dev/null` =~ /^(\d+)$/ and return $1;
42 for my $nproc (qw(nproc gnproc)) { # GNU coreutils nproc
43 `$nproc 2>/dev/null` =~ /^(\d+)$/ and return $1;
46 # should we bother with `sysctl hw.ncpu`? Those only give
47 # us total processor count, not online processor count.
51 sub nproc_shards ($) {
53 my $n = $creat_opt->{nproc} if ref($creat_opt) eq 'HASH';
56 # assume 2 cores if not detectable or zero
57 state $NPROC_DETECTED = detect_nproc() || 2;
59 $n = $NPROC_MAX_DEFAULT if $n > $NPROC_MAX_DEFAULT;
62 # subtract for the main process and git-fast-import
67 sub count_shards ($) {
70 my $xpfx = $self->{xpfx};
72 # always load existing shards in case core count changes:
73 # Also, shard count may change while -watch is running
74 # due to "xcpdb --reshard"
77 foreach my $shard (<$xpfx/*>) {
78 -d $shard && $shard =~ m!/[0-9]+\z! or next;
79 $XapianDatabase //= do {
80 require PublicInbox::Search;
81 PublicInbox::Search::load_xapian();
82 $PublicInbox::Search::X{Database};
85 $XapianDatabase->new($shard)->close;
94 # $creat may be any true value, or 0/undef. A hashref is true,
95 # and $creat->{nproc} may be set to an integer
96 my ($class, $v2ibx, $creat) = @_;
97 $v2ibx = PublicInbox::InboxWritable->new($v2ibx);
98 my $dir = $v2ibx->assert_usable_dir;
102 File::Path::mkpath($dir);
104 die "$dir does not exist\n";
107 $v2ibx->umask_prepare;
109 my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION;
112 im => undef, # PublicInbox::Import
118 over => PublicInbox::OverIdx->new("$xpfx/over.sqlite3"),
119 lock_path => "$dir/inbox.lock",
120 # limit each git repo (epoch) to 1GB or so
121 rotate_bytes => int((1024 * 1024 * 1024) / $PACKING_FACTOR),
122 last_commit => [], # git epoch -> commit
124 $self->{over}->{-no_fsync} = 1 if $v2ibx->{-no_fsync};
125 $self->{shards} = count_shards($self) || nproc_shards($creat);
131 my ($self, $shards, $skip_epoch, $skip_artnum) = @_;
132 if (defined $shards) {
133 $self->{parallel} = 0 if $shards == 0;
134 $self->{shards} = $shards if $shards > 0;
137 $self->{mm}->skip_artnum($skip_artnum) if defined $skip_artnum;
139 git_dir_latest($self, \$epoch_max);
140 if (defined $skip_epoch && $epoch_max == -1) {
141 $epoch_max = $skip_epoch;
143 $self->git_init($epoch_max >= 0 ? $epoch_max : 0);
147 # returns undef on duplicate or spam
148 # mimics Import::add and wraps it for v2
150 my ($self, $eml, $check_cb) = @_;
151 $self->{ibx}->with_umask(\&_add, $self, $eml, $check_cb);
154 # indexes a message, returns true if checkpointing is needed
156 my ($self, $msgref, $mime, $smsg) = @_;
157 $smsg->{bytes} = $smsg->{raw_bytes} + crlf_adjust($$msgref);
158 $self->{over}->add_overview($mime, $smsg);
159 my $idx = idx_shard($self, $smsg->{num} % $self->{shards});
160 $idx->index_raw($msgref, $mime, $smsg);
161 my $n = $self->{transact_bytes} += $smsg->{raw_bytes};
162 $n >= $self->{batch_bytes};
166 my ($self, $mime, $check_cb) = @_;
170 $mime = $check_cb->($mime, $self->{ibx}) or return;
173 # All pipes (> $^F) known to Perl 5.6+ have FD_CLOEXEC set,
174 # as does SQLite 3.4.1+ (released in 2007-07-20), and
175 # Xapian 1.3.2+ (released 2015-03-15).
176 # For the most part, we can spawn git-fast-import without
177 # leaking FDs to it...
180 my ($num, $mid0) = v2_num_for($self, $mime);
181 defined $num or return; # duplicate
182 defined $mid0 or die "BUG: \$mid0 undefined\n";
183 my $im = $self->importer;
184 my $smsg = bless { mid => $mid0, num => $num }, 'PublicInbox::Smsg';
185 my $cmt = $im->add($mime, undef, $smsg); # sets $smsg->{ds|ts|blob}
186 $cmt = $im->get_mark($cmt);
187 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
189 my $msgref = delete $smsg->{-raw_email};
190 if (do_idx($self, $msgref, $mime, $smsg)) {
198 my ($self, $mime) = @_;
199 my $mids = mids($mime);
201 my $mid = $mids->[0];
202 my $num = $self->{mm}->mid_insert($mid);
203 if (defined $num) { # common case
207 # crap, Message-ID is already known, hope somebody just resent:
208 foreach my $m (@$mids) {
209 # read-only lookup now safe to do after above barrier
210 # easy, don't store duplicates
211 # note: do not add more diagnostic info here since
212 # it gets noisy on public-inbox-watch restarts
213 return () if content_exists($self, $mime, $m);
216 # AltId may pre-populate article numbers (e.g. X-Mail-Count
217 # or NNTP article number), use that article number if it's
219 my $altid = $self->{ibx}->{altid};
220 if ($altid && grep(/:file=msgmap\.sqlite3\z/, @$altid)) {
221 my $num = $self->{mm}->num_for($mid);
223 if (defined $num && !$self->{over}->get_art($num)) {
229 warn "<$mid> reused for mismatched content\n";
231 # try the rest of the mids
232 for(my $i = $#$mids; $i >= 1; $i--) {
234 $num = $self->{mm}->mid_insert($m);
236 warn "alternative <$m> for <$mid> found\n";
241 # none of the existing Message-IDs are good, generate a new one:
242 v2_num_for_harder($self, $mime);
245 sub v2_num_for_harder {
246 my ($self, $eml) = @_;
248 my $dig = content_digest($eml);
249 my $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
250 my $num = $self->{mm}->mid_insert($mid0);
251 unless (defined $num) {
252 # it's hard to spoof the last Received: header
253 my @recvd = $eml->header_raw('Received');
254 $dig->add("Received: $_") foreach (@recvd);
255 $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
256 $num = $self->{mm}->mid_insert($mid0);
258 # fall back to a random Message-ID and give up determinism:
259 until (defined($num)) {
261 $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
262 warn "using random Message-ID <$mid0> as fallback\n";
263 $num = $self->{mm}->mid_insert($mid0);
266 PublicInbox::Import::append_mid($eml, $mid0);
271 my ($self, $shard_i) = @_;
272 $self->{idx_shards}->[$shard_i];
275 sub _idx_init { # with_umask callback
276 my ($self, $opt) = @_;
277 $self->lock_acquire unless $opt && $opt->{-skip_lock};
278 $self->{over}->create;
280 # xcpdb can change shard count while -watch is idle
281 my $nshards = count_shards($self);
282 $self->{shards} = $nshards if $nshards && $nshards != $self->{shards};
283 $self->{batch_bytes} = $opt->{batch_size} //
284 $PublicInbox::SearchIdx::BATCH_BYTES;
285 $self->{batch_bytes} *= $self->{shards} if $self->{parallel};
287 # need to create all shards before initializing msgmap FD
288 # idx_shards must be visible to all forked processes
289 my $max = $self->{shards} - 1;
290 my $idx = $self->{idx_shards} = [];
291 push @$idx, PublicInbox::SearchIdxShard->new($self, $_) for (0..$max);
293 # Now that all subprocesses are up, we can open the FDs
295 my $mm = $self->{mm} = PublicInbox::Msgmap->new_file(
296 "$self->{ibx}->{inboxdir}/msgmap.sqlite3",
297 $self->{ibx}->{-no_fsync} ? 2 : 1);
298 $mm->{dbh}->begin_work;
303 my ($self, $opt) = @_;
304 return if $self->{idx_shards};
305 my $ibx = $self->{ibx};
307 # do not leak read-only FDs to child processes, we only have these
308 # FDs for duplicate detection so they should not be
309 # frequently activated.
310 delete @$ibx{qw(mm search)};
313 $self->{parallel} = 0 if ($ibx->{indexlevel}//'') eq 'basic';
314 if ($self->{parallel}) {
315 pipe(my ($r, $w)) or die "pipe failed: $!";
316 # pipe for barrier notifications doesn't need to be big,
318 fcntl($w, 1031, 4096) if $^O eq 'linux';
319 $self->{bnote} = [ $r, $w ];
324 $ibx->with_umask(\&_idx_init, $self, $opt);
327 # returns an array mapping [ epoch => latest_commit ]
328 # latest_commit may be undef if nothing was done to that epoch
329 # $replace_map = { $object_id => $strref, ... }
330 sub _replace_oids ($$$) {
331 my ($self, $mime, $replace_map) = @_;
333 my $pfx = "$self->{ibx}->{inboxdir}/git";
334 my $rewrites = []; # epoch => commit
335 my $max = $self->{epoch_max};
337 unless (defined($max)) {
338 defined(my $latest = git_dir_latest($self, \$max)) or return;
339 $self->{epoch_max} = $max;
342 foreach my $i (0..$max) {
343 my $git_dir = "$pfx/$i.git";
345 my $git = PublicInbox::Git->new($git_dir);
346 my $im = $self->import_init($git, 0, 1);
347 $rewrites->[$i] = $im->replace_oids($mime, $replace_map);
353 sub content_hashes ($) {
355 my @chashes = ( content_hash($mime) );
357 # We still support Email::MIME, here, and
358 # Email::MIME->as_string doesn't always round-trip, so we may
359 # use a second content_hash
360 my $rt = content_hash(PublicInbox::Eml->new(\($mime->as_string)));
361 push @chashes, $rt if $chashes[0] ne $rt;
365 sub content_matches ($$) {
366 my ($chashes, $existing) = @_;
367 my $chash = content_hash($existing);
368 foreach (@$chashes) {
369 return 1 if $_ eq $chash
374 # used for removing or replacing (purging)
375 sub rewrite_internal ($$;$$$) {
376 my ($self, $old_eml, $cmt_msg, $new_eml, $sref) = @_;
378 my ($im, $need_reindex, $replace_map);
380 $replace_map = {}; # oid => sref
381 $need_reindex = [] if $new_eml;
383 $im = $self->importer;
385 my $over = $self->{over};
386 my $chashes = content_hashes($old_eml);
388 my $mids = mids($old_eml);
390 # We avoid introducing new blobs into git since the raw content
391 # can be slightly different, so we do not need the user-supplied
392 # message now that we have the mids and content_hash
396 foreach my $mid (@$mids) {
397 my %gone; # num => [ smsg, $mime, raw ]
399 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
400 my $msg = get_blob($self, $smsg);
401 if (!defined($msg)) {
402 warn "broken smsg for $mid\n";
406 my $cur = PublicInbox::Eml->new($msg);
407 if (content_matches($chashes, $cur)) {
408 $gone{$smsg->{num}} = [ $smsg, $cur, \$orig ];
411 my $n = scalar keys %gone;
414 warn "BUG: multiple articles linked to <$mid>\n",
415 join(',', sort keys %gone), "\n";
417 foreach my $num (keys %gone) {
418 my ($smsg, $mime, $orig) = @{$gone{$num}};
419 # $removed should only be set once assuming
420 # no bugs in our deduplication code:
421 $removed = [ undef, $mime, $smsg ];
422 my $oid = $smsg->{blob};
424 $replace_map->{$oid} = $sref;
426 ($mark, undef) = $im->remove($orig, $cmt_msg);
427 $removed->[0] = $mark;
430 if ($need_reindex) { # ->replace
431 push @$need_reindex, $smsg;
432 } else { # ->purge or ->remove
433 $self->{mm}->num_delete($num);
435 unindex_oid_remote($self, $oid, $mid);
440 my $cmt = $im->get_mark($mark);
441 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
443 if ($replace_map && scalar keys %$replace_map) {
444 my $rewrites = _replace_oids($self, $new_eml, $replace_map);
445 return { rewrites => $rewrites, need_reindex => $need_reindex };
447 defined($mark) ? $removed : undef;
450 # public (see PublicInbox::Import->remove), but note the 3rd element
451 # (retval[2]) is not part of the stable API shared with Import->remove
453 my ($self, $eml, $cmt_msg) = @_;
454 my $r = $self->{ibx}->with_umask(\&rewrite_internal,
455 $self, $eml, $cmt_msg);
456 defined($r) && defined($r->[0]) ? @$r: undef;
459 sub _replace ($$;$$) {
460 my ($self, $old_eml, $new_eml, $sref) = @_;
461 my $arg = [ $self, $old_eml, undef, $new_eml, $sref ];
462 my $rewritten = $self->{ibx}->with_umask(\&rewrite_internal,
463 $self, $old_eml, undef, $new_eml, $sref) or return;
465 my $rewrites = $rewritten->{rewrites};
466 # ->done is called if there are rewrites since we gc+prune from git
467 $self->idx_init if @$rewrites;
469 for my $i (0..$#$rewrites) {
470 defined(my $cmt = $rewrites->[$i]) or next;
471 $self->{last_commit}->[$i] = $cmt;
478 my ($self, $mime) = @_;
479 my $rewritten = _replace($self, $mime, undef, \'') or return;
480 $rewritten->{rewrites}
483 # returns the git object_id of $fh, does not write the object to FS
484 sub git_hash_raw ($$) {
485 my ($self, $raw) = @_;
486 # grab the expected OID we have to reindex:
487 pipe(my($in, $w)) or die "pipe: $!";
488 my $git_dir = $self->{ibx}->git->{git_dir};
489 my $cmd = ['git', "--git-dir=$git_dir", qw(hash-object --stdin)];
490 my $r = popen_rd($cmd, undef, { 0 => $in });
491 print $w $$raw or die "print \$w: $!";
492 close $w or die "close \$w: $!";
494 chomp(my $oid = <$r>);
495 close $r or die "git hash-object failed: $?";
496 $oid =~ /\A$OID\z/ or die "OID not expected: $oid";
500 sub _check_mids_match ($$$) {
501 my ($old_list, $new_list, $hdrs) = @_;
502 my %old_mids = map { $_ => 1 } @$old_list;
503 my %new_mids = map { $_ => 1 } @$new_list;
504 my @old = keys %old_mids;
505 my @new = keys %new_mids;
506 my $err = "$hdrs may not be changed when replacing\n";
507 die $err if scalar(@old) != scalar(@new);
508 delete @new_mids{@old};
509 delete @old_mids{@new};
510 die $err if (scalar(keys %old_mids) || scalar(keys %new_mids));
513 # Changing Message-IDs or References with ->replace isn't supported.
514 # The rules for dealing with messages with multiple or conflicting
515 # Message-IDs are pretty complex and rethreading hasn't been fully
517 sub check_mids_match ($$) {
518 my ($old, $new) = @_;
519 _check_mids_match(mids($old), mids($new), 'Message-ID(s)');
520 _check_mids_match(references($old), references($new),
521 'References/In-Reply-To');
526 my ($self, $old_mime, $new_mime) = @_;
528 check_mids_match($old_mime, $new_mime);
530 # mutt will always add Content-Length:, Status:, Lines: when editing
531 PublicInbox::Import::drop_unwanted_headers($new_mime);
533 my $raw = $new_mime->as_string;
534 my $expect_oid = git_hash_raw($self, \$raw);
535 my $rewritten = _replace($self, $old_mime, $new_mime, \$raw) or return;
536 my $need_reindex = $rewritten->{need_reindex};
538 # just in case we have bugs in deduplication code:
539 my $n = scalar(@$need_reindex);
541 my $list = join(', ', map {
542 "$_->{num}: <$_->{mid}>"
545 W: rewritten $n messages matching content of original message (expected: 1).
546 W: possible bug in public-inbox, NNTP article IDs and Message-IDs follow:
551 # make sure we really got the OID:
552 my ($blob, $type, $bytes) = $self->{ibx}->git->check($expect_oid);
553 $blob eq $expect_oid or die "BUG: $expect_oid not found after replace";
555 # don't leak FDs to Xapian:
556 $self->{ibx}->git->cleanup;
558 # reindex modified messages:
559 for my $smsg (@$need_reindex) {
560 my $new_smsg = bless {
565 }, 'PublicInbox::Smsg';
566 my $sync = { autime => $smsg->{ds}, cotime => $smsg->{ts} };
567 $new_smsg->populate($new_mime, $sync);
568 do_idx($self, \$raw, $new_mime, $new_smsg);
570 $rewritten->{rewrites};
573 sub last_epoch_commit ($$;$) {
574 my ($self, $i, $cmt) = @_;
575 my $v = PublicInbox::Search::SCHEMA_VERSION();
576 $self->{mm}->last_commit_xap($v, $i, $cmt);
579 sub set_last_commits ($) {
581 defined(my $epoch_max = $self->{epoch_max}) or return;
582 my $last_commit = $self->{last_commit};
583 foreach my $i (0..$epoch_max) {
584 defined(my $cmt = $last_commit->[$i]) or next;
585 $last_commit->[$i] = undef;
586 last_epoch_commit($self, $i, $cmt);
592 $self->{bnote} or return;
594 my $barrier = { map { $_ => 1 } (0..$n) };
598 my ($self, $barrier) = @_;
599 my $bnote = $self->{bnote} or return;
601 while (scalar keys %$barrier) {
602 defined(my $l = readline($r)) or die "EOF on barrier_wait: $!";
603 $l =~ /\Abarrier (\d+)/ or die "bad line on barrier_wait: $l";
604 delete $barrier->{$1} or die "bad shard[$1] on barrier wait";
609 sub checkpoint ($;$) {
610 my ($self, $wait) = @_;
612 if (my $im = $self->{im}) {
619 my $shards = $self->{idx_shards};
621 my $dbh = $self->{mm}->{dbh};
623 # SQLite msgmap data is second in importance
626 # SQLite overview is third
627 $self->{over}->commit_lazy;
629 # Now deal with Xapian
631 my $barrier = $self->barrier_init(scalar @$shards);
633 # each shard needs to issue a barrier command
634 $_->remote_barrier for @$shards;
636 # wait for each Xapian shard
637 $self->barrier_wait($barrier);
639 $_->remote_commit for @$shards;
642 # last_commit is special, don't commit these until
643 # remote shards are done:
645 set_last_commits($self);
650 $self->{total_bytes} += $self->{transact_bytes};
651 $self->{transact_bytes} = 0;
654 # issue a write barrier to ensure all data is visible to other processes
655 # and read-only ops. Order of data importance is: git > SQLite > Xapian
657 sub barrier { checkpoint($_[0], 1) };
663 if (my $im = delete $self->{im}) {
664 eval { $im->done }; # PublicInbox::Import::done
665 $err .= "import done: $@\n" if $@;
668 eval { checkpoint($self) };
669 $err .= "checkpoint: $@\n" if $@;
671 if (my $mm = delete $self->{mm}) {
672 my $m = $err ? 'rollback' : 'commit';
673 eval { $mm->{dbh}->$m };
674 $err .= "msgmap $m: $@\n" if $@;
676 my $shards = delete $self->{idx_shards};
679 eval { $_->remote_close };
680 $err .= "shard close: $@\n" if $@;
683 eval { $self->{over}->disconnect };
684 $err .= "over disconnect: $@\n" if $@;
685 delete $self->{bnote};
686 my $nbytes = $self->{total_bytes};
687 $self->{total_bytes} = 0;
688 $self->lock_release(!!$nbytes) if $shards;
689 $self->{ibx}->git->cleanup;
693 sub fill_alternates ($$) {
694 my ($self, $epoch) = @_;
696 my $pfx = "$self->{ibx}->{inboxdir}/git";
697 my $all = "$self->{ibx}->{inboxdir}/all.git";
698 PublicInbox::Import::init_bare($all) unless -d $all;
699 my $info_dir = "$all/objects/info";
700 my $alt = "$info_dir/alternates";
704 open(my $fh, '<', $alt) or die "open < $alt: $!\n";
705 $mode = (stat($fh))[2] & 07777;
707 # we assign a sort score to every alternate and favor
708 # the newest (highest numbered) one because loose objects
709 # require scanning epochs and only the latest epoch is
710 # expected to see loose objects
712 my $other = 0; # in case admin adds non-epoch repos
714 if (m!\A\Q../../\E([0-9]+)\.git/objects\z!) {
720 } split(/\n+/, do { local $/; <$fh> });
723 foreach my $i (0..$epoch) {
724 my $dir = "../../git/$i.git/objects";
725 if (!exists($alt{$dir}) && -d "$pfx/$i.git") {
732 my $fh = File::Temp->new(TEMPLATE => 'alt-XXXXXXXX', DIR => $info_dir);
733 my $tmp = $fh->filename;
734 print $fh join("\n", sort { $alt{$b} <=> $alt{$a} } keys %alt), "\n"
735 or die "print $tmp: $!\n";
736 chmod($mode, $fh) or die "fchmod $tmp: $!\n";
737 close $fh or die "close $tmp $!\n";
738 rename($tmp, $alt) or die "rename $tmp => $alt: $!\n";
739 $fh->unlink_on_destroy(0);
743 my ($self, $epoch) = @_;
744 my $git_dir = "$self->{ibx}->{inboxdir}/git/$epoch.git";
745 PublicInbox::Import::init_bare($git_dir);
746 my @cmd = (qw/git config/, "--file=$git_dir/config",
747 'include.path', '../../all.git/config');
748 PublicInbox::Import::run_die(\@cmd);
749 fill_alternates($self, $epoch);
754 my ($self, $max) = @_;
756 my $pfx = "$self->{ibx}->{inboxdir}/git";
757 return unless -d $pfx;
759 opendir my $dh, $pfx or die "opendir $pfx: $!\n";
760 while (defined(my $git_dir = readdir($dh))) {
761 $git_dir =~ m!\A([0-9]+)\.git\z! or next;
764 $latest = "$pfx/$git_dir";
772 my $im = $self->{im};
774 if ($im->{bytes_added} < $self->{rotate_bytes}) {
781 my $git_dir = $self->git_init(++$self->{epoch_max});
782 my $git = PublicInbox::Git->new($git_dir);
783 return $self->import_init($git, 0);
788 my $latest = git_dir_latest($self, \$max);
789 if (defined $latest) {
790 my $git = PublicInbox::Git->new($latest);
791 my $packed_bytes = $git->packed_bytes;
792 my $unpacked_bytes = $packed_bytes / $PACKING_FACTOR;
794 if ($unpacked_bytes >= $self->{rotate_bytes}) {
797 $self->{epoch_max} = $max;
798 return $self->import_init($git, $packed_bytes);
801 $self->{epoch_max} = $epoch;
802 $latest = $self->git_init($epoch);
803 $self->import_init(PublicInbox::Git->new($latest), 0);
807 my ($self, $git, $packed_bytes, $tmp) = @_;
808 my $im = PublicInbox::Import->new($git, undef, undef, $self->{ibx});
809 $im->{bytes_added} = int($packed_bytes / $PACKING_FACTOR);
810 $im->{lock_path} = undef;
811 $im->{path_type} = 'v2';
812 $self->{im} = $im unless $tmp;
818 my ($mid, $cur, $new) = @_;
820 my $ah = File::Temp->new(TEMPLATE => 'email-cur-XXXXXXXX', TMPDIR => 1);
821 print $ah $cur->as_string or die "print: $!";
822 $ah->flush or die "flush: $!";
823 PublicInbox::Import::drop_unwanted_headers($new);
824 my $bh = File::Temp->new(TEMPLATE => 'email-new-XXXXXXXX', TMPDIR => 1);
825 print $bh $new->as_string or die "print: $!";
826 $bh->flush or die "flush: $!";
827 my $cmd = [ qw(diff -u), $ah->filename, $bh->filename ];
828 print STDERR "# MID conflict <$mid>\n";
829 my $pid = spawn($cmd, undef, { 1 => 2 });
830 waitpid($pid, 0) == $pid or die "diff did not finish";
834 my ($self, $smsg) = @_;
835 if (my $im = $self->{im}) {
836 my $msg = $im->cat_blob($smsg->{blob});
839 # older message, should be in alternates
840 $self->{ibx}->msg_by_smsg($smsg);
843 sub content_exists ($$$) {
844 my ($self, $mime, $mid) = @_;
845 my $over = $self->{over};
846 my $chashes = content_hashes($mime);
848 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
849 my $msg = get_blob($self, $smsg);
850 if (!defined($msg)) {
851 warn "broken smsg for $mid\n";
854 my $cur = PublicInbox::Eml->new($msg);
855 return 1 if content_matches($chashes, $cur);
857 # XXX DEBUG_DIFF is experimental and may be removed
858 diff($mid, $cur, $mime) if $ENV{DEBUG_DIFF};
865 if (my $shards = $self->{idx_shards}) {
866 $_->atfork_child foreach @$shards;
868 if (my $im = $self->{im}) {
871 die "unexpected mm" if $self->{mm};
872 close $self->{bnote}->[0] or die "close bnote[0]: $!\n";
876 sub reindex_checkpoint ($$) {
877 my ($self, $sync) = @_;
879 $self->{ibx}->git->cleanup; # *async_wait
880 ${$sync->{need_checkpoint}} = 0;
881 my $mm_tmp = $sync->{mm_tmp};
882 $mm_tmp->atfork_prepare if $mm_tmp;
883 $self->done; # release lock
885 if (my $pr = $sync->{-opt}->{-progress}) {
886 $pr->(sprintf($sync->{-regen_fmt}, ${$sync->{nr}}));
889 # allow -watch or -mda to write...
890 $self->idx_init($sync->{-opt}); # reacquire lock
891 $mm_tmp->atfork_parent if $mm_tmp;
894 sub index_oid { # cat_async callback
895 my ($bref, $oid, $type, $size, $arg) = @_;
896 return if $size == 0; # purged
898 my $eml = PublicInbox::Eml->new($$bref);
899 my $mids = mids($eml);
900 my $chash = content_hash($eml);
901 my $self = $arg->{v2w};
903 if (scalar(@$mids) == 0) {
904 warn "E: $oid has no Message-ID, skipping\n";
908 # {unindexed} is unlikely
909 if ((my $unindexed = $arg->{unindexed}) && scalar(@$mids) == 1) {
910 $num = delete($unindexed->{$mids->[0]});
913 $self->{mm}->mid_set($num, $mid0);
914 delete($arg->{unindexed}) if !keys(%$unindexed);
917 if (!defined($num)) { # reuse if reindexing (or duplicates)
918 my $over = $self->{over};
919 for my $mid (@$mids) {
920 ($num, $mid0) = $over->num_mid0_for_oid($oid, $mid);
921 last if defined $num;
924 $mid0 //= do { # is this a number we got before?
925 $num = $arg->{mm_tmp}->num_for($mids->[0]);
926 defined($num) ? $mids->[0] : undef;
928 if (!defined($num)) {
929 for (my $i = $#$mids; $i >= 1; $i--) {
930 $num = $arg->{mm_tmp}->num_for($mids->[$i]);
938 $arg->{mm_tmp}->num_delete($num);
939 } else { # never seen
940 $num = $self->{mm}->mid_insert($mids->[0]);
943 } else { # rare, try the rest of them, backwards
944 for (my $i = $#$mids; $i >= 1; $i--) {
945 $num = $self->{mm}->mid_insert($mids->[$i]);
953 if (!defined($num)) {
954 warn "E: $oid <", join('> <', @$mids), "> is a duplicate\n";
963 }, 'PublicInbox::Smsg';
964 $smsg->populate($eml, $arg);
965 if (do_idx($self, $bref, $eml, $smsg)) {
966 ${$arg->{need_checkpoint}} = 1;
970 # only update last_commit for $i on reindex iff newer than current
971 sub update_last_commit ($$$$) {
972 my ($self, $git, $i, $cmt) = @_;
973 my $last = last_epoch_commit($self, $i);
974 if (defined $last && is_ancestor($git, $last, $cmt)) {
975 my @cmd = (qw(rev-list --count), "$last..$cmt");
976 chomp(my $n = $git->qx(@cmd));
977 return if $n ne '' && $n == 0;
979 last_epoch_commit($self, $i, $cmt);
982 sub git_dir_n ($$) { "$_[0]->{ibx}->{inboxdir}/git/$_[1].git" }
984 sub last_commits ($$) {
985 my ($self, $epoch_max) = @_;
987 for (my $i = $epoch_max; $i >= 0; $i--) {
988 $heads->[$i] = last_epoch_commit($self, $i);
993 # returns a revision range for git-log(1)
994 sub log_range ($$$$$) {
995 my ($self, $sync, $git, $i, $tip) = @_;
996 my $opt = $sync->{-opt};
997 my $pr = $opt->{-progress} if (($opt->{verbose} || 0) > 1);
998 my $cur = $sync->{ranges}->[$i] or do {
999 $pr->("$i.git indexing all of $tip") if $pr;
1000 return $tip; # all of it
1003 # fast equality check to avoid (v)fork+execve overhead
1005 $sync->{ranges}->[$i] = undef;
1009 my $range = "$cur..$tip";
1010 $pr->("$i.git checking contiguity... ") if $pr;
1011 if (is_ancestor($git, $cur, $tip)) { # common case
1012 $pr->("OK\n") if $pr;
1013 my $n = $git->qx(qw(rev-list --count), $range);
1016 $sync->{ranges}->[$i] = undef;
1017 $pr->("$i.git has nothing new\n") if $pr;
1018 return; # nothing to do
1020 $pr->("$i.git has $n changes since $cur\n") if $pr;
1022 $pr->("FAIL\n") if $pr;
1024 discontiguous range: $range
1025 Rewritten history? (in $git->{git_dir})
1027 chomp(my $base = $git->qx('merge-base', $tip, $cur));
1029 $range = "$base..$tip";
1030 warn "found merge-base: $base\n"
1033 warn "discarding history at $cur\n";
1036 reindexing $git->{git_dir} starting at
1039 $sync->{unindex_range}->{$i} = "$base..$cur";
1044 sub sync_prepare ($$$) {
1045 my ($self, $sync, $epoch_max) = @_;
1046 my $pr = $sync->{-opt}->{-progress};
1048 my $head = $self->{ibx}->{ref_head} || 'refs/heads/master';
1050 # reindex stops at the current heads and we later rerun index_sync
1052 my $reindex_heads = last_commits($self, $epoch_max) if $sync->{reindex};
1054 for (my $i = $epoch_max; $i >= 0; $i--) {
1055 my $git_dir = git_dir_n($self, $i);
1056 -d $git_dir or next; # missing epochs are fine
1057 my $git = PublicInbox::Git->new($git_dir);
1058 if ($reindex_heads) {
1059 $head = $reindex_heads->[$i] or next;
1061 chomp(my $tip = $git->qx(qw(rev-parse -q --verify), $head));
1063 next if $?; # new repo
1064 my $range = log_range($self, $sync, $git, $i, $tip) or next;
1065 # can't use 'rev-list --count' if we use --diff-filter
1066 $pr->("$i.git counting $range ... ") if $pr;
1067 # Don't bump num_highwater on --reindex by using {D}.
1068 # We intentionally do NOT use {D} in the non-reindex case
1069 # because we want NNTP article number gaps from unindexed
1070 # messages to show up in mirrors, too.
1071 $sync->{D} //= $sync->{reindex} ? {} : undef; # OID_BIN => NR
1072 my $stk = log2stack($sync, $git, $range, $self->{ibx});
1073 my $nr = $stk ? $stk->num_records : 0;
1074 $pr->("$nr\n") if $pr;
1075 $sync->{stacks}->[$i] = $stk if $stk;
1079 # XXX this should not happen unless somebody bypasses checks in
1080 # our code and blindly injects "d" file history into git repos
1081 if (my @leftovers = keys %{delete($sync->{D}) // {}}) {
1082 warn('W: unindexing '.scalar(@leftovers)." leftovers\n");
1083 my $arg = { v2w => $self };
1084 my $all = $self->{ibx}->git;
1085 for my $oid (@leftovers) {
1086 $oid = unpack('H*', $oid);
1087 $self->{current_info} = "leftover $oid";
1088 $all->cat_async($oid, \&unindex_oid, $arg);
1090 $all->cat_async_wait;
1092 if (!$regen_max && !keys(%{$self->{unindex_range}})) {
1093 $sync->{-regen_fmt} = "%u/?\n";
1097 # reindex should NOT see new commits anymore, if we do,
1098 # it's a problem and we need to notice it via die()
1099 my $pad = length($regen_max) + 1;
1100 $sync->{-regen_fmt} = "% ${pad}u/$regen_max\n";
1101 $sync->{nr} = \(my $nr = 0);
1102 return -1 if $sync->{reindex};
1103 $regen_max + $self->{mm}->num_highwater() || 0;
1106 sub unindex_oid_remote ($$$) {
1107 my ($self, $oid, $mid) = @_;
1108 my @removed = $self->{over}->remove_oid($oid, $mid);
1109 for my $num (@removed) {
1110 my $idx = idx_shard($self, $num % $self->{shards});
1111 $idx->remote_remove($oid, $num);
1115 sub unindex_oid ($$;$) { # git->cat_async callback
1116 my ($bref, $oid, $type, $size, $sync) = @_;
1117 my $self = $sync->{v2w};
1118 my $unindexed = $sync->{in_unindex} ? $sync->{unindexed} : undef;
1119 my $mm = $self->{mm};
1120 my $mids = mids(PublicInbox::Eml->new($bref));
1122 my $over = $self->{over};
1123 foreach my $mid (@$mids) {
1126 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
1127 $gone{$smsg->{num}} = 1 if $oid eq $smsg->{blob};
1129 my $n = scalar(keys(%gone)) or next;
1131 warn "BUG: multiple articles linked to $oid\n",
1132 join(',',sort keys %gone), "\n";
1134 foreach my $num (keys %gone) {
1136 my $mid0 = $mm->mid_for($num);
1137 $unindexed->{$mid0} = $num;
1139 $mm->num_delete($num);
1141 unindex_oid_remote($self, $oid, $mid);
1145 # this is rare, it only happens when we get discontiguous history in
1146 # a mirror because the source used -purge or -edit
1147 sub unindex ($$$$) {
1148 my ($self, $sync, $git, $unindex_range) = @_;
1149 my $unindexed = $sync->{unindexed} //= {}; # $mid0 => $num
1150 my $before = scalar keys %$unindexed;
1151 # order does not matter, here:
1152 my @cmd = qw(log --raw -r
1153 --no-notes --no-color --no-abbrev --no-renames);
1154 my $fh = $git->popen(@cmd, $unindex_range);
1155 my $all = $self->{ibx}->git;
1156 local $sync->{in_unindex} = 1;
1158 /\A:\d{6} 100644 $OID ($OID) [AM]\tm$/o or next;
1159 $all->cat_async($1, \&unindex_oid, $sync);
1161 close $fh or die "git log failed: \$?=$?";
1162 $all->cat_async_wait;
1164 return unless $sync->{-opt}->{prune};
1165 my $after = scalar keys %$unindexed;
1166 return if $before == $after;
1168 # ensure any blob can not longer be accessed via dumb HTTP
1169 PublicInbox::Import::run_die(['git', "--git-dir=$git->{git_dir}",
1170 qw(-c gc.reflogExpire=now gc --prune=all --quiet)]);
1173 sub sync_ranges ($$$) {
1174 my ($self, $sync, $epoch_max) = @_;
1175 my $reindex = $sync->{reindex};
1177 return last_commits($self, $epoch_max) unless $reindex;
1178 return [] if ref($reindex) ne 'HASH';
1180 my $ranges = $reindex->{from}; # arrayref;
1181 if (ref($ranges) ne 'ARRAY') {
1182 die 'BUG: $reindex->{from} not an ARRAY';
1187 sub index_xap_only { # git->cat_async callback
1188 my ($bref, $oid, $type, $size, $smsg) = @_;
1189 my $self = $smsg->{v2w};
1190 my $idx = idx_shard($self, $smsg->{num} % $self->{shards});
1191 $smsg->{raw_bytes} = $size;
1192 $idx->index_raw($bref, undef, $smsg);
1193 $self->{transact_bytes} += $size;
1196 sub index_xap_step ($$$;$) {
1197 my ($self, $sync, $beg, $step) = @_;
1198 my $end = $sync->{art_end};
1199 return if $beg > $end; # nothing to do
1201 $step //= $self->{shards};
1202 my $ibx = $self->{ibx};
1203 if (my $pr = $sync->{-opt}->{-progress}) {
1204 $pr->("Xapian indexlevel=$ibx->{indexlevel} ".
1205 "$beg..$end (% $step)\n");
1207 for (my $num = $beg; $num <= $end; $num += $step) {
1208 my $smsg = $ibx->over->get_art($num) or next;
1209 $smsg->{v2w} = $self;
1210 $ibx->git->cat_async($smsg->{blob}, \&index_xap_only, $smsg);
1211 if ($self->{transact_bytes} >= $self->{batch_bytes}) {
1212 ${$sync->{nr}} = $num;
1213 reindex_checkpoint($self, $sync);
1218 sub index_epoch ($$$) {
1219 my ($self, $sync, $i) = @_;
1221 my $git_dir = git_dir_n($self, $i);
1222 -d $git_dir or return; # missing epochs are fine
1223 my $git = PublicInbox::Git->new($git_dir);
1224 if (my $unindex_range = delete $sync->{unindex_range}->{$i}) { # rare
1225 unindex($self, $sync, $git, $unindex_range);
1227 defined(my $stk = $sync->{stacks}->[$i]) or return;
1228 $sync->{stacks}->[$i] = undef;
1229 my $all = $self->{ibx}->git;
1230 while (my ($f, $at, $ct, $oid) = $stk->pop_rec) {
1231 $self->{current_info} = "$i.git $oid";
1233 my $arg = { %$sync, autime => $at, cotime => $ct };
1234 if ($sync->{max_size}) {
1235 $all->check_async($oid, \&check_size, $arg);
1237 $all->cat_async($oid, \&index_oid, $arg);
1239 } elsif ($f eq 'd') {
1240 $all->cat_async($oid, \&unindex_oid, $sync);
1242 if (${$sync->{need_checkpoint}}) {
1243 reindex_checkpoint($self, $sync);
1246 $all->check_async_wait;
1247 $all->cat_async_wait;
1248 update_last_commit($self, $git, $i, $stk->{latest_cmt});
1252 my ($self, $opt, $sync, $art_beg) = @_;
1253 my $seq = $opt->{sequential_shard};
1255 local $self->{parallel} = 0 if $seq;
1256 $self->idx_init($opt); # acquire lock
1257 if (my $art_end = $self->{ibx}->mm->max) {
1259 need_checkpoint => \(my $bool = 0),
1262 nr => \(my $nr = 0),
1263 -regen_fmt => "%u/?\n",
1265 $sync->{art_end} = $art_end;
1266 if ($seq || !$self->{parallel}) {
1267 my $shard_end = $self->{shards} - 1;
1268 for (0..$shard_end) {
1269 index_xap_step($self, $sync, $art_beg + $_)
1271 } else { # parallel (maybe)
1272 index_xap_step($self, $sync, $art_beg, 1);
1275 $self->{ibx}->git->cat_async_wait;
1279 # public, called by public-inbox-index
1281 my ($self, $opt) = @_;
1282 $opt //= $_[1] //= {};
1283 goto \&xapian_only if $opt->{xapian_only};
1285 my $pr = $opt->{-progress};
1287 my $latest = git_dir_latest($self, \$epoch_max);
1288 return unless defined $latest;
1290 my $seq = $opt->{sequential_shard};
1291 my $art_beg; # the NNTP article number we start xapian_only at
1292 my $idxlevel = $self->{ibx}->{indexlevel};
1293 local $self->{ibx}->{indexlevel} = 'basic' if $seq;
1295 $self->idx_init($opt); # acquire lock
1296 fill_alternates($self, $epoch_max);
1297 $self->{over}->rethread_prepare($opt);
1299 need_checkpoint => \(my $bool = 0),
1300 unindex_range => {}, # EPOCH => oid_old..oid_new
1301 reindex => $opt->{reindex},
1305 $sync->{ranges} = sync_ranges($self, $sync, $epoch_max);
1306 if (sync_prepare($self, $sync, $epoch_max)) {
1307 # tmp_clone seems to fail if inside a transaction, so
1308 # we rollback here (because we opened {mm} for reading)
1309 # Note: we do NOT rely on DBI transactions for atomicity;
1310 # only for batch performance.
1311 $self->{mm}->{dbh}->rollback;
1312 $self->{mm}->{dbh}->begin_work;
1314 $self->{mm}->tmp_clone($self->{ibx}->{inboxdir});
1316 # xapian_only works incrementally w/o --reindex
1317 if ($seq && !$opt->{reindex}) {
1318 $art_beg = $sync->{mm_tmp}->max;
1319 $art_beg++ if defined($art_beg);
1322 if ($sync->{max_size} = $opt->{max_size}) {
1323 $sync->{index_oid} = \&index_oid;
1325 # work forwards through history
1326 index_epoch($self, $sync, $_) for (0..$epoch_max);
1327 $self->{over}->rethread_done($opt);
1330 if (my $nr = $sync->{nr}) {
1331 my $pr = $sync->{-opt}->{-progress};
1332 $pr->('all.git '.sprintf($sync->{-regen_fmt}, $$nr)) if $pr;
1335 # deal with Xapian shards sequentially
1336 if ($seq && delete($sync->{mm_tmp})) {
1337 $self->{ibx}->{indexlevel} = $idxlevel;
1338 xapian_only($self, $opt, $sync, $art_beg);
1341 # reindex does not pick up new changes, so we rerun w/o it:
1342 if ($opt->{reindex}) {
1345 delete @again{qw(rethread reindex -skip_lock)};
1346 index_sync($self, \%again);