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;
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);
20 use PublicInbox::SearchIdx qw(log2stack crlf_adjust is_ancestor check_size);
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 my $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 # always load existing shards in case core count changes:
69 # Also, shard count may change while -watch is running
70 my $srch = $self->{ibx}->search or return 0;
71 delete $self->{ibx}->{search};
76 # $creat may be any true value, or 0/undef. A hashref is true,
77 # and $creat->{nproc} may be set to an integer
78 my ($class, $v2ibx, $creat) = @_;
79 $v2ibx = PublicInbox::InboxWritable->new($v2ibx);
80 my $dir = $v2ibx->assert_usable_dir;
84 File::Path::mkpath($dir);
86 die "$dir does not exist\n";
89 $v2ibx->umask_prepare;
91 my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION;
94 im => undef, # PublicInbox::Import
100 oidx => PublicInbox::OverIdx->new("$xpfx/over.sqlite3"),
101 lock_path => "$dir/inbox.lock",
102 # limit each git repo (epoch) to 1GB or so
103 rotate_bytes => int((1024 * 1024 * 1024) / $PACKING_FACTOR),
104 last_commit => [], # git epoch -> commit
106 $self->{oidx}->{-no_fsync} = 1 if $v2ibx->{-no_fsync};
107 $self->{shards} = count_shards($self) || nproc_shards($creat);
113 my ($self, $shards, $skip_epoch, $skip_artnum) = @_;
114 if (defined $shards) {
115 $self->{parallel} = 0 if $shards == 0;
116 $self->{shards} = $shards if $shards > 0;
119 $self->{mm}->skip_artnum($skip_artnum) if defined $skip_artnum;
121 git_dir_latest($self, \$epoch_max);
122 if (defined $skip_epoch && $epoch_max == -1) {
123 $epoch_max = $skip_epoch;
125 $self->git_init($epoch_max >= 0 ? $epoch_max : 0);
129 # returns undef on duplicate or spam
130 # mimics Import::add and wraps it for v2
132 my ($self, $eml, $check_cb) = @_;
133 $self->{ibx}->with_umask(\&_add, $self, $eml, $check_cb);
137 my ($self, $num) = @_;
138 $self->{idx_shards}->[$num % scalar(@{$self->{idx_shards}})];
141 # indexes a message, returns true if checkpointing is needed
143 my ($self, $msgref, $mime, $smsg) = @_;
144 $smsg->{bytes} = $smsg->{raw_bytes} + crlf_adjust($$msgref);
145 $self->{oidx}->add_overview($mime, $smsg);
146 my $idx = idx_shard($self, $smsg->{num});
147 $idx->index_raw($msgref, $mime, $smsg);
148 my $n = $self->{transact_bytes} += $smsg->{raw_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 my $msgref = delete $smsg->{-raw_email};
177 if (do_idx($self, $msgref, $mime, $smsg)) {
185 my ($self, $mime) = @_;
186 my $mids = mids($mime);
188 my $mid = $mids->[0];
189 my $num = $self->{mm}->mid_insert($mid);
190 if (defined $num) { # common case
194 # crap, Message-ID is already known, hope somebody just resent:
195 foreach my $m (@$mids) {
196 # read-only lookup now safe to do after above barrier
197 # easy, don't store duplicates
198 # note: do not add more diagnostic info here since
199 # it gets noisy on public-inbox-watch restarts
200 return () if content_exists($self, $mime, $m);
203 # AltId may pre-populate article numbers (e.g. X-Mail-Count
204 # or NNTP article number), use that article number if it's
206 my $altid = $self->{ibx}->{altid};
207 if ($altid && grep(/:file=msgmap\.sqlite3\z/, @$altid)) {
208 my $num = $self->{mm}->num_for($mid);
210 if (defined $num && !$self->{oidx}->get_art($num)) {
216 warn "<$mid> reused for mismatched content\n";
218 # try the rest of the mids
219 for(my $i = $#$mids; $i >= 1; $i--) {
221 $num = $self->{mm}->mid_insert($m);
223 warn "alternative <$m> for <$mid> found\n";
228 # none of the existing Message-IDs are good, generate a new one:
229 v2_num_for_harder($self, $mime);
232 sub v2_num_for_harder {
233 my ($self, $eml) = @_;
235 my $dig = content_digest($eml);
236 my $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
237 my $num = $self->{mm}->mid_insert($mid0);
238 unless (defined $num) {
239 # it's hard to spoof the last Received: header
240 my @recvd = $eml->header_raw('Received');
241 $dig->add("Received: $_") foreach (@recvd);
242 $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
243 $num = $self->{mm}->mid_insert($mid0);
245 # fall back to a random Message-ID and give up determinism:
246 until (defined($num)) {
248 $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
249 warn "using random Message-ID <$mid0> as fallback\n";
250 $num = $self->{mm}->mid_insert($mid0);
253 PublicInbox::Import::append_mid($eml, $mid0);
257 sub _idx_init { # with_umask callback
258 my ($self, $opt) = @_;
259 $self->lock_acquire unless $opt && $opt->{-skip_lock};
260 $self->{oidx}->create;
262 # xcpdb can change shard count while -watch is idle
263 my $nshards = count_shards($self);
264 $self->{shards} = $nshards if $nshards && $nshards != $self->{shards};
265 $self->{batch_bytes} = $opt->{batch_size} //
266 $PublicInbox::SearchIdx::BATCH_BYTES;
267 $self->{batch_bytes} *= $self->{shards} if $self->{parallel};
269 # need to create all shards before initializing msgmap FD
270 # idx_shards must be visible to all forked processes
271 my $max = $self->{shards} - 1;
272 my $idx = $self->{idx_shards} = [];
273 push @$idx, PublicInbox::SearchIdxShard->new($self, $_) for (0..$max);
274 my $ibx = $self->{ibx} or return; # ExtIdxSearch
276 # Now that all subprocesses are up, we can open the FDs
278 my $mm = $self->{mm} = PublicInbox::Msgmap->new_file(
279 "$ibx->{inboxdir}/msgmap.sqlite3",
280 $ibx->{-no_fsync} ? 2 : 1);
281 $mm->{dbh}->begin_work;
284 sub parallel_init ($$) {
285 my ($self, $indexlevel) = @_;
286 if (($indexlevel // 'full') eq 'basic') {
287 $self->{parallel} = 0;
289 pipe(my ($r, $w)) or die "pipe failed: $!";
290 # pipe for barrier notifications doesn't need to be big,
292 fcntl($w, 1031, 4096) if $^O eq 'linux';
293 $self->{bnote} = [ $r, $w ];
300 my ($self, $opt) = @_;
301 return if $self->{idx_shards};
302 my $ibx = $self->{ibx};
304 # do not leak read-only FDs to child processes, we only have these
305 # FDs for duplicate detection so they should not be
306 # frequently activated.
307 delete @$ibx{qw(mm search)};
310 parallel_init($self, $ibx->{indexlevel});
312 $ibx->with_umask(\&_idx_init, $self, $opt);
315 # returns an array mapping [ epoch => latest_commit ]
316 # latest_commit may be undef if nothing was done to that epoch
317 # $replace_map = { $object_id => $strref, ... }
318 sub _replace_oids ($$$) {
319 my ($self, $mime, $replace_map) = @_;
321 my $pfx = "$self->{ibx}->{inboxdir}/git";
322 my $rewrites = []; # epoch => commit
323 my $max = $self->{epoch_max};
325 unless (defined($max)) {
326 defined(my $latest = git_dir_latest($self, \$max)) or return;
327 $self->{epoch_max} = $max;
330 foreach my $i (0..$max) {
331 my $git_dir = "$pfx/$i.git";
333 my $git = PublicInbox::Git->new($git_dir);
334 my $im = $self->import_init($git, 0, 1);
335 $rewrites->[$i] = $im->replace_oids($mime, $replace_map);
341 sub content_hashes ($) {
343 my @chashes = ( content_hash($mime) );
345 # We still support Email::MIME, here, and
346 # Email::MIME->as_string doesn't always round-trip, so we may
347 # use a second content_hash
348 my $rt = content_hash(PublicInbox::Eml->new(\($mime->as_string)));
349 push @chashes, $rt if $chashes[0] ne $rt;
353 sub content_matches ($$) {
354 my ($chashes, $existing) = @_;
355 my $chash = content_hash($existing);
356 foreach (@$chashes) {
357 return 1 if $_ eq $chash
362 # used for removing or replacing (purging)
363 sub rewrite_internal ($$;$$$) {
364 my ($self, $old_eml, $cmt_msg, $new_eml, $sref) = @_;
366 my ($im, $need_reindex, $replace_map);
368 $replace_map = {}; # oid => sref
369 $need_reindex = [] if $new_eml;
371 $im = $self->importer;
373 my $oidx = $self->{oidx};
374 my $chashes = content_hashes($old_eml);
376 my $mids = mids($old_eml);
378 # We avoid introducing new blobs into git since the raw content
379 # can be slightly different, so we do not need the user-supplied
380 # message now that we have the mids and content_hash
384 foreach my $mid (@$mids) {
385 my %gone; # num => [ smsg, $mime, raw ]
387 while (my $smsg = $oidx->next_by_mid($mid, \$id, \$prev)) {
388 my $msg = get_blob($self, $smsg);
389 if (!defined($msg)) {
390 warn "broken smsg for $mid\n";
394 my $cur = PublicInbox::Eml->new($msg);
395 if (content_matches($chashes, $cur)) {
396 $gone{$smsg->{num}} = [ $smsg, $cur, \$orig ];
399 my $n = scalar keys %gone;
402 warn "BUG: multiple articles linked to <$mid>\n",
403 join(',', sort keys %gone), "\n";
405 foreach my $num (keys %gone) {
406 my ($smsg, $mime, $orig) = @{$gone{$num}};
407 # $removed should only be set once assuming
408 # no bugs in our deduplication code:
409 $removed = [ undef, $mime, $smsg ];
410 my $oid = $smsg->{blob};
412 $replace_map->{$oid} = $sref;
414 ($mark, undef) = $im->remove($orig, $cmt_msg);
415 $removed->[0] = $mark;
418 if ($need_reindex) { # ->replace
419 push @$need_reindex, $smsg;
420 } else { # ->purge or ->remove
421 $self->{mm}->num_delete($num);
423 unindex_oid_remote($self, $oid, $mid);
428 my $cmt = $im->get_mark($mark);
429 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
431 if ($replace_map && scalar keys %$replace_map) {
432 my $rewrites = _replace_oids($self, $new_eml, $replace_map);
433 return { rewrites => $rewrites, need_reindex => $need_reindex };
435 defined($mark) ? $removed : undef;
438 # public (see PublicInbox::Import->remove), but note the 3rd element
439 # (retval[2]) is not part of the stable API shared with Import->remove
441 my ($self, $eml, $cmt_msg) = @_;
442 my $r = $self->{ibx}->with_umask(\&rewrite_internal,
443 $self, $eml, $cmt_msg);
444 defined($r) && defined($r->[0]) ? @$r: undef;
447 sub _replace ($$;$$) {
448 my ($self, $old_eml, $new_eml, $sref) = @_;
449 my $arg = [ $self, $old_eml, undef, $new_eml, $sref ];
450 my $rewritten = $self->{ibx}->with_umask(\&rewrite_internal,
451 $self, $old_eml, undef, $new_eml, $sref) or return;
453 my $rewrites = $rewritten->{rewrites};
454 # ->done is called if there are rewrites since we gc+prune from git
455 $self->idx_init if @$rewrites;
457 for my $i (0..$#$rewrites) {
458 defined(my $cmt = $rewrites->[$i]) or next;
459 $self->{last_commit}->[$i] = $cmt;
466 my ($self, $mime) = @_;
467 my $rewritten = _replace($self, $mime, undef, \'') or return;
468 $rewritten->{rewrites}
471 # returns the git object_id of $fh, does not write the object to FS
472 sub git_hash_raw ($$) {
473 my ($self, $raw) = @_;
474 # grab the expected OID we have to reindex:
475 pipe(my($in, $w)) or die "pipe: $!";
476 my $git_dir = $self->git->{git_dir};
477 my $cmd = ['git', "--git-dir=$git_dir", qw(hash-object --stdin)];
478 my $r = popen_rd($cmd, undef, { 0 => $in });
479 print $w $$raw or die "print \$w: $!";
480 close $w or die "close \$w: $!";
482 chomp(my $oid = <$r>);
483 close $r or die "git hash-object failed: $?";
484 $oid =~ /\A$OID\z/ or die "OID not expected: $oid";
488 sub _check_mids_match ($$$) {
489 my ($old_list, $new_list, $hdrs) = @_;
490 my %old_mids = map { $_ => 1 } @$old_list;
491 my %new_mids = map { $_ => 1 } @$new_list;
492 my @old = keys %old_mids;
493 my @new = keys %new_mids;
494 my $err = "$hdrs may not be changed when replacing\n";
495 die $err if scalar(@old) != scalar(@new);
496 delete @new_mids{@old};
497 delete @old_mids{@new};
498 die $err if (scalar(keys %old_mids) || scalar(keys %new_mids));
501 # Changing Message-IDs or References with ->replace isn't supported.
502 # The rules for dealing with messages with multiple or conflicting
503 # Message-IDs are pretty complex and rethreading hasn't been fully
505 sub check_mids_match ($$) {
506 my ($old, $new) = @_;
507 _check_mids_match(mids($old), mids($new), 'Message-ID(s)');
508 _check_mids_match(references($old), references($new),
509 'References/In-Reply-To');
514 my ($self, $old_mime, $new_mime) = @_;
516 check_mids_match($old_mime, $new_mime);
518 # mutt will always add Content-Length:, Status:, Lines: when editing
519 PublicInbox::Import::drop_unwanted_headers($new_mime);
521 my $raw = $new_mime->as_string;
522 my $expect_oid = git_hash_raw($self, \$raw);
523 my $rewritten = _replace($self, $old_mime, $new_mime, \$raw) or return;
524 my $need_reindex = $rewritten->{need_reindex};
526 # just in case we have bugs in deduplication code:
527 my $n = scalar(@$need_reindex);
529 my $list = join(', ', map {
530 "$_->{num}: <$_->{mid}>"
533 W: rewritten $n messages matching content of original message (expected: 1).
534 W: possible bug in public-inbox, NNTP article IDs and Message-IDs follow:
539 # make sure we really got the OID:
540 my ($blob, $type, $bytes) = $self->git->check($expect_oid);
541 $blob eq $expect_oid or die "BUG: $expect_oid not found after replace";
543 # don't leak FDs to Xapian:
546 # reindex modified messages:
547 for my $smsg (@$need_reindex) {
548 my $new_smsg = bless {
553 }, 'PublicInbox::Smsg';
554 my $sync = { autime => $smsg->{ds}, cotime => $smsg->{ts} };
555 $new_smsg->populate($new_mime, $sync);
556 do_idx($self, \$raw, $new_mime, $new_smsg);
558 $rewritten->{rewrites};
561 sub last_epoch_commit ($$;$) {
562 my ($self, $i, $cmt) = @_;
563 my $v = PublicInbox::Search::SCHEMA_VERSION();
564 $self->{mm}->last_commit_xap($v, $i, $cmt);
567 sub set_last_commits ($) {
569 defined(my $epoch_max = $self->{epoch_max}) or return;
570 my $last_commit = $self->{last_commit};
571 foreach my $i (0..$epoch_max) {
572 defined(my $cmt = $last_commit->[$i]) or next;
573 $last_commit->[$i] = undef;
574 last_epoch_commit($self, $i, $cmt);
580 $self->{bnote} or return;
582 my $barrier = { map { $_ => 1 } (0..$n) };
586 my ($self, $barrier) = @_;
587 my $bnote = $self->{bnote} or return;
589 while (scalar keys %$barrier) {
590 defined(my $l = readline($r)) or die "EOF on barrier_wait: $!";
591 $l =~ /\Abarrier (\d+)/ or die "bad line on barrier_wait: $l";
592 delete $barrier->{$1} or die "bad shard[$1] on barrier wait";
597 sub checkpoint ($;$) {
598 my ($self, $wait) = @_;
600 if (my $im = $self->{im}) {
607 my $shards = $self->{idx_shards};
609 my $dbh = $self->{mm}->{dbh};
611 # SQLite msgmap data is second in importance
614 # SQLite overview is third
615 $self->{oidx}->commit_lazy;
617 # Now deal with Xapian
619 my $barrier = $self->barrier_init(scalar @$shards);
621 # each shard needs to issue a barrier command
622 $_->shard_barrier for @$shards;
624 # wait for each Xapian shard
625 $self->barrier_wait($barrier);
627 $_->shard_commit for @$shards;
630 # last_commit is special, don't commit these until
631 # remote shards are done:
633 set_last_commits($self);
638 $self->{total_bytes} += $self->{transact_bytes};
639 $self->{transact_bytes} = 0;
642 # issue a write barrier to ensure all data is visible to other processes
643 # and read-only ops. Order of data importance is: git > SQLite > Xapian
645 sub barrier { checkpoint($_[0], 1) };
647 # true if locked and active
648 sub active { !!$_[0]->{im} }
654 if (my $im = delete $self->{im}) {
655 eval { $im->done }; # PublicInbox::Import::done
656 $err .= "import done: $@\n" if $@;
659 eval { checkpoint($self) };
660 $err .= "checkpoint: $@\n" if $@;
662 if (my $mm = delete $self->{mm}) {
663 my $m = $err ? 'rollback' : 'commit';
664 eval { $mm->{dbh}->$m };
665 $err .= "msgmap $m: $@\n" if $@;
667 my $shards = delete $self->{idx_shards};
670 eval { $_->shard_close };
671 $err .= "shard close: $@\n" if $@;
674 eval { $self->{oidx}->dbh_close };
675 $err .= "over close: $@\n" if $@;
676 delete $self->{bnote};
677 my $nbytes = $self->{total_bytes};
678 $self->{total_bytes} = 0;
679 $self->lock_release(!!$nbytes) if $shards;
684 sub write_alternates ($$$) {
685 my ($info_dir, $mode, $out) = @_;
686 my $fh = File::Temp->new(TEMPLATE => 'alt-XXXXXXXX', DIR => $info_dir);
687 my $tmp = $fh->filename;
688 print $fh @$out or die "print $tmp: $!\n";
689 chmod($mode, $fh) or die "fchmod $tmp: $!\n";
690 close $fh or die "close $tmp $!\n";
691 my $alt = "$info_dir/alternates";
692 rename($tmp, $alt) or die "rename $tmp => $alt: $!\n";
693 $fh->unlink_on_destroy(0);
696 sub fill_alternates ($$) {
697 my ($self, $epoch) = @_;
699 my $pfx = "$self->{ibx}->{inboxdir}/git";
700 my $all = "$self->{ibx}->{inboxdir}/all.git";
701 PublicInbox::Import::init_bare($all) unless -d $all;
702 my $info_dir = "$all/objects/info";
703 my $alt = "$info_dir/alternates";
707 open(my $fh, '<', $alt) or die "open < $alt: $!\n";
708 $mode = (stat($fh))[2] & 07777;
710 # we assign a sort score to every alternate and favor
711 # the newest (highest numbered) one because loose objects
712 # require scanning epochs and only the latest epoch is
713 # expected to see loose objects
715 my $other = 0; # in case admin adds non-epoch repos
717 if (m!\A\Q../../\E([0-9]+)\.git/objects\z!) {
723 } split(/\n+/, do { local $/; <$fh> });
726 foreach my $i (0..$epoch) {
727 my $dir = "../../git/$i.git/objects";
728 if (!exists($alt{$dir}) && -d "$pfx/$i.git") {
734 write_alternates($info_dir, $mode,
735 [join("\n", sort { $alt{$b} <=> $alt{$a} } keys %alt), "\n"]);
739 my ($self, $epoch) = @_;
740 my $git_dir = "$self->{ibx}->{inboxdir}/git/$epoch.git";
741 PublicInbox::Import::init_bare($git_dir);
742 my @cmd = (qw/git config/, "--file=$git_dir/config",
743 'include.path', '../../all.git/config');
744 PublicInbox::Import::run_die(\@cmd);
745 fill_alternates($self, $epoch);
750 my ($self, $max) = @_;
752 my $pfx = "$self->{ibx}->{inboxdir}/git";
753 return unless -d $pfx;
755 opendir my $dh, $pfx or die "opendir $pfx: $!\n";
756 while (defined(my $git_dir = readdir($dh))) {
757 $git_dir =~ m!\A([0-9]+)\.git\z! or next;
760 $latest = "$pfx/$git_dir";
768 my $im = $self->{im};
770 if ($im->{bytes_added} < $self->{rotate_bytes}) {
777 my $git_dir = $self->git_init(++$self->{epoch_max});
778 my $git = PublicInbox::Git->new($git_dir);
779 return $self->import_init($git, 0);
784 my $latest = git_dir_latest($self, \$max);
785 if (defined $latest) {
786 my $git = PublicInbox::Git->new($latest);
787 my $packed_bytes = $git->packed_bytes;
788 my $unpacked_bytes = $packed_bytes / $PACKING_FACTOR;
790 if ($unpacked_bytes >= $self->{rotate_bytes}) {
793 $self->{epoch_max} = $max;
794 return $self->import_init($git, $packed_bytes);
797 $self->{epoch_max} = $epoch;
798 $latest = $self->git_init($epoch);
799 $self->import_init(PublicInbox::Git->new($latest), 0);
803 my ($self, $git, $packed_bytes, $tmp) = @_;
804 my $im = PublicInbox::Import->new($git, undef, undef, $self->{ibx});
805 $im->{bytes_added} = int($packed_bytes / $PACKING_FACTOR);
806 $im->{lock_path} = undef;
807 $im->{path_type} = 'v2';
808 $self->{im} = $im unless $tmp;
814 my ($mid, $cur, $new) = @_;
816 my $ah = File::Temp->new(TEMPLATE => 'email-cur-XXXXXXXX', TMPDIR => 1);
817 print $ah $cur->as_string or die "print: $!";
818 $ah->flush or die "flush: $!";
819 PublicInbox::Import::drop_unwanted_headers($new);
820 my $bh = File::Temp->new(TEMPLATE => 'email-new-XXXXXXXX', TMPDIR => 1);
821 print $bh $new->as_string or die "print: $!";
822 $bh->flush or die "flush: $!";
823 my $cmd = [ qw(diff -u), $ah->filename, $bh->filename ];
824 print STDERR "# MID conflict <$mid>\n";
825 my $pid = spawn($cmd, undef, { 1 => 2 });
826 waitpid($pid, 0) == $pid or die "diff did not finish";
830 my ($self, $smsg) = @_;
831 if (my $im = $self->{im}) {
832 my $msg = $im->cat_blob($smsg->{blob});
835 # older message, should be in alternates
836 $self->{ibx}->msg_by_smsg($smsg);
839 sub content_exists ($$$) {
840 my ($self, $mime, $mid) = @_;
841 my $oidx = $self->{oidx};
842 my $chashes = content_hashes($mime);
844 while (my $smsg = $oidx->next_by_mid($mid, \$id, \$prev)) {
845 my $msg = get_blob($self, $smsg);
846 if (!defined($msg)) {
847 warn "broken smsg for $mid\n";
850 my $cur = PublicInbox::Eml->new($msg);
851 return 1 if content_matches($chashes, $cur);
853 # XXX DEBUG_DIFF is experimental and may be removed
854 diff($mid, $cur, $mime) if $ENV{DEBUG_DIFF};
861 if (my $shards = $self->{idx_shards}) {
862 $_->atfork_child foreach @$shards;
864 if (my $im = $self->{im}) {
867 die "unexpected mm" if $self->{mm};
868 close $self->{bnote}->[0] or die "close bnote[0]: $!\n";
872 sub reindex_checkpoint ($$) {
873 my ($self, $sync) = @_;
875 $self->git->cleanup; # *async_wait
876 ${$sync->{need_checkpoint}} = 0;
877 my $mm_tmp = $sync->{mm_tmp};
878 $mm_tmp->atfork_prepare if $mm_tmp;
879 $self->done; # release lock
881 if (my $pr = $sync->{-opt}->{-progress}) {
882 $pr->(sprintf($sync->{-regen_fmt}, ${$sync->{nr}}));
885 # allow -watch or -mda to write...
886 $self->idx_init($sync->{-opt}); # reacquire lock
887 $mm_tmp->atfork_parent if $mm_tmp;
890 sub index_oid { # cat_async callback
891 my ($bref, $oid, $type, $size, $arg) = @_;
892 return if $size == 0; # purged
894 my $eml = PublicInbox::Eml->new($$bref);
895 my $mids = mids($eml);
896 my $chash = content_hash($eml);
897 my $self = $arg->{v2w};
899 if (scalar(@$mids) == 0) {
900 warn "E: $oid has no Message-ID, skipping\n";
904 # {unindexed} is unlikely
905 if ((my $unindexed = $arg->{unindexed}) && scalar(@$mids) == 1) {
906 $num = delete($unindexed->{$mids->[0]});
909 $self->{mm}->mid_set($num, $mid0);
910 delete($arg->{unindexed}) if !keys(%$unindexed);
913 if (!defined($num)) { # reuse if reindexing (or duplicates)
914 my $oidx = $self->{oidx};
915 for my $mid (@$mids) {
916 ($num, $mid0) = $oidx->num_mid0_for_oid($oid, $mid);
917 last if defined $num;
920 $mid0 //= do { # is this a number we got before?
921 $num = $arg->{mm_tmp}->num_for($mids->[0]);
922 defined($num) ? $mids->[0] : undef;
924 if (!defined($num)) {
925 for (my $i = $#$mids; $i >= 1; $i--) {
926 $num = $arg->{mm_tmp}->num_for($mids->[$i]);
934 $arg->{mm_tmp}->num_delete($num);
935 } else { # never seen
936 $num = $self->{mm}->mid_insert($mids->[0]);
939 } else { # rare, try the rest of them, backwards
940 for (my $i = $#$mids; $i >= 1; $i--) {
941 $num = $self->{mm}->mid_insert($mids->[$i]);
949 if (!defined($num)) {
950 warn "E: $oid <", join('> <', @$mids), "> is a duplicate\n";
959 }, 'PublicInbox::Smsg';
960 $smsg->populate($eml, $arg);
961 if (do_idx($self, $bref, $eml, $smsg)) {
962 ${$arg->{need_checkpoint}} = 1;
966 # only update last_commit for $i on reindex iff newer than current
967 sub update_last_commit {
968 my ($self, $git, $i, $cmt) = @_;
969 my $last = last_epoch_commit($self, $i);
970 if (defined $last && is_ancestor($git, $last, $cmt)) {
971 my @cmd = (qw(rev-list --count), "$last..$cmt");
972 chomp(my $n = $git->qx(@cmd));
973 return if $n ne '' && $n == 0;
975 last_epoch_commit($self, $i, $cmt);
978 sub git_dir_n ($$) { "$_[0]->{ibx}->{inboxdir}/git/$_[1].git" }
980 sub last_commits ($$) {
981 my ($self, $epoch_max) = @_;
983 for (my $i = $epoch_max; $i >= 0; $i--) {
984 $heads->[$i] = last_epoch_commit($self, $i);
989 # returns a revision range for git-log(1)
990 sub log_range ($$$$$) {
991 my ($self, $sync, $git, $i, $tip) = @_;
992 my $opt = $sync->{-opt};
993 my $pr = $opt->{-progress} if (($opt->{verbose} || 0) > 1);
994 my $cur = $sync->{ranges}->[$i] or do {
995 $pr->("$i.git indexing all of $tip\n") if $pr;
996 return $tip; # all of it
999 # fast equality check to avoid (v)fork+execve overhead
1001 $sync->{ranges}->[$i] = undef;
1005 my $range = "$cur..$tip";
1006 $pr->("$i.git checking contiguity... ") if $pr;
1007 if (is_ancestor($git, $cur, $tip)) { # common case
1008 $pr->("OK\n") if $pr;
1009 my $n = $git->qx(qw(rev-list --count), $range);
1012 $sync->{ranges}->[$i] = undef;
1013 $pr->("$i.git has nothing new\n") if $pr;
1014 return; # nothing to do
1016 $pr->("$i.git has $n changes since $cur\n") if $pr;
1018 $pr->("FAIL\n") if $pr;
1020 discontiguous range: $range
1021 Rewritten history? (in $git->{git_dir})
1023 chomp(my $base = $git->qx('merge-base', $tip, $cur));
1025 $range = "$base..$tip";
1026 warn "found merge-base: $base\n"
1029 warn "discarding history at $cur\n";
1032 reindexing $git->{git_dir} starting at
1035 $sync->{unindex_range}->{$i} = "$base..$cur";
1040 sub sync_prepare ($$$) {
1041 my ($self, $sync, $epoch_max) = @_;
1042 my $pr = $sync->{-opt}->{-progress};
1044 my $head = $self->{ibx}->{ref_head} || 'HEAD';
1046 # reindex stops at the current heads and we later rerun index_sync
1048 my $reindex_heads = $self->last_commits($epoch_max) if $sync->{reindex};
1050 for (my $i = $epoch_max; $i >= 0; $i--) {
1051 my $git_dir = git_dir_n($self, $i);
1052 -d $git_dir or next; # missing epochs are fine
1053 my $git = PublicInbox::Git->new($git_dir);
1054 if ($reindex_heads) {
1055 $head = $reindex_heads->[$i] or next;
1057 chomp(my $tip = $git->qx(qw(rev-parse -q --verify), $head));
1059 next if $?; # new repo
1060 my $range = log_range($self, $sync, $git, $i, $tip) or next;
1061 # can't use 'rev-list --count' if we use --diff-filter
1062 $pr->("$i.git counting $range ... ") if $pr;
1063 # Don't bump num_highwater on --reindex by using {D}.
1064 # We intentionally do NOT use {D} in the non-reindex case
1065 # because we want NNTP article number gaps from unindexed
1066 # messages to show up in mirrors, too.
1067 $sync->{D} //= $sync->{reindex} ? {} : undef; # OID_BIN => NR
1068 my $stk = log2stack($sync, $git, $range, $self->{ibx});
1069 my $nr = $stk ? $stk->num_records : 0;
1070 $pr->("$nr\n") if $pr;
1071 $sync->{stacks}->[$i] = $stk if $stk;
1075 # XXX this should not happen unless somebody bypasses checks in
1076 # our code and blindly injects "d" file history into git repos
1077 if (my @leftovers = keys %{delete($sync->{D}) // {}}) {
1078 warn('W: unindexing '.scalar(@leftovers)." leftovers\n");
1079 my $arg = { v2w => $self };
1080 for my $oid (@leftovers) {
1081 $oid = unpack('H*', $oid);
1082 $self->{current_info} = "leftover $oid";
1083 $self->git->cat_async($oid, \&unindex_oid, $arg);
1085 $self->git->cat_async_wait;
1088 $sync->{-regen_fmt} = "%u/?\n";
1092 # reindex should NOT see new commits anymore, if we do,
1093 # it's a problem and we need to notice it via die()
1094 my $pad = length($regen_max) + 1;
1095 $sync->{-regen_fmt} = "% ${pad}u/$regen_max\n";
1096 $sync->{nr} = \(my $nr = 0);
1097 return -1 if $sync->{reindex};
1098 $regen_max + $self->{mm}->num_highwater() || 0;
1101 sub unindex_oid_remote ($$$) {
1102 my ($self, $oid, $mid) = @_;
1103 my @removed = $self->{oidx}->remove_oid($oid, $mid);
1104 for my $num (@removed) {
1105 my $idx = idx_shard($self, $num);
1106 $idx->shard_remove($oid, $num);
1110 sub unindex_oid ($$;$) { # git->cat_async callback
1111 my ($bref, $oid, $type, $size, $sync) = @_;
1112 my $self = $sync->{v2w};
1113 my $unindexed = $sync->{in_unindex} ? $sync->{unindexed} : undef;
1114 my $mm = $self->{mm};
1115 my $mids = mids(PublicInbox::Eml->new($bref));
1117 my $oidx = $self->{oidx};
1118 foreach my $mid (@$mids) {
1121 while (my $smsg = $oidx->next_by_mid($mid, \$id, \$prev)) {
1122 $gone{$smsg->{num}} = 1 if $oid eq $smsg->{blob};
1124 my $n = scalar(keys(%gone)) or next;
1126 warn "BUG: multiple articles linked to $oid\n",
1127 join(',',sort keys %gone), "\n";
1129 foreach my $num (keys %gone) {
1131 my $mid0 = $mm->mid_for($num);
1132 $unindexed->{$mid0} = $num;
1134 $mm->num_delete($num);
1136 unindex_oid_remote($self, $oid, $mid);
1140 sub git { $_[0]->{ibx}->git }
1142 # this is rare, it only happens when we get discontiguous history in
1143 # a mirror because the source used -purge or -edit
1144 sub unindex ($$$$) {
1145 my ($self, $sync, $git, $unindex_range) = @_;
1146 my $unindexed = $sync->{unindexed} //= {}; # $mid0 => $num
1147 my $before = scalar keys %$unindexed;
1148 # order does not matter, here:
1149 my @cmd = qw(log --raw -r
1150 --no-notes --no-color --no-abbrev --no-renames);
1151 my $fh = $git->popen(@cmd, $unindex_range);
1152 local $sync->{in_unindex} = 1;
1154 /\A:\d{6} 100644 $OID ($OID) [AM]\tm$/o or next;
1155 $self->git->cat_async($1, \&unindex_oid, $sync);
1157 close $fh or die "git log failed: \$?=$?";
1158 $self->git->cat_async_wait;
1160 return unless $sync->{-opt}->{prune};
1161 my $after = scalar keys %$unindexed;
1162 return if $before == $after;
1164 # ensure any blob can not longer be accessed via dumb HTTP
1165 PublicInbox::Import::run_die(['git', "--git-dir=$git->{git_dir}",
1166 qw(-c gc.reflogExpire=now gc --prune=all --quiet)]);
1169 sub sync_ranges ($$$) {
1170 my ($self, $sync, $epoch_max) = @_;
1171 my $reindex = $sync->{reindex};
1173 return last_commits($self, $epoch_max) unless $reindex;
1174 return [] if ref($reindex) ne 'HASH';
1176 my $ranges = $reindex->{from}; # arrayref;
1177 if (ref($ranges) ne 'ARRAY') {
1178 die 'BUG: $reindex->{from} not an ARRAY';
1183 sub index_xap_only { # git->cat_async callback
1184 my ($bref, $oid, $type, $size, $smsg) = @_;
1185 my $self = $smsg->{v2w};
1186 my $idx = idx_shard($self, $smsg->{num});
1187 $smsg->{raw_bytes} = $size;
1188 $idx->index_raw($bref, undef, $smsg);
1189 $self->{transact_bytes} += $size;
1192 sub index_xap_step ($$$;$) {
1193 my ($self, $sync, $beg, $step) = @_;
1194 my $end = $sync->{art_end};
1195 return if $beg > $end; # nothing to do
1197 $step //= $self->{shards};
1198 my $ibx = $self->{ibx};
1199 if (my $pr = $sync->{-opt}->{-progress}) {
1200 $pr->("Xapian indexlevel=$ibx->{indexlevel} ".
1201 "$beg..$end (% $step)\n");
1203 for (my $num = $beg; $num <= $end; $num += $step) {
1204 my $smsg = $ibx->over->get_art($num) or next;
1205 $smsg->{v2w} = $self;
1206 $ibx->git->cat_async($smsg->{blob}, \&index_xap_only, $smsg);
1207 if ($self->{transact_bytes} >= $self->{batch_bytes}) {
1208 ${$sync->{nr}} = $num;
1209 reindex_checkpoint($self, $sync);
1214 sub index_epoch ($$$) {
1215 my ($self, $sync, $i) = @_;
1217 my $git_dir = git_dir_n($self, $i);
1218 -d $git_dir or return; # missing epochs are fine
1219 my $git = PublicInbox::Git->new($git_dir);
1220 if (my $unindex_range = delete $sync->{unindex_range}->{$i}) { # rare
1221 unindex($self, $sync, $git, $unindex_range);
1223 defined(my $stk = $sync->{stacks}->[$i]) or return;
1224 $sync->{stacks}->[$i] = undef;
1225 my $all = $self->git;
1226 while (my ($f, $at, $ct, $oid) = $stk->pop_rec) {
1227 $self->{current_info} = "$i.git $oid";
1229 my $arg = { %$sync, autime => $at, cotime => $ct };
1230 if ($sync->{max_size}) {
1231 $all->check_async($oid, \&check_size, $arg);
1233 $all->cat_async($oid, \&index_oid, $arg);
1235 } elsif ($f eq 'd') {
1236 $all->cat_async($oid, \&unindex_oid, $sync);
1238 if (${$sync->{need_checkpoint}}) {
1239 reindex_checkpoint($self, $sync);
1242 $all->async_wait_all;
1243 $self->update_last_commit($git, $i, $stk->{latest_cmt});
1247 my ($self, $opt, $sync, $art_beg) = @_;
1248 my $seq = $opt->{sequential_shard};
1250 local $self->{parallel} = 0 if $seq;
1251 $self->idx_init($opt); # acquire lock
1252 if (my $art_end = $self->{ibx}->mm->max) {
1254 need_checkpoint => \(my $bool = 0),
1257 nr => \(my $nr = 0),
1258 -regen_fmt => "%u/?\n",
1260 $sync->{art_end} = $art_end;
1261 if ($seq || !$self->{parallel}) {
1262 my $shard_end = $self->{shards} - 1;
1263 for my $i (0..$shard_end) {
1264 index_xap_step($self, $sync, $art_beg + $i);
1265 if ($i != $shard_end) {
1266 reindex_checkpoint($self, $sync);
1269 } else { # parallel (maybe)
1270 index_xap_step($self, $sync, $art_beg, 1);
1273 $self->git->cat_async_wait;
1277 # public, called by public-inbox-index
1279 my ($self, $opt) = @_;
1281 return xapian_only($self, $opt) if $opt->{xapian_only};
1283 my $pr = $opt->{-progress};
1285 my $latest = git_dir_latest($self, \$epoch_max);
1286 return unless defined $latest;
1288 my $seq = $opt->{sequential_shard};
1289 my $art_beg; # the NNTP article number we start xapian_only at
1290 my $idxlevel = $self->{ibx}->{indexlevel};
1291 local $self->{ibx}->{indexlevel} = 'basic' if $seq;
1293 $self->idx_init($opt); # acquire lock
1294 fill_alternates($self, $epoch_max);
1295 $self->{oidx}->rethread_prepare($opt);
1297 need_checkpoint => \(my $bool = 0),
1298 unindex_range => {}, # EPOCH => oid_old..oid_new
1299 reindex => $opt->{reindex},
1303 $sync->{ranges} = sync_ranges($self, $sync, $epoch_max);
1304 if (sync_prepare($self, $sync, $epoch_max)) {
1305 # tmp_clone seems to fail if inside a transaction, so
1306 # we rollback here (because we opened {mm} for reading)
1307 # Note: we do NOT rely on DBI transactions for atomicity;
1308 # only for batch performance.
1309 $self->{mm}->{dbh}->rollback;
1310 $self->{mm}->{dbh}->begin_work;
1312 $self->{mm}->tmp_clone($self->{ibx}->{inboxdir});
1314 # xapian_only works incrementally w/o --reindex
1315 if ($seq && !$opt->{reindex}) {
1316 $art_beg = $sync->{mm_tmp}->max;
1317 $art_beg++ if defined($art_beg);
1320 if ($sync->{max_size} = $opt->{max_size}) {
1321 $sync->{index_oid} = \&index_oid;
1323 # work forwards through history
1324 index_epoch($self, $sync, $_) for (0..$epoch_max);
1325 $self->{oidx}->rethread_done($opt);
1328 if (my $nr = $sync->{nr}) {
1329 my $pr = $sync->{-opt}->{-progress};
1330 $pr->('all.git '.sprintf($sync->{-regen_fmt}, $$nr)) if $pr;
1333 # deal with Xapian shards sequentially
1334 if ($seq && delete($sync->{mm_tmp})) {
1335 $self->{ibx}->{indexlevel} = $idxlevel;
1336 xapian_only($self, $opt, $sync, $art_beg);
1339 # --reindex on the command-line
1340 if ($opt->{reindex} && !ref($opt->{reindex}) && $idxlevel ne 'basic') {
1341 $self->lock_acquire;
1342 my $s0 = PublicInbox::SearchIdx->new($self->{ibx}, 0, 0);
1343 if (my $xdb = $s0->idx_acquire) {
1344 my $n = $xdb->get_metadata('has_threadid');
1345 $xdb->set_metadata('has_threadid', '1') if $n ne '1';
1348 $self->lock_release;
1351 # reindex does not pick up new changes, so we rerun w/o it:
1352 if ($opt->{reindex}) {
1355 delete @again{qw(rethread reindex -skip_lock)};
1356 index_sync($self, \%again);