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
23 use File::Temp qw(tempfile);
25 my $OID = qr/[a-f0-9]{40,}/;
26 # an estimate of the post-packed size to the raw uncompressed size
27 my $PACKING_FACTOR = 0.4;
29 # SATA storage lags behind what CPUs are capable of, so relying on
30 # nproc(1) can be misleading and having extra Xapian shards is a
31 # waste of FDs and space. It can also lead to excessive IO latency
32 # and slow things down. Users on NVME or other fast storage can
33 # use the NPROC env or switches in our script/public-inbox-* programs
34 # to increase Xapian shards
35 our $NPROC_MAX_DEFAULT = 4;
38 for my $nproc (qw(nproc gnproc)) { # GNU coreutils nproc
39 `$nproc 2>/dev/null` =~ /^(\d+)$/ and return $1;
42 # getconf(1) is POSIX, but *NPROCESSORS* vars are not
43 for (qw(_NPROCESSORS_ONLN NPROCESSORS_ONLN)) {
44 `getconf $_ 2>/dev/null` =~ /^(\d+)$/ and return $1;
47 # should we bother with `sysctl hw.ncpu`? Those only give
48 # us total processor count, not online processor count.
52 sub nproc_shards ($) {
54 my $n = $creat_opt->{nproc} if ref($creat_opt) eq 'HASH';
57 # assume 2 cores if not detectable or zero
58 state $NPROC_DETECTED = detect_nproc() || 2;
60 $n = $NPROC_MAX_DEFAULT if $n > $NPROC_MAX_DEFAULT;
63 # subtract for the main process and git-fast-import
68 sub count_shards ($) {
71 my $xpfx = $self->{xpfx};
73 # always load existing shards in case core count changes:
74 # Also, shard count may change while -watch is running
75 # due to "xcpdb --reshard"
78 foreach my $shard (<$xpfx/*>) {
79 -d $shard && $shard =~ m!/[0-9]+\z! or next;
80 $XapianDatabase //= do {
81 require PublicInbox::Search;
82 PublicInbox::Search::load_xapian();
83 $PublicInbox::Search::X{Database};
86 $XapianDatabase->new($shard)->close;
95 # $creat may be any true value, or 0/undef. A hashref is true,
96 # and $creat->{nproc} may be set to an integer
97 my ($class, $v2ibx, $creat) = @_;
98 $v2ibx = PublicInbox::InboxWritable->new($v2ibx);
99 my $dir = $v2ibx->assert_usable_dir;
103 File::Path::mkpath($dir);
105 die "$dir does not exist\n";
108 $v2ibx->umask_prepare;
110 my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION;
113 im => undef, # PublicInbox::Import
119 over => PublicInbox::OverIdx->new("$xpfx/over.sqlite3"),
120 lock_path => "$dir/inbox.lock",
121 # limit each git repo (epoch) to 1GB or so
122 rotate_bytes => int((1024 * 1024 * 1024) / $PACKING_FACTOR),
123 last_commit => [], # git epoch -> commit
125 $self->{over}->{-no_sync} = 1 if $v2ibx->{-no_sync};
126 $self->{shards} = count_shards($self) || nproc_shards($creat);
132 my ($self, $shards, $skip_epoch, $skip_artnum) = @_;
133 if (defined $shards) {
134 $self->{parallel} = 0 if $shards == 0;
135 $self->{shards} = $shards if $shards > 0;
138 $self->{mm}->skip_artnum($skip_artnum) if defined $skip_artnum;
140 git_dir_latest($self, \$epoch_max);
141 if (defined $skip_epoch && $epoch_max == -1) {
142 $epoch_max = $skip_epoch;
144 $self->git_init($epoch_max >= 0 ? $epoch_max : 0);
148 # returns undef on duplicate or spam
149 # mimics Import::add and wraps it for v2
151 my ($self, $eml, $check_cb) = @_;
152 $self->{ibx}->with_umask(\&_add, $self, $eml, $check_cb);
155 # indexes a message, returns true if checkpointing is needed
157 my ($self, $msgref, $mime, $smsg) = @_;
158 $smsg->{bytes} = $smsg->{raw_bytes} + crlf_adjust($$msgref);
159 $self->{over}->add_overview($mime, $smsg);
160 my $idx = idx_shard($self, $smsg->{num} % $self->{shards});
161 $idx->index_raw($msgref, $mime, $smsg);
162 my $n = $self->{transact_bytes} += $smsg->{raw_bytes};
163 $n >= ($PublicInbox::SearchIdx::BATCH_BYTES * $self->{shards});
167 my ($self, $mime, $check_cb) = @_;
171 $mime = $check_cb->($mime, $self->{ibx}) or return;
174 # All pipes (> $^F) known to Perl 5.6+ have FD_CLOEXEC set,
175 # as does SQLite 3.4.1+ (released in 2007-07-20), and
176 # Xapian 1.3.2+ (released 2015-03-15).
177 # For the most part, we can spawn git-fast-import without
178 # leaking FDs to it...
181 my ($num, $mid0) = v2_num_for($self, $mime);
182 defined $num or return; # duplicate
183 defined $mid0 or die "BUG: \$mid0 undefined\n";
184 my $im = $self->importer;
185 my $smsg = bless { mid => $mid0, num => $num }, 'PublicInbox::Smsg';
186 my $cmt = $im->add($mime, undef, $smsg); # sets $smsg->{ds|ts|blob}
187 $cmt = $im->get_mark($cmt);
188 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
190 my $msgref = delete $smsg->{-raw_email};
191 if (do_idx($self, $msgref, $mime, $smsg)) {
199 my ($self, $mime) = @_;
200 my $mids = mids($mime);
202 my $mid = $mids->[0];
203 my $num = $self->{mm}->mid_insert($mid);
204 if (defined $num) { # common case
208 # crap, Message-ID is already known, hope somebody just resent:
209 foreach my $m (@$mids) {
210 # read-only lookup now safe to do after above barrier
211 # easy, don't store duplicates
212 # note: do not add more diagnostic info here since
213 # it gets noisy on public-inbox-watch restarts
214 return () if content_exists($self, $mime, $m);
217 # AltId may pre-populate article numbers (e.g. X-Mail-Count
218 # or NNTP article number), use that article number if it's
220 my $altid = $self->{ibx}->{altid};
221 if ($altid && grep(/:file=msgmap\.sqlite3\z/, @$altid)) {
222 my $num = $self->{mm}->num_for($mid);
224 if (defined $num && !$self->{over}->get_art($num)) {
230 warn "<$mid> reused for mismatched content\n";
232 # try the rest of the mids
233 for(my $i = $#$mids; $i >= 1; $i--) {
235 $num = $self->{mm}->mid_insert($m);
237 warn "alternative <$m> for <$mid> found\n";
242 # none of the existing Message-IDs are good, generate a new one:
243 v2_num_for_harder($self, $mime);
246 sub v2_num_for_harder {
247 my ($self, $eml) = @_;
249 my $dig = content_digest($eml);
250 my $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
251 my $num = $self->{mm}->mid_insert($mid0);
252 unless (defined $num) {
253 # it's hard to spoof the last Received: header
254 my @recvd = $eml->header_raw('Received');
255 $dig->add("Received: $_") foreach (@recvd);
256 $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
257 $num = $self->{mm}->mid_insert($mid0);
259 # fall back to a random Message-ID and give up determinism:
260 until (defined($num)) {
262 $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
263 warn "using random Message-ID <$mid0> as fallback\n";
264 $num = $self->{mm}->mid_insert($mid0);
267 PublicInbox::Import::append_mid($eml, $mid0);
272 my ($self, $shard_i) = @_;
273 $self->{idx_shards}->[$shard_i];
276 sub _idx_init { # with_umask callback
277 my ($self, $opt) = @_;
278 $self->lock_acquire unless $opt && $opt->{-skip_lock};
279 $self->{over}->create;
281 # xcpdb can change shard count while -watch is idle
282 my $nshards = count_shards($self);
283 $self->{shards} = $nshards if $nshards && $nshards != $self->{shards};
285 # need to create all shards before initializing msgmap FD
286 # idx_shards must be visible to all forked processes
287 my $max = $self->{shards} - 1;
288 my $idx = $self->{idx_shards} = [];
289 push @$idx, PublicInbox::SearchIdxShard->new($self, $_) for (0..$max);
291 # Now that all subprocesses are up, we can open the FDs
293 my $mm = $self->{mm} = PublicInbox::Msgmap->new_file(
294 "$self->{ibx}->{inboxdir}/msgmap.sqlite3",
295 $self->{ibx}->{-no_sync} ? 2 : 1);
296 $mm->{dbh}->begin_work;
301 my ($self, $opt) = @_;
302 return if $self->{idx_shards};
303 my $ibx = $self->{ibx};
305 # do not leak read-only FDs to child processes, we only have these
306 # FDs for duplicate detection so they should not be
307 # frequently activated.
308 delete @$ibx{qw(mm search)};
311 $self->{parallel} = 0 if ($ibx->{indexlevel}//'') eq 'basic';
312 if ($self->{parallel}) {
313 pipe(my ($r, $w)) or die "pipe failed: $!";
314 # pipe for barrier notifications doesn't need to be big,
316 fcntl($w, 1031, 4096) if $^O eq 'linux';
317 $self->{bnote} = [ $r, $w ];
322 $ibx->with_umask(\&_idx_init, $self, $opt);
325 # returns an array mapping [ epoch => latest_commit ]
326 # latest_commit may be undef if nothing was done to that epoch
327 # $replace_map = { $object_id => $strref, ... }
328 sub _replace_oids ($$$) {
329 my ($self, $mime, $replace_map) = @_;
331 my $pfx = "$self->{ibx}->{inboxdir}/git";
332 my $rewrites = []; # epoch => commit
333 my $max = $self->{epoch_max};
335 unless (defined($max)) {
336 defined(my $latest = git_dir_latest($self, \$max)) or return;
337 $self->{epoch_max} = $max;
340 foreach my $i (0..$max) {
341 my $git_dir = "$pfx/$i.git";
343 my $git = PublicInbox::Git->new($git_dir);
344 my $im = $self->import_init($git, 0, 1);
345 $rewrites->[$i] = $im->replace_oids($mime, $replace_map);
351 sub content_hashes ($) {
353 my @chashes = ( content_hash($mime) );
355 # We still support Email::MIME, here, and
356 # Email::MIME->as_string doesn't always round-trip, so we may
357 # use a second content_hash
358 my $rt = content_hash(PublicInbox::Eml->new(\($mime->as_string)));
359 push @chashes, $rt if $chashes[0] ne $rt;
363 sub content_matches ($$) {
364 my ($chashes, $existing) = @_;
365 my $chash = content_hash($existing);
366 foreach (@$chashes) {
367 return 1 if $_ eq $chash
372 # used for removing or replacing (purging)
373 sub rewrite_internal ($$;$$$) {
374 my ($self, $old_eml, $cmt_msg, $new_eml, $sref) = @_;
376 my ($im, $need_reindex, $replace_map);
378 $replace_map = {}; # oid => sref
379 $need_reindex = [] if $new_eml;
381 $im = $self->importer;
383 my $over = $self->{over};
384 my $chashes = content_hashes($old_eml);
386 my $mids = mids($old_eml);
388 # We avoid introducing new blobs into git since the raw content
389 # can be slightly different, so we do not need the user-supplied
390 # message now that we have the mids and content_hash
394 foreach my $mid (@$mids) {
395 my %gone; # num => [ smsg, $mime, raw ]
397 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
398 my $msg = get_blob($self, $smsg);
399 if (!defined($msg)) {
400 warn "broken smsg for $mid\n";
404 my $cur = PublicInbox::Eml->new($msg);
405 if (content_matches($chashes, $cur)) {
406 $gone{$smsg->{num}} = [ $smsg, $cur, \$orig ];
409 my $n = scalar keys %gone;
412 warn "BUG: multiple articles linked to <$mid>\n",
413 join(',', sort keys %gone), "\n";
415 foreach my $num (keys %gone) {
416 my ($smsg, $mime, $orig) = @{$gone{$num}};
417 # $removed should only be set once assuming
418 # no bugs in our deduplication code:
419 $removed = [ undef, $mime, $smsg ];
420 my $oid = $smsg->{blob};
422 $replace_map->{$oid} = $sref;
424 ($mark, undef) = $im->remove($orig, $cmt_msg);
425 $removed->[0] = $mark;
428 if ($need_reindex) { # ->replace
429 push @$need_reindex, $smsg;
430 } else { # ->purge or ->remove
431 $self->{mm}->num_delete($num);
433 unindex_oid_remote($self, $oid, $mid);
438 my $cmt = $im->get_mark($mark);
439 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
441 if ($replace_map && scalar keys %$replace_map) {
442 my $rewrites = _replace_oids($self, $new_eml, $replace_map);
443 return { rewrites => $rewrites, need_reindex => $need_reindex };
445 defined($mark) ? $removed : undef;
448 # public (see PublicInbox::Import->remove), but note the 3rd element
449 # (retval[2]) is not part of the stable API shared with Import->remove
451 my ($self, $eml, $cmt_msg) = @_;
452 my $r = $self->{ibx}->with_umask(\&rewrite_internal,
453 $self, $eml, $cmt_msg);
454 defined($r) && defined($r->[0]) ? @$r: undef;
457 sub _replace ($$;$$) {
458 my ($self, $old_eml, $new_eml, $sref) = @_;
459 my $arg = [ $self, $old_eml, undef, $new_eml, $sref ];
460 my $rewritten = $self->{ibx}->with_umask(\&rewrite_internal,
461 $self, $old_eml, undef, $new_eml, $sref) or return;
463 my $rewrites = $rewritten->{rewrites};
464 # ->done is called if there are rewrites since we gc+prune from git
465 $self->idx_init if @$rewrites;
467 for my $i (0..$#$rewrites) {
468 defined(my $cmt = $rewrites->[$i]) or next;
469 $self->{last_commit}->[$i] = $cmt;
476 my ($self, $mime) = @_;
477 my $rewritten = _replace($self, $mime, undef, \'') or return;
478 $rewritten->{rewrites}
481 # returns the git object_id of $fh, does not write the object to FS
482 sub git_hash_raw ($$) {
483 my ($self, $raw) = @_;
484 # grab the expected OID we have to reindex:
485 pipe(my($in, $w)) or die "pipe: $!";
486 my $git_dir = $self->{ibx}->git->{git_dir};
487 my $cmd = ['git', "--git-dir=$git_dir", qw(hash-object --stdin)];
488 my $r = popen_rd($cmd, undef, { 0 => $in });
489 print $w $$raw or die "print \$w: $!";
490 close $w or die "close \$w: $!";
492 chomp(my $oid = <$r>);
493 close $r or die "git hash-object failed: $?";
494 $oid =~ /\A$OID\z/ or die "OID not expected: $oid";
498 sub _check_mids_match ($$$) {
499 my ($old_list, $new_list, $hdrs) = @_;
500 my %old_mids = map { $_ => 1 } @$old_list;
501 my %new_mids = map { $_ => 1 } @$new_list;
502 my @old = keys %old_mids;
503 my @new = keys %new_mids;
504 my $err = "$hdrs may not be changed when replacing\n";
505 die $err if scalar(@old) != scalar(@new);
506 delete @new_mids{@old};
507 delete @old_mids{@new};
508 die $err if (scalar(keys %old_mids) || scalar(keys %new_mids));
511 # Changing Message-IDs or References with ->replace isn't supported.
512 # The rules for dealing with messages with multiple or conflicting
513 # Message-IDs are pretty complex and rethreading hasn't been fully
515 sub check_mids_match ($$) {
516 my ($old, $new) = @_;
517 _check_mids_match(mids($old), mids($new), 'Message-ID(s)');
518 _check_mids_match(references($old), references($new),
519 'References/In-Reply-To');
524 my ($self, $old_mime, $new_mime) = @_;
526 check_mids_match($old_mime, $new_mime);
528 # mutt will always add Content-Length:, Status:, Lines: when editing
529 PublicInbox::Import::drop_unwanted_headers($new_mime);
531 my $raw = $new_mime->as_string;
532 my $expect_oid = git_hash_raw($self, \$raw);
533 my $rewritten = _replace($self, $old_mime, $new_mime, \$raw) or return;
534 my $need_reindex = $rewritten->{need_reindex};
536 # just in case we have bugs in deduplication code:
537 my $n = scalar(@$need_reindex);
539 my $list = join(', ', map {
540 "$_->{num}: <$_->{mid}>"
543 W: rewritten $n messages matching content of original message (expected: 1).
544 W: possible bug in public-inbox, NNTP article IDs and Message-IDs follow:
549 # make sure we really got the OID:
550 my ($blob, $type, $bytes) = $self->{ibx}->git->check($expect_oid);
551 $blob eq $expect_oid or die "BUG: $expect_oid not found after replace";
553 # don't leak FDs to Xapian:
554 $self->{ibx}->git->cleanup;
556 # reindex modified messages:
557 for my $smsg (@$need_reindex) {
558 my $new_smsg = bless {
563 }, 'PublicInbox::Smsg';
564 my $sync = { autime => $smsg->{ds}, cotime => $smsg->{ts} };
565 $new_smsg->populate($new_mime, $sync);
566 do_idx($self, \$raw, $new_mime, $new_smsg);
568 $rewritten->{rewrites};
571 sub last_epoch_commit ($$;$) {
572 my ($self, $i, $cmt) = @_;
573 my $v = PublicInbox::Search::SCHEMA_VERSION();
574 $self->{mm}->last_commit_xap($v, $i, $cmt);
577 sub set_last_commits ($) {
579 defined(my $epoch_max = $self->{epoch_max}) or return;
580 my $last_commit = $self->{last_commit};
581 foreach my $i (0..$epoch_max) {
582 defined(my $cmt = $last_commit->[$i]) or next;
583 $last_commit->[$i] = undef;
584 last_epoch_commit($self, $i, $cmt);
590 $self->{bnote} or return;
592 my $barrier = { map { $_ => 1 } (0..$n) };
596 my ($self, $barrier) = @_;
597 my $bnote = $self->{bnote} or return;
599 while (scalar keys %$barrier) {
600 defined(my $l = readline($r)) or die "EOF on barrier_wait: $!";
601 $l =~ /\Abarrier (\d+)/ or die "bad line on barrier_wait: $l";
602 delete $barrier->{$1} or die "bad shard[$1] on barrier wait";
607 sub checkpoint ($;$) {
608 my ($self, $wait) = @_;
610 if (my $im = $self->{im}) {
617 my $shards = $self->{idx_shards};
619 my $dbh = $self->{mm}->{dbh};
621 # SQLite msgmap data is second in importance
624 # SQLite overview is third
625 $self->{over}->commit_lazy;
627 # Now deal with Xapian
629 my $barrier = $self->barrier_init(scalar @$shards);
631 # each shard needs to issue a barrier command
632 $_->remote_barrier for @$shards;
634 # wait for each Xapian shard
635 $self->barrier_wait($barrier);
637 $_->remote_commit for @$shards;
640 # last_commit is special, don't commit these until
641 # remote shards are done:
643 set_last_commits($self);
648 $self->{total_bytes} += $self->{transact_bytes};
649 $self->{transact_bytes} = 0;
652 # issue a write barrier to ensure all data is visible to other processes
653 # and read-only ops. Order of data importance is: git > SQLite > Xapian
655 sub barrier { checkpoint($_[0], 1) };
661 if (my $im = delete $self->{im}) {
662 eval { $im->done }; # PublicInbox::Import::done
663 $err .= "import done: $@\n" if $@;
666 eval { checkpoint($self) };
667 $err .= "checkpoint: $@\n" if $@;
669 if (my $mm = delete $self->{mm}) {
670 my $m = $err ? 'rollback' : 'commit';
671 eval { $mm->{dbh}->$m };
672 $err .= "msgmap $m: $@\n" if $@;
674 my $shards = delete $self->{idx_shards};
677 eval { $_->remote_close };
678 $err .= "shard close: $@\n" if $@;
681 eval { $self->{over}->disconnect };
682 $err .= "over disconnect: $@\n" if $@;
683 delete $self->{bnote};
684 my $nbytes = $self->{total_bytes};
685 $self->{total_bytes} = 0;
686 $self->lock_release(!!$nbytes) if $shards;
687 $self->{ibx}->git->cleanup;
691 sub fill_alternates ($$) {
692 my ($self, $epoch) = @_;
694 my $pfx = "$self->{ibx}->{inboxdir}/git";
695 my $all = "$self->{ibx}->{inboxdir}/all.git";
696 PublicInbox::Import::init_bare($all) unless -d $all;
697 my $info_dir = "$all/objects/info";
698 my $alt = "$info_dir/alternates";
702 open(my $fh, '<', $alt) or die "open < $alt: $!\n";
703 $mode = (stat($fh))[2] & 07777;
705 # we assign a sort score to every alternate and favor
706 # the newest (highest numbered) one because loose objects
707 # require scanning epochs and only the latest epoch is
708 # expected to see loose objects
710 my $other = 0; # in case admin adds non-epoch repos
712 if (m!\A\Q../../\E([0-9]+)\.git/objects\z!) {
718 } split(/\n+/, do { local $/; <$fh> });
721 foreach my $i (0..$epoch) {
722 my $dir = "../../git/$i.git/objects";
723 if (!exists($alt{$dir}) && -d "$pfx/$i.git") {
730 my ($fh, $tmp) = tempfile('alt-XXXXXXXX', DIR => $info_dir);
731 print $fh join("\n", sort { $alt{$b} <=> $alt{$a} } keys %alt), "\n"
732 or die "print $tmp: $!\n";
733 chmod($mode, $fh) or die "fchmod $tmp: $!\n";
734 close $fh or die "close $tmp $!\n";
735 rename($tmp, $alt) or die "rename $tmp => $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, $an) = tempfile('email-cur-XXXXXXXX', TMPDIR => 1);
817 print $ah $cur->as_string or die "print: $!";
818 close $ah or die "close: $!";
819 my ($bh, $bn) = tempfile('email-new-XXXXXXXX', TMPDIR => 1);
820 PublicInbox::Import::drop_unwanted_headers($new);
821 print $bh $new->as_string or die "print: $!";
822 close $bh or die "close: $!";
823 my $cmd = [ qw(diff -u), $an, $bn ];
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";
831 my ($self, $smsg) = @_;
832 if (my $im = $self->{im}) {
833 my $msg = $im->cat_blob($smsg->{blob});
836 # older message, should be in alternates
837 $self->{ibx}->msg_by_smsg($smsg);
840 sub content_exists ($$$) {
841 my ($self, $mime, $mid) = @_;
842 my $over = $self->{over};
843 my $chashes = content_hashes($mime);
845 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
846 my $msg = get_blob($self, $smsg);
847 if (!defined($msg)) {
848 warn "broken smsg for $mid\n";
851 my $cur = PublicInbox::Eml->new($msg);
852 return 1 if content_matches($chashes, $cur);
854 # XXX DEBUG_DIFF is experimental and may be removed
855 diff($mid, $cur, $mime) if $ENV{DEBUG_DIFF};
862 if (my $shards = $self->{idx_shards}) {
863 $_->atfork_child foreach @$shards;
865 if (my $im = $self->{im}) {
868 die "unexpected mm" if $self->{mm};
869 close $self->{bnote}->[0] or die "close bnote[0]: $!\n";
873 sub reindex_checkpoint ($$) {
874 my ($self, $sync) = @_;
876 $self->{ibx}->git->cleanup; # *async_wait
877 ${$sync->{need_checkpoint}} = 0;
878 $sync->{mm_tmp}->atfork_prepare;
879 $self->done; # release lock
881 if (my $pr = $sync->{-opt}->{-progress}) {
882 $pr->(sprintf($sync->{-regen_fmt}, ${$sync->{nr}}));
885 # allow -watch or -mda to write...
886 $self->idx_init; # reacquire lock
887 $sync->{mm_tmp}->atfork_parent;
890 sub index_oid { # cat_async callback
891 my ($bref, $oid, $type, $size, $arg) = @_;
892 return if $size == 0; # purged
894 my $eml = PublicInbox::Eml->new($$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 $over = $self->{over};
915 for my $mid (@$mids) {
916 ($num, $mid0) = $over->num_mid0_for_oid($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 ($$) {
981 my ($self, $epoch_max) = @_;
983 for (my $i = $epoch_max; $i >= 0; $i--) {
984 $heads->[$i] = last_epoch_commit($self, $i);
989 # returns a revision range for git-log(1)
990 sub log_range ($$$$$) {
991 my ($self, $sync, $git, $i, $tip) = @_;
992 my $opt = $sync->{-opt};
993 my $pr = $opt->{-progress} if (($opt->{verbose} || 0) > 1);
994 my $cur = $sync->{ranges}->[$i] or do {
995 $pr->("$i.git indexing all of $tip") if $pr;
996 return $tip; # all of it
999 # fast equality check to avoid (v)fork+execve overhead
1001 $sync->{ranges}->[$i] = undef;
1005 my $range = "$cur..$tip";
1006 $pr->("$i.git checking contiguity... ") if $pr;
1007 if (is_ancestor($git, $cur, $tip)) { # common case
1008 $pr->("OK\n") if $pr;
1009 my $n = $git->qx(qw(rev-list --count), $range);
1012 $sync->{ranges}->[$i] = undef;
1013 $pr->("$i.git has nothing new\n") if $pr;
1014 return; # nothing to do
1016 $pr->("$i.git has $n changes since $cur\n") if $pr;
1018 $pr->("FAIL\n") if $pr;
1020 discontiguous range: $range
1021 Rewritten history? (in $git->{git_dir})
1023 chomp(my $base = $git->qx('merge-base', $tip, $cur));
1025 $range = "$base..$tip";
1026 warn "found merge-base: $base\n"
1029 warn "discarding history at $cur\n";
1032 reindexing $git->{git_dir} starting at
1035 $sync->{unindex_range}->{$i} = "$base..$cur";
1040 sub sync_prepare ($$$) {
1041 my ($self, $sync, $epoch_max) = @_;
1042 my $pr = $sync->{-opt}->{-progress};
1044 my $head = $self->{ibx}->{ref_head} || 'refs/heads/master';
1046 # reindex stops at the current heads and we later rerun index_sync
1048 my $reindex_heads = last_commits($self, $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 my $all = $self->{ibx}->git;
1081 for my $oid (@leftovers) {
1082 $oid = unpack('H*', $oid);
1083 $self->{current_info} = "leftover $oid";
1084 $all->cat_async($oid, \&unindex_oid, $arg);
1086 $all->cat_async_wait;
1088 return 0 if (!$regen_max && !keys(%{$self->{unindex_range}}));
1090 # reindex should NOT see new commits anymore, if we do,
1091 # it's a problem and we need to notice it via die()
1092 my $pad = length($regen_max) + 1;
1093 $sync->{-regen_fmt} = "% ${pad}u/$regen_max\n";
1094 $sync->{nr} = \(my $nr = 0);
1095 return -1 if $sync->{reindex};
1096 $regen_max + $self->{mm}->num_highwater() || 0;
1099 sub unindex_oid_remote ($$$) {
1100 my ($self, $oid, $mid) = @_;
1101 my @removed = $self->{over}->remove_oid($oid, $mid);
1102 for my $num (@removed) {
1103 my $idx = idx_shard($self, $num % $self->{shards});
1104 $idx->remote_remove($oid, $num);
1108 sub unindex_oid ($$;$) { # git->cat_async callback
1109 my ($bref, $oid, $type, $size, $sync) = @_;
1110 my $self = $sync->{v2w};
1111 my $unindexed = $sync->{in_unindex} ? $sync->{unindexed} : undef;
1112 my $mm = $self->{mm};
1113 my $mids = mids(PublicInbox::Eml->new($bref));
1115 my $over = $self->{over};
1116 foreach my $mid (@$mids) {
1119 while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
1120 $gone{$smsg->{num}} = 1 if $oid eq $smsg->{blob};
1122 my $n = scalar(keys(%gone)) or next;
1124 warn "BUG: multiple articles linked to $oid\n",
1125 join(',',sort keys %gone), "\n";
1127 foreach my $num (keys %gone) {
1129 my $mid0 = $mm->mid_for($num);
1130 $unindexed->{$mid0} = $num;
1132 $mm->num_delete($num);
1134 unindex_oid_remote($self, $oid, $mid);
1138 # this is rare, it only happens when we get discontiguous history in
1139 # a mirror because the source used -purge or -edit
1140 sub unindex ($$$$) {
1141 my ($self, $sync, $git, $unindex_range) = @_;
1142 my $unindexed = $sync->{unindexed} //= {}; # $mid0 => $num
1143 my $before = scalar keys %$unindexed;
1144 # order does not matter, here:
1145 my @cmd = qw(log --raw -r
1146 --no-notes --no-color --no-abbrev --no-renames);
1147 my $fh = $git->popen(@cmd, $unindex_range);
1148 my $all = $self->{ibx}->git;
1149 local $sync->{in_unindex} = 1;
1151 /\A:\d{6} 100644 $OID ($OID) [AM]\tm$/o or next;
1152 $all->cat_async($1, \&unindex_oid, $sync);
1154 close $fh or die "git log failed: \$?=$?";
1155 $all->cat_async_wait;
1157 return unless $sync->{-opt}->{prune};
1158 my $after = scalar keys %$unindexed;
1159 return if $before == $after;
1161 # ensure any blob can not longer be accessed via dumb HTTP
1162 PublicInbox::Import::run_die(['git', "--git-dir=$git->{git_dir}",
1163 qw(-c gc.reflogExpire=now gc --prune=all --quiet)]);
1166 sub sync_ranges ($$$) {
1167 my ($self, $sync, $epoch_max) = @_;
1168 my $reindex = $sync->{reindex};
1170 return last_commits($self, $epoch_max) unless $reindex;
1171 return [] if ref($reindex) ne 'HASH';
1173 my $ranges = $reindex->{from}; # arrayref;
1174 if (ref($ranges) ne 'ARRAY') {
1175 die 'BUG: $reindex->{from} not an ARRAY';
1180 sub index_epoch ($$$) {
1181 my ($self, $sync, $i) = @_;
1183 my $git_dir = git_dir_n($self, $i);
1184 -d $git_dir or return; # missing epochs are fine
1185 my $git = PublicInbox::Git->new($git_dir);
1186 if (my $unindex_range = delete $sync->{unindex_range}->{$i}) { # rare
1187 unindex($self, $sync, $git, $unindex_range);
1189 defined(my $stk = $sync->{stacks}->[$i]) or return;
1190 $sync->{stacks}->[$i] = undef;
1191 my $all = $self->{ibx}->git;
1192 while (my ($f, $at, $ct, $oid) = $stk->pop_rec) {
1193 $self->{current_info} = "$i.git $oid";
1195 my $arg = { %$sync, autime => $at, cotime => $ct };
1196 if ($sync->{index_max_size}) {
1197 $all->check_async($oid, \&check_size, $arg);
1199 $all->cat_async($oid, \&index_oid, $arg);
1201 } elsif ($f eq 'd') {
1202 $all->cat_async($oid, \&unindex_oid, $sync);
1204 if (${$sync->{need_checkpoint}}) {
1205 reindex_checkpoint($self, $sync);
1208 $all->check_async_wait;
1209 $all->cat_async_wait;
1210 update_last_commit($self, $git, $i, $stk->{latest_cmt});
1213 # public, called by public-inbox-index
1215 my ($self, $opt) = @_;
1217 my $pr = $opt->{-progress};
1219 my $latest = git_dir_latest($self, \$epoch_max);
1220 return unless defined $latest;
1221 $self->idx_init($opt); # acquire lock
1222 fill_alternates($self, $epoch_max);
1223 $self->{over}->rethread_prepare($opt);
1225 need_checkpoint => \(my $bool = 0),
1226 unindex_range => {}, # EPOCH => oid_old..oid_new
1227 reindex => $opt->{reindex},
1231 $sync->{ranges} = sync_ranges($self, $sync, $epoch_max);
1232 if (sync_prepare($self, $sync, $epoch_max)) {
1233 # tmp_clone seems to fail if inside a transaction, so
1234 # we rollback here (because we opened {mm} for reading)
1235 # Note: we do NOT rely on DBI transactions for atomicity;
1236 # only for batch performance.
1237 $self->{mm}->{dbh}->rollback;
1238 $self->{mm}->{dbh}->begin_work;
1240 $self->{mm}->tmp_clone($self->{ibx}->{inboxdir});
1242 if ($sync->{index_max_size} = $self->{ibx}->{index_max_size}) {
1243 $sync->{index_oid} = \&index_oid;
1245 # work forwards through history
1246 index_epoch($self, $sync, $_) for (0..$epoch_max);
1247 $self->{over}->rethread_done($opt);
1250 if (my $nr = $sync->{nr}) {
1251 my $pr = $sync->{-opt}->{-progress};
1252 $pr->('all.git '.sprintf($sync->{-regen_fmt}, $$nr)) if $pr;
1255 # reindex does not pick up new changes, so we rerun w/o it:
1256 if ($opt->{reindex}) {
1259 delete @again{qw(rethread reindex -skip_lock)};
1260 index_sync($self, \%again);