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 if (my $ibx = $self->{ibx}) {
69 # always load existing shards in case core count changes:
70 # Also, shard count may change while -watch is running
71 my $srch = $ibx->search or return 0;
72 delete $ibx->{search};
74 } else { # ExtSearchIdx
75 $self->{nshard} // do {
76 if ($self->xdb_sharded) {
77 $self->{nshard} // die 'BUG: {nshard} unset';
86 # $creat may be any true value, or 0/undef. A hashref is true,
87 # and $creat->{nproc} may be set to an integer
88 my ($class, $v2ibx, $creat) = @_;
89 $v2ibx = PublicInbox::InboxWritable->new($v2ibx);
90 my $dir = $v2ibx->assert_usable_dir;
94 File::Path::mkpath($dir);
96 die "$dir does not exist\n";
99 $v2ibx->umask_prepare;
101 my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION;
104 im => undef, # PublicInbox::Import
110 oidx => PublicInbox::OverIdx->new("$xpfx/over.sqlite3"),
111 lock_path => "$dir/inbox.lock",
112 # limit each git repo (epoch) to 1GB or so
113 rotate_bytes => int((1024 * 1024 * 1024) / $PACKING_FACTOR),
114 last_commit => [], # git epoch -> commit
116 $self->{oidx}->{-no_fsync} = 1 if $v2ibx->{-no_fsync};
117 $self->{shards} = count_shards($self) || nproc_shards($creat);
123 my ($self, $shards, $skip_epoch, $skip_artnum) = @_;
124 if (defined $shards) {
125 $self->{parallel} = 0 if $shards == 0;
126 $self->{shards} = $shards if $shards > 0;
129 $self->{mm}->skip_artnum($skip_artnum) if defined $skip_artnum;
131 $self->{ibx}->git_dir_latest(\$epoch_max);
132 if (defined $skip_epoch && $epoch_max == -1) {
133 $epoch_max = $skip_epoch;
135 $self->git_init($epoch_max >= 0 ? $epoch_max : 0);
139 # returns undef on duplicate or spam
140 # mimics Import::add and wraps it for v2
142 my ($self, $eml, $check_cb) = @_;
143 $self->{ibx}->with_umask(\&_add, $self, $eml, $check_cb);
147 my ($self, $num) = @_;
148 $self->{idx_shards}->[$num % scalar(@{$self->{idx_shards}})];
151 # indexes a message, returns true if checkpointing is needed
153 my ($self, $msgref, $mime, $smsg) = @_;
154 $smsg->{bytes} = $smsg->{raw_bytes} + crlf_adjust($$msgref);
155 $self->{oidx}->add_overview($mime, $smsg);
156 my $idx = idx_shard($self, $smsg->{num});
157 $idx->index_raw($msgref, $mime, $smsg);
158 my $n = $self->{transact_bytes} += $smsg->{raw_bytes};
159 $n >= $self->{batch_bytes};
163 my ($self, $mime, $check_cb) = @_;
167 $mime = $check_cb->($mime, $self->{ibx}) or return;
170 # All pipes (> $^F) known to Perl 5.6+ have FD_CLOEXEC set,
171 # as does SQLite 3.4.1+ (released in 2007-07-20), and
172 # Xapian 1.3.2+ (released 2015-03-15).
173 # For the most part, we can spawn git-fast-import without
174 # leaking FDs to it...
177 my ($num, $mid0) = v2_num_for($self, $mime);
178 defined $num or return; # duplicate
179 defined $mid0 or die "BUG: \$mid0 undefined\n";
180 my $im = $self->importer;
181 my $smsg = bless { mid => $mid0, num => $num }, 'PublicInbox::Smsg';
182 my $cmt = $im->add($mime, undef, $smsg); # sets $smsg->{ds|ts|blob}
183 $cmt = $im->get_mark($cmt);
184 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
186 my $msgref = delete $smsg->{-raw_email};
187 if (do_idx($self, $msgref, $mime, $smsg)) {
195 my ($self, $mime) = @_;
196 my $mids = mids($mime);
198 my $mid = $mids->[0];
199 my $num = $self->{mm}->mid_insert($mid);
200 if (defined $num) { # common case
204 # crap, Message-ID is already known, hope somebody just resent:
205 foreach my $m (@$mids) {
206 # read-only lookup now safe to do after above barrier
207 # easy, don't store duplicates
208 # note: do not add more diagnostic info here since
209 # it gets noisy on public-inbox-watch restarts
210 return () if content_exists($self, $mime, $m);
213 # AltId may pre-populate article numbers (e.g. X-Mail-Count
214 # or NNTP article number), use that article number if it's
216 my $altid = $self->{ibx}->{altid};
217 if ($altid && grep(/:file=msgmap\.sqlite3\z/, @$altid)) {
218 my $num = $self->{mm}->num_for($mid);
220 if (defined $num && !$self->{oidx}->get_art($num)) {
226 warn "<$mid> reused for mismatched content\n";
228 # try the rest of the mids
229 for(my $i = $#$mids; $i >= 1; $i--) {
231 $num = $self->{mm}->mid_insert($m);
233 warn "alternative <$m> for <$mid> found\n";
238 # none of the existing Message-IDs are good, generate a new one:
239 v2_num_for_harder($self, $mime);
242 sub v2_num_for_harder {
243 my ($self, $eml) = @_;
245 my $dig = content_digest($eml);
246 my $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
247 my $num = $self->{mm}->mid_insert($mid0);
248 unless (defined $num) {
249 # it's hard to spoof the last Received: header
250 my @recvd = $eml->header_raw('Received');
251 $dig->add("Received: $_") foreach (@recvd);
252 $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
253 $num = $self->{mm}->mid_insert($mid0);
255 # fall back to a random Message-ID and give up determinism:
256 until (defined($num)) {
258 $mid0 = PublicInbox::Import::digest2mid($dig, $eml);
259 warn "using random Message-ID <$mid0> as fallback\n";
260 $num = $self->{mm}->mid_insert($mid0);
263 PublicInbox::Import::append_mid($eml, $mid0);
267 sub _idx_init { # with_umask callback
268 my ($self, $opt) = @_;
269 $self->lock_acquire unless $opt && $opt->{-skip_lock};
270 $self->{oidx}->create;
272 # xcpdb can change shard count while -watch is idle
273 my $nshards = count_shards($self);
274 $self->{shards} = $nshards if $nshards && $nshards != $self->{shards};
275 $self->{batch_bytes} = $opt->{batch_size} //
276 $PublicInbox::SearchIdx::BATCH_BYTES;
277 $self->{batch_bytes} *= $self->{shards} if $self->{parallel};
279 # need to create all shards before initializing msgmap FD
280 # idx_shards must be visible to all forked processes
281 my $max = $self->{shards} - 1;
282 my $idx = $self->{idx_shards} = [];
283 push @$idx, PublicInbox::SearchIdxShard->new($self, $_) for (0..$max);
284 my $ibx = $self->{ibx} or return; # ExtIdxSearch
286 # Now that all subprocesses are up, we can open the FDs
288 my $mm = $self->{mm} = PublicInbox::Msgmap->new_file(
289 "$ibx->{inboxdir}/msgmap.sqlite3",
290 $ibx->{-no_fsync} ? 2 : 1);
291 $mm->{dbh}->begin_work;
294 sub parallel_init ($$) {
295 my ($self, $indexlevel) = @_;
296 if (($indexlevel // 'full') eq 'basic') {
297 $self->{parallel} = 0;
299 pipe(my ($r, $w)) or die "pipe failed: $!";
300 # pipe for barrier notifications doesn't need to be big,
302 fcntl($w, 1031, 4096) if $^O eq 'linux';
303 $self->{bnote} = [ $r, $w ];
310 my ($self, $opt) = @_;
311 return if $self->{idx_shards};
312 my $ibx = $self->{ibx};
314 # do not leak read-only FDs to child processes, we only have these
315 # FDs for duplicate detection so they should not be
316 # frequently activated.
317 delete @$ibx{qw(mm search)};
320 parallel_init($self, $ibx->{indexlevel});
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 $ibx = $self->{ibx};
332 my $pfx = "$ibx->{inboxdir}/git";
333 my $rewrites = []; # epoch => commit
334 my $max = $self->{epoch_max};
336 unless (defined($max)) {
337 defined(my $latest = $ibx->git_dir_latest(\$max)) or return;
338 $self->{epoch_max} = $max;
341 foreach my $i (0..$max) {
342 my $git_dir = "$pfx/$i.git";
344 my $git = PublicInbox::Git->new($git_dir);
345 my $im = $self->import_init($git, 0, 1);
346 $rewrites->[$i] = $im->replace_oids($mime, $replace_map);
352 sub content_hashes ($) {
354 my @chashes = ( content_hash($mime) );
356 # We still support Email::MIME, here, and
357 # Email::MIME->as_string doesn't always round-trip, so we may
358 # use a second content_hash
359 my $rt = content_hash(PublicInbox::Eml->new(\($mime->as_string)));
360 push @chashes, $rt if $chashes[0] ne $rt;
364 sub content_matches ($$) {
365 my ($chashes, $existing) = @_;
366 my $chash = content_hash($existing);
367 foreach (@$chashes) {
368 return 1 if $_ eq $chash
373 # used for removing or replacing (purging)
374 sub rewrite_internal ($$;$$$) {
375 my ($self, $old_eml, $cmt_msg, $new_eml, $sref) = @_;
377 my ($im, $need_reindex, $replace_map);
379 $replace_map = {}; # oid => sref
380 $need_reindex = [] if $new_eml;
382 $im = $self->importer;
384 my $oidx = $self->{oidx};
385 my $chashes = content_hashes($old_eml);
387 my $mids = mids($old_eml);
389 # We avoid introducing new blobs into git since the raw content
390 # can be slightly different, so we do not need the user-supplied
391 # message now that we have the mids and content_hash
395 foreach my $mid (@$mids) {
396 my %gone; # num => [ smsg, $mime, raw ]
398 while (my $smsg = $oidx->next_by_mid($mid, \$id, \$prev)) {
399 my $msg = get_blob($self, $smsg);
400 if (!defined($msg)) {
401 warn "broken smsg for $mid\n";
405 my $cur = PublicInbox::Eml->new($msg);
406 if (content_matches($chashes, $cur)) {
407 $gone{$smsg->{num}} = [ $smsg, $cur, \$orig ];
410 my $n = scalar keys %gone;
413 warn "BUG: multiple articles linked to <$mid>\n",
414 join(',', sort keys %gone), "\n";
416 foreach my $num (keys %gone) {
417 my ($smsg, $mime, $orig) = @{$gone{$num}};
418 # $removed should only be set once assuming
419 # no bugs in our deduplication code:
420 $removed = [ undef, $mime, $smsg ];
421 my $oid = $smsg->{blob};
423 $replace_map->{$oid} = $sref;
425 ($mark, undef) = $im->remove($orig, $cmt_msg);
426 $removed->[0] = $mark;
429 if ($need_reindex) { # ->replace
430 push @$need_reindex, $smsg;
431 } else { # ->purge or ->remove
432 $self->{mm}->num_delete($num);
434 unindex_oid_aux($self, $oid, $mid);
439 my $cmt = $im->get_mark($mark);
440 $self->{last_commit}->[$self->{epoch_max}] = $cmt;
442 if ($replace_map && scalar keys %$replace_map) {
443 my $rewrites = _replace_oids($self, $new_eml, $replace_map);
444 return { rewrites => $rewrites, need_reindex => $need_reindex };
446 defined($mark) ? $removed : undef;
449 # public (see PublicInbox::Import->remove), but note the 3rd element
450 # (retval[2]) is not part of the stable API shared with Import->remove
452 my ($self, $eml, $cmt_msg) = @_;
453 my $r = $self->{ibx}->with_umask(\&rewrite_internal,
454 $self, $eml, $cmt_msg);
455 defined($r) && defined($r->[0]) ? @$r: undef;
458 sub _replace ($$;$$) {
459 my ($self, $old_eml, $new_eml, $sref) = @_;
460 my $arg = [ $self, $old_eml, undef, $new_eml, $sref ];
461 my $rewritten = $self->{ibx}->with_umask(\&rewrite_internal,
462 $self, $old_eml, undef, $new_eml, $sref) or return;
464 my $rewrites = $rewritten->{rewrites};
465 # ->done is called if there are rewrites since we gc+prune from git
466 $self->idx_init if @$rewrites;
468 for my $i (0..$#$rewrites) {
469 defined(my $cmt = $rewrites->[$i]) or next;
470 $self->{last_commit}->[$i] = $cmt;
477 my ($self, $mime) = @_;
478 my $rewritten = _replace($self, $mime, undef, \'') or return;
479 $rewritten->{rewrites}
482 # returns the git object_id of $fh, does not write the object to FS
483 sub git_hash_raw ($$) {
484 my ($self, $raw) = @_;
485 # grab the expected OID we have to reindex:
486 pipe(my($in, $w)) or die "pipe: $!";
487 my $git_dir = $self->git->{git_dir};
488 my $cmd = ['git', "--git-dir=$git_dir", qw(hash-object --stdin)];
489 my $r = popen_rd($cmd, undef, { 0 => $in });
490 print $w $$raw or die "print \$w: $!";
491 close $w or die "close \$w: $!";
493 chomp(my $oid = <$r>);
494 close $r or die "git hash-object failed: $?";
495 $oid =~ /\A$OID\z/ or die "OID not expected: $oid";
499 sub _check_mids_match ($$$) {
500 my ($old_list, $new_list, $hdrs) = @_;
501 my %old_mids = map { $_ => 1 } @$old_list;
502 my %new_mids = map { $_ => 1 } @$new_list;
503 my @old = keys %old_mids;
504 my @new = keys %new_mids;
505 my $err = "$hdrs may not be changed when replacing\n";
506 die $err if scalar(@old) != scalar(@new);
507 delete @new_mids{@old};
508 delete @old_mids{@new};
509 die $err if (scalar(keys %old_mids) || scalar(keys %new_mids));
512 # Changing Message-IDs or References with ->replace isn't supported.
513 # The rules for dealing with messages with multiple or conflicting
514 # Message-IDs are pretty complex and rethreading hasn't been fully
516 sub check_mids_match ($$) {
517 my ($old, $new) = @_;
518 _check_mids_match(mids($old), mids($new), 'Message-ID(s)');
519 _check_mids_match(references($old), references($new),
520 'References/In-Reply-To');
525 my ($self, $old_mime, $new_mime) = @_;
527 check_mids_match($old_mime, $new_mime);
529 # mutt will always add Content-Length:, Status:, Lines: when editing
530 PublicInbox::Import::drop_unwanted_headers($new_mime);
532 my $raw = $new_mime->as_string;
533 my $expect_oid = git_hash_raw($self, \$raw);
534 my $rewritten = _replace($self, $old_mime, $new_mime, \$raw) or return;
535 my $need_reindex = $rewritten->{need_reindex};
537 # just in case we have bugs in deduplication code:
538 my $n = scalar(@$need_reindex);
540 my $list = join(', ', map {
541 "$_->{num}: <$_->{mid}>"
544 W: rewritten $n messages matching content of original message (expected: 1).
545 W: possible bug in public-inbox, NNTP article IDs and Message-IDs follow:
550 # make sure we really got the OID:
551 my ($blob, $type, $bytes) = $self->git->check($expect_oid);
552 $blob eq $expect_oid or die "BUG: $expect_oid not found after replace";
554 # don't leak FDs to Xapian:
557 # reindex modified messages:
558 for my $smsg (@$need_reindex) {
559 my $new_smsg = bless {
564 }, 'PublicInbox::Smsg';
565 my $sync = { autime => $smsg->{ds}, cotime => $smsg->{ts} };
566 $new_smsg->populate($new_mime, $sync);
567 do_idx($self, \$raw, $new_mime, $new_smsg);
569 $rewritten->{rewrites};
572 sub last_epoch_commit ($$;$) {
573 my ($self, $i, $cmt) = @_;
574 my $v = PublicInbox::Search::SCHEMA_VERSION();
575 $self->{mm}->last_commit_xap($v, $i, $cmt);
578 sub set_last_commits ($) { # this is NOT for ExtSearchIdx
580 defined(my $epoch_max = $self->{epoch_max}) or return;
581 my $last_commit = $self->{last_commit};
582 foreach my $i (0..$epoch_max) {
583 defined(my $cmt = $last_commit->[$i]) or next;
584 $last_commit->[$i] = undef;
585 last_epoch_commit($self, $i, $cmt);
591 $self->{bnote} or return;
593 my $barrier = { map { $_ => 1 } (0..$n) };
597 my ($self, $barrier) = @_;
598 my $bnote = $self->{bnote} or return;
600 while (scalar keys %$barrier) {
601 defined(my $l = readline($r)) or die "EOF on barrier_wait: $!";
602 $l =~ /\Abarrier (\d+)/ or die "bad line on barrier_wait: $l";
603 delete $barrier->{$1} or die "bad shard[$1] on barrier wait";
608 sub checkpoint ($;$) {
609 my ($self, $wait) = @_;
611 if (my $im = $self->{im}) {
618 my $shards = $self->{idx_shards};
620 my $mm = $self->{mm};
621 my $dbh = $mm->{dbh} if $mm;
623 # SQLite msgmap data is second in importance
624 $dbh->commit if $dbh;
626 # SQLite overview is third
627 $self->{oidx}->commit_lazy;
629 # Now deal with Xapian
631 my $barrier = barrier_init($self, scalar @$shards);
633 # each shard needs to issue a barrier command
634 $_->shard_barrier for @$shards;
636 # wait for each Xapian shard
637 barrier_wait($self, $barrier);
639 $_->shard_commit for @$shards;
642 my $midx = $self->{midx}; # misc index
643 $midx->commit_txn if $midx;
645 # last_commit is special, don't commit these until
646 # Xapian shards are done:
647 $dbh->begin_work if $dbh;
648 set_last_commits($self);
653 $midx->begin_txn if $midx;
655 $self->{total_bytes} += $self->{transact_bytes};
656 $self->{transact_bytes} = 0;
659 # issue a write barrier to ensure all data is visible to other processes
660 # and read-only ops. Order of data importance is: git > SQLite > Xapian
662 sub barrier { checkpoint($_[0], 1) };
664 # true if locked and active
665 sub active { !!$_[0]->{im} }
671 if (my $im = delete $self->{im}) {
672 eval { $im->done }; # PublicInbox::Import::done
673 $err .= "import done: $@\n" if $@;
676 eval { checkpoint($self) };
677 $err .= "checkpoint: $@\n" if $@;
679 if (my $mm = delete $self->{mm}) {
680 my $m = $err ? 'rollback' : 'commit';
681 eval { $mm->{dbh}->$m };
682 $err .= "msgmap $m: $@\n" if $@;
684 my $shards = delete $self->{idx_shards};
687 eval { $_->shard_close };
688 $err .= "shard close: $@\n" if $@;
691 eval { $self->{oidx}->dbh_close };
692 $err .= "over close: $@\n" if $@;
693 delete $self->{midx};
694 delete $self->{bnote};
695 my $nbytes = $self->{total_bytes};
696 $self->{total_bytes} = 0;
697 $self->lock_release(!!$nbytes) if $shards;
702 sub write_alternates ($$$) {
703 my ($info_dir, $mode, $out) = @_;
704 my $fh = File::Temp->new(TEMPLATE => 'alt-XXXXXXXX', DIR => $info_dir);
705 my $tmp = $fh->filename;
706 print $fh @$out or die "print $tmp: $!\n";
707 chmod($mode, $fh) or die "fchmod $tmp: $!\n";
708 close $fh or die "close $tmp $!\n";
709 my $alt = "$info_dir/alternates";
710 rename($tmp, $alt) or die "rename $tmp => $alt: $!\n";
711 $fh->unlink_on_destroy(0);
714 sub fill_alternates ($$) {
715 my ($self, $epoch) = @_;
717 my $pfx = "$self->{ibx}->{inboxdir}/git";
718 my $all = "$self->{ibx}->{inboxdir}/all.git";
719 PublicInbox::Import::init_bare($all) unless -d $all;
720 my $info_dir = "$all/objects/info";
721 my $alt = "$info_dir/alternates";
725 open(my $fh, '<', $alt) or die "open < $alt: $!\n";
726 $mode = (stat($fh))[2] & 07777;
728 # we assign a sort score to every alternate and favor
729 # the newest (highest numbered) one because loose objects
730 # require scanning epochs and only the latest epoch is
731 # expected to see loose objects
733 my $other = 0; # in case admin adds non-epoch repos
735 if (m!\A\Q../../\E([0-9]+)\.git/objects\z!) {
741 } split(/\n+/, do { local $/; <$fh> });
744 foreach my $i (0..$epoch) {
745 my $dir = "../../git/$i.git/objects";
746 if (!exists($alt{$dir}) && -d "$pfx/$i.git") {
752 write_alternates($info_dir, $mode,
753 [join("\n", sort { $alt{$b} <=> $alt{$a} } keys %alt), "\n"]);
757 my ($self, $epoch) = @_;
758 my $git_dir = "$self->{ibx}->{inboxdir}/git/$epoch.git";
759 PublicInbox::Import::init_bare($git_dir);
760 my @cmd = (qw/git config/, "--file=$git_dir/config",
761 'include.path', '../../all.git/config');
762 PublicInbox::Import::run_die(\@cmd);
763 fill_alternates($self, $epoch);
769 my $im = $self->{im};
771 if ($im->{bytes_added} < $self->{rotate_bytes}) {
778 my $git_dir = $self->git_init(++$self->{epoch_max});
779 my $git = PublicInbox::Git->new($git_dir);
780 return $self->import_init($git, 0);
785 my $latest = $self->{ibx}->git_dir_latest(\$max);
786 if (defined $latest) {
787 my $git = PublicInbox::Git->new($latest);
788 my $packed_bytes = $git->packed_bytes;
789 my $unpacked_bytes = $packed_bytes / $PACKING_FACTOR;
791 if ($unpacked_bytes >= $self->{rotate_bytes}) {
794 $self->{epoch_max} = $max;
795 return $self->import_init($git, $packed_bytes);
798 $self->{epoch_max} = $epoch;
799 $latest = $self->git_init($epoch);
800 $self->import_init(PublicInbox::Git->new($latest), 0);
804 my ($self, $git, $packed_bytes, $tmp) = @_;
805 my $im = PublicInbox::Import->new($git, undef, undef, $self->{ibx});
806 $im->{bytes_added} = int($packed_bytes / $PACKING_FACTOR);
807 $im->{lock_path} = undef;
808 $im->{path_type} = 'v2';
809 $self->{im} = $im unless $tmp;
815 my ($mid, $cur, $new) = @_;
817 my $ah = File::Temp->new(TEMPLATE => 'email-cur-XXXXXXXX', TMPDIR => 1);
818 print $ah $cur->as_string or die "print: $!";
819 $ah->flush or die "flush: $!";
820 PublicInbox::Import::drop_unwanted_headers($new);
821 my $bh = File::Temp->new(TEMPLATE => 'email-new-XXXXXXXX', TMPDIR => 1);
822 print $bh $new->as_string or die "print: $!";
823 $bh->flush or die "flush: $!";
824 my $cmd = [ qw(diff -u), $ah->filename, $bh->filename ];
825 print STDERR "# MID conflict <$mid>\n";
826 my $pid = spawn($cmd, undef, { 1 => 2 });
827 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 ($$$) {
841 my ($self, $mime, $mid) = @_;
842 my $oidx = $self->{oidx};
843 my $chashes = content_hashes($mime);
845 while (my $smsg = $oidx->next_by_mid($mid, \$id, \$prev)) {
846 my $msg = get_blob($self, $smsg);
847 if (!defined($msg)) {
848 warn "broken smsg for $mid\n";
851 my $cur = PublicInbox::Eml->new($msg);
852 return 1 if content_matches($chashes, $cur);
854 # XXX DEBUG_DIFF is experimental and may be removed
855 diff($mid, $cur, $mime) if $ENV{DEBUG_DIFF};
862 if (my $shards = $self->{idx_shards}) {
863 $_->atfork_child foreach @$shards;
865 if (my $im = $self->{im}) {
868 die "unexpected mm" if $self->{mm};
869 close $self->{bnote}->[0] or die "close bnote[0]: $!\n";
873 sub reindex_checkpoint ($$) {
874 my ($self, $sync) = @_;
876 $self->git->async_wait_all;
877 $self->update_last_commit($sync);
878 ${$sync->{need_checkpoint}} = 0;
879 my $mm_tmp = $sync->{mm_tmp};
880 $mm_tmp->atfork_prepare if $mm_tmp;
881 die 'BUG: {im} during reindex' if $self->{im};
882 if ($self->{ibx_map}) {
883 checkpoint($self, 1); # no need to release lock on pure index
885 $self->done; # release lock
888 if (my $pr = $sync->{-opt}->{-progress}) {
889 $pr->(sprintf($sync->{-regen_fmt}, ${$sync->{nr}}));
892 # allow -watch or -mda to write...
893 $self->idx_init($sync->{-opt}); # reacquire lock
894 $mm_tmp->atfork_parent if $mm_tmp;
897 sub index_oid { # cat_async callback
898 my ($bref, $oid, $type, $size, $arg) = @_;
899 my $self = $arg->{self};
900 local $self->{current_info} = "$self->{current_info} $oid";
901 return if $size == 0; # purged
903 my $eml = PublicInbox::Eml->new($$bref);
904 my $mids = mids($eml);
905 my $chash = content_hash($eml);
907 if (scalar(@$mids) == 0) {
908 warn "E: $oid has no Message-ID, skipping\n";
912 # {unindexed} is unlikely
913 if ((my $unindexed = $arg->{unindexed}) && scalar(@$mids) == 1) {
914 $num = delete($unindexed->{$mids->[0]});
917 $self->{mm}->mid_set($num, $mid0);
918 delete($arg->{unindexed}) if !keys(%$unindexed);
921 if (!defined($num)) { # reuse if reindexing (or duplicates)
922 my $oidx = $self->{oidx};
923 for my $mid (@$mids) {
924 ($num, $mid0) = $oidx->num_mid0_for_oid($oid, $mid);
925 last if defined $num;
928 $mid0 //= do { # is this a number we got before?
929 $num = $arg->{mm_tmp}->num_for($mids->[0]);
930 defined($num) ? $mids->[0] : undef;
932 if (!defined($num)) {
933 for (my $i = $#$mids; $i >= 1; $i--) {
934 $num = $arg->{mm_tmp}->num_for($mids->[$i]);
942 $arg->{mm_tmp}->num_delete($num);
943 } else { # never seen
944 $num = $self->{mm}->mid_insert($mids->[0]);
947 } else { # rare, try the rest of them, backwards
948 for (my $i = $#$mids; $i >= 1; $i--) {
949 $num = $self->{mm}->mid_insert($mids->[$i]);
957 if (!defined($num)) {
958 warn "E: $oid <", join('> <', @$mids), "> is a duplicate\n";
967 }, 'PublicInbox::Smsg';
968 $smsg->populate($eml, $arg);
969 if (do_idx($self, $bref, $eml, $smsg)) {
970 ${$arg->{need_checkpoint}} = 1;
972 ${$arg->{latest_cmt}} = $arg->{cur_cmt} // die 'BUG: {cur_cmt} missing';
975 # only update last_commit for $i on reindex iff newer than current
976 sub update_last_commit {
977 my ($self, $sync, $stk) = @_;
978 my $unit = $sync->{unit} // return;
979 my $latest_cmt = $stk ? $stk->{latest_cmt} : ${$sync->{latest_cmt}};
980 defined($latest_cmt) or return;
981 my $last = last_epoch_commit($self, $unit->{epoch});
982 if (defined $last && is_ancestor($self->git, $last, $latest_cmt)) {
983 my @cmd = (qw(rev-list --count), "$last..$latest_cmt");
984 chomp(my $n = $unit->{git}->qx(@cmd));
985 return if $n ne '' && $n == 0;
987 last_epoch_commit($self, $unit->{epoch}, $latest_cmt);
991 my ($self, $sync) = @_;
993 for (my $i = $sync->{epoch_max}; $i >= 0; $i--) {
994 $heads->[$i] = last_epoch_commit($self, $i);
999 # returns a revision range for git-log(1)
1000 sub log_range ($$$) {
1001 my ($sync, $unit, $tip) = @_;
1002 my $opt = $sync->{-opt};
1003 my $pr = $opt->{-progress} if (($opt->{verbose} || 0) > 1);
1004 my $i = $unit->{epoch};
1005 my $cur = $sync->{ranges}->[$i] or do {
1006 $pr->("$i.git indexing all of $tip\n") if $pr;
1007 return $tip; # all of it
1010 # fast equality check to avoid (v)fork+execve overhead
1012 $sync->{ranges}->[$i] = undef;
1016 my $range = "$cur..$tip";
1017 $pr->("$i.git checking contiguity... ") if $pr;
1018 my $git = $unit->{git};
1019 if (is_ancestor($sync->{self}->git, $cur, $tip)) { # common case
1020 $pr->("OK\n") if $pr;
1021 my $n = $git->qx(qw(rev-list --count), $range);
1024 $sync->{ranges}->[$i] = undef;
1025 $pr->("$i.git has nothing new\n") if $pr;
1026 return; # nothing to do
1028 $pr->("$i.git has $n changes since $cur\n") if $pr;
1030 $pr->("FAIL\n") if $pr;
1032 discontiguous range: $range
1033 Rewritten history? (in $git->{git_dir})
1035 chomp(my $base = $git->qx('merge-base', $tip, $cur));
1037 $range = "$base..$tip";
1038 warn "found merge-base: $base\n"
1041 warn "discarding history at $cur\n";
1044 reindexing $git->{git_dir}
1047 # $cur^0 may no longer exist if pruned by git
1048 if ($git->qx(qw(rev-parse -q --verify), "$cur^0")) {
1049 $unit->{unindex_range} = "$base..$cur";
1050 } elsif ($base && $git->qx(qw(rev-parse -q --verify), $base)) {
1051 $unit->{unindex_range} = "$base..";
1053 warn "W: unable to unindex before $range\n";
1059 # overridden by ExtSearchIdx
1060 sub artnum_max { $_[0]->{mm}->num_highwater }
1062 sub sync_prepare ($$) {
1063 my ($self, $sync) = @_;
1064 $sync->{ranges} = sync_ranges($self, $sync);
1065 my $pr = $sync->{-opt}->{-progress};
1067 my $head = $sync->{ibx}->{ref_head} || 'HEAD';
1070 ($pfx) = ($sync->{ibx}->{inboxdir} =~ m!([^/]+)\z!g);
1071 $pfx //= $sync->{ibx}->{inboxdir};
1074 # reindex stops at the current heads and we later rerun index_sync
1076 my $reindex_heads = $self->last_commits($sync) if $sync->{reindex};
1078 if ($sync->{max_size} = $sync->{-opt}->{max_size}) {
1079 $sync->{index_oid} = $self->can('index_oid');
1081 for (my $i = $sync->{epoch_max}; $i >= 0; $i--) {
1082 my $git_dir = $sync->{ibx}->git_dir_n($i);
1083 -d $git_dir or next; # missing epochs are fine
1084 my $git = PublicInbox::Git->new($git_dir);
1085 my $unit = { git => $git, epoch => $i };
1086 if ($reindex_heads) {
1087 $head = $reindex_heads->[$i] or next;
1089 chomp(my $tip = $git->qx(qw(rev-parse -q --verify), $head));
1090 next if $?; # new repo
1092 my $range = log_range($sync, $unit, $tip) or next;
1093 # can't use 'rev-list --count' if we use --diff-filter
1094 $pr->("$pfx $i.git counting $range ... ") if $pr;
1095 # Don't bump num_highwater on --reindex by using {D}.
1096 # We intentionally do NOT use {D} in the non-reindex case
1097 # because we want NNTP article number gaps from unindexed
1098 # messages to show up in mirrors, too.
1099 $sync->{D} //= $sync->{reindex} ? {} : undef; # OID_BIN => NR
1100 my $stk = log2stack($sync, $git, $range);
1101 return 0 if $sync->{quit};
1102 my $nr = $stk ? $stk->num_records : 0;
1103 $pr->("$nr\n") if $pr;
1104 $unit->{stack} = $stk; # may be undef
1105 unshift @{$sync->{todo}}, $unit;
1108 return 0 if $sync->{quit};
1110 # XXX this should not happen unless somebody bypasses checks in
1111 # our code and blindly injects "d" file history into git repos
1112 if (my @leftovers = keys %{delete($sync->{D}) // {}}) {
1113 warn('W: unindexing '.scalar(@leftovers)." leftovers\n");
1114 local $self->{current_info} = 'leftover ';
1115 my $unindex_oid = $self->can('unindex_oid');
1116 for my $oid (@leftovers) {
1117 last if $sync->{quit};
1118 $oid = unpack('H*', $oid);
1119 my $req = { %$sync, oid => $oid };
1120 $self->git->cat_async($oid, $unindex_oid, $req);
1122 $self->git->cat_async_wait;
1124 return 0 if $sync->{quit};
1126 $sync->{-regen_fmt} = "%u/?\n";
1130 # reindex should NOT see new commits anymore, if we do,
1131 # it's a problem and we need to notice it via die()
1132 my $pad = length($regen_max) + 1;
1133 $sync->{-regen_fmt} = "% ${pad}u/$regen_max\n";
1134 $sync->{nr} = \(my $nr = 0);
1135 return -1 if $sync->{reindex};
1136 $regen_max + $self->artnum_max || 0;
1139 sub unindex_oid_aux ($$$) {
1140 my ($self, $oid, $mid) = @_;
1141 my @removed = $self->{oidx}->remove_oid($oid, $mid);
1142 for my $num (@removed) {
1143 my $idx = idx_shard($self, $num);
1144 $idx->shard_remove($oid, $num);
1148 sub unindex_oid ($$;$) { # git->cat_async callback
1149 my ($bref, $oid, $type, $size, $sync) = @_;
1150 my $self = $sync->{self};
1151 local $self->{current_info} = "$self->{current_info} $oid";
1152 my $unindexed = $sync->{in_unindex} ? $sync->{unindexed} : undef;
1153 my $mm = $self->{mm};
1154 my $mids = mids(PublicInbox::Eml->new($bref));
1156 my $oidx = $self->{oidx};
1157 foreach my $mid (@$mids) {
1160 while (my $smsg = $oidx->next_by_mid($mid, \$id, \$prev)) {
1161 $gone{$smsg->{num}} = 1 if $oid eq $smsg->{blob};
1163 my $n = scalar(keys(%gone)) or next;
1165 warn "BUG: multiple articles linked to $oid\n",
1166 join(',',sort keys %gone), "\n";
1168 foreach my $num (keys %gone) {
1170 my $mid0 = $mm->mid_for($num);
1171 $unindexed->{$mid0} = $num;
1173 $mm->num_delete($num);
1175 unindex_oid_aux($self, $oid, $mid);
1179 sub git { $_[0]->{ibx}->git }
1181 # this is rare, it only happens when we get discontiguous history in
1182 # a mirror because the source used -purge or -edit
1183 sub unindex_todo ($$$) {
1184 my ($self, $sync, $unit) = @_;
1185 my $unindex_range = delete($unit->{unindex_range}) // return;
1186 my $unindexed = $sync->{unindexed} //= {}; # $mid0 => $num
1187 my $before = scalar keys %$unindexed;
1188 # order does not matter, here:
1189 my $fh = $unit->{git}->popen(qw(log --raw -r --no-notes --no-color
1190 --no-abbrev --no-renames), $unindex_range);
1191 local $sync->{in_unindex} = 1;
1192 my $unindex_oid = $self->can('unindex_oid');
1194 /\A:\d{6} 100644 $OID ($OID) [AM]\tm$/o or next;
1195 $self->git->cat_async($1, $unindex_oid, { %$sync, oid => $1 });
1197 close $fh or die "git log failed: \$?=$?";
1198 $self->git->cat_async_wait;
1200 return unless $sync->{-opt}->{prune};
1201 my $after = scalar keys %$unindexed;
1202 return if $before == $after;
1204 # ensure any blob can not longer be accessed via dumb HTTP
1205 PublicInbox::Import::run_die(['git',
1206 "--git-dir=$unit->{git}->{git_dir}",
1207 qw(-c gc.reflogExpire=now gc --prune=all --quiet)]);
1210 sub sync_ranges ($$) {
1211 my ($self, $sync) = @_;
1212 my $reindex = $sync->{reindex};
1213 return $self->last_commits($sync) unless $reindex;
1214 return [] if ref($reindex) ne 'HASH';
1216 my $ranges = $reindex->{from}; # arrayref;
1217 if (ref($ranges) ne 'ARRAY') {
1218 die 'BUG: $reindex->{from} not an ARRAY';
1223 sub index_xap_only { # git->cat_async callback
1224 my ($bref, $oid, $type, $size, $smsg) = @_;
1225 my $self = $smsg->{self};
1226 my $idx = idx_shard($self, $smsg->{num});
1227 $smsg->{raw_bytes} = $size;
1228 $idx->index_raw($bref, undef, $smsg);
1229 $self->{transact_bytes} += $size;
1232 sub index_xap_step ($$$;$) {
1233 my ($self, $sync, $beg, $step) = @_;
1234 my $end = $sync->{art_end};
1235 return if $beg > $end; # nothing to do
1237 $step //= $self->{shards};
1238 my $ibx = $self->{ibx};
1239 if (my $pr = $sync->{-opt}->{-progress}) {
1240 $pr->("Xapian indexlevel=$ibx->{indexlevel} ".
1241 "$beg..$end (% $step)\n");
1243 for (my $num = $beg; $num <= $end; $num += $step) {
1244 last if $sync->{quit};
1245 my $smsg = $ibx->over->get_art($num) or next;
1246 $smsg->{self} = $self;
1247 $ibx->git->cat_async($smsg->{blob}, \&index_xap_only, $smsg);
1248 if ($self->{transact_bytes} >= $self->{batch_bytes}) {
1249 ${$sync->{nr}} = $num;
1250 reindex_checkpoint($self, $sync);
1255 sub index_todo ($$$) {
1256 my ($self, $sync, $unit) = @_;
1257 return if $sync->{quit};
1258 unindex_todo($self, $sync, $unit);
1259 my $stk = delete($unit->{stack}) or return;
1260 my $all = $self->git;
1261 my $index_oid = $self->can('index_oid');
1262 my $unindex_oid = $self->can('unindex_oid');
1264 if ($unit->{git}->{git_dir} =~ m!/([^/]+)/git/([0-9]+\.git)\z!) {
1265 $pfx = "$1 $2"; # v2
1267 ($pfx) = ($unit->{git}->{git_dir} =~ m!/([^/]+)\z!g);
1268 $pfx //= $unit->{git}->{git_dir};
1270 local $self->{current_info} = "$pfx ";
1271 local $sync->{latest_cmt} = \(my $latest_cmt);
1272 local $sync->{unit} = $unit;
1273 while (my ($f, $at, $ct, $oid, $cmt) = $stk->pop_rec) {
1274 if ($sync->{quit}) {
1275 warn "waiting to quit...\n";
1276 $all->async_wait_all;
1277 $self->update_last_commit($sync);
1288 if ($sync->{max_size}) {
1289 $all->check_async($oid, \&check_size, $req);
1291 $all->cat_async($oid, $index_oid, $req);
1293 } elsif ($f eq 'd') {
1294 $all->cat_async($oid, $unindex_oid, $req);
1296 if (${$sync->{need_checkpoint}}) {
1297 reindex_checkpoint($self, $sync);
1300 $all->async_wait_all;
1301 $self->update_last_commit($sync, $stk);
1305 my ($self, $opt, $sync, $art_beg) = @_;
1306 my $seq = $opt->{sequential_shard};
1308 local $self->{parallel} = 0 if $seq;
1309 $self->idx_init($opt); # acquire lock
1310 if (my $art_end = $self->{ibx}->mm->max) {
1312 need_checkpoint => \(my $bool = 0),
1315 nr => \(my $nr = 0),
1316 -regen_fmt => "%u/?\n",
1318 $sync->{art_end} = $art_end;
1319 if ($seq || !$self->{parallel}) {
1320 my $shard_end = $self->{shards} - 1;
1321 for my $i (0..$shard_end) {
1322 last if $sync->{quit};
1323 index_xap_step($self, $sync, $art_beg + $i);
1324 if ($i != $shard_end) {
1325 reindex_checkpoint($self, $sync);
1328 } else { # parallel (maybe)
1329 index_xap_step($self, $sync, $art_beg, 1);
1332 $self->git->cat_async_wait;
1336 # public, called by public-inbox-index
1338 my ($self, $opt) = @_;
1340 return xapian_only($self, $opt) if $opt->{xapian_only};
1342 my $pr = $opt->{-progress};
1344 my $latest = $self->{ibx}->git_dir_latest(\$epoch_max);
1345 return unless defined $latest;
1347 my $seq = $opt->{sequential_shard};
1348 my $art_beg; # the NNTP article number we start xapian_only at
1349 my $idxlevel = $self->{ibx}->{indexlevel};
1350 local $self->{ibx}->{indexlevel} = 'basic' if $seq;
1352 $self->idx_init($opt); # acquire lock
1353 fill_alternates($self, $epoch_max);
1354 $self->{oidx}->rethread_prepare($opt);
1356 need_checkpoint => \(my $bool = 0),
1357 reindex => $opt->{reindex},
1360 ibx => $self->{ibx},
1361 epoch_max => $epoch_max,
1363 my $quit = PublicInbox::SearchIdx::quit_cb($sync);
1364 local $SIG{QUIT} = $quit;
1365 local $SIG{INT} = $quit;
1366 local $SIG{TERM} = $quit;
1368 if (sync_prepare($self, $sync)) {
1369 # tmp_clone seems to fail if inside a transaction, so
1370 # we rollback here (because we opened {mm} for reading)
1371 # Note: we do NOT rely on DBI transactions for atomicity;
1372 # only for batch performance.
1373 $self->{mm}->{dbh}->rollback;
1374 $self->{mm}->{dbh}->begin_work;
1376 $self->{mm}->tmp_clone($self->{ibx}->{inboxdir});
1378 # xapian_only works incrementally w/o --reindex
1379 if ($seq && !$opt->{reindex}) {
1380 $art_beg = $sync->{mm_tmp}->max || -1;
1384 # work forwards through history
1385 index_todo($self, $sync, $_) for @{delete($sync->{todo}) // []};
1386 $self->{oidx}->rethread_done($opt) unless $sync->{quit};
1389 if (my $nr = $sync->{nr}) {
1390 my $pr = $sync->{-opt}->{-progress};
1391 $pr->('all.git '.sprintf($sync->{-regen_fmt}, $$nr)) if $pr;
1395 # deal with Xapian shards sequentially
1396 if ($seq && delete($sync->{mm_tmp})) {
1397 if ($sync->{quit}) {
1400 $self->{ibx}->{indexlevel} = $idxlevel;
1401 xapian_only($self, $opt, $sync, $art_beg);
1402 $quit_warn = 1 if $sync->{quit};
1406 # --reindex on the command-line
1407 if (!$sync->{quit} && $opt->{reindex} &&
1408 !ref($opt->{reindex}) && $idxlevel ne 'basic') {
1409 $self->lock_acquire;
1410 my $s0 = PublicInbox::SearchIdx->new($self->{ibx}, 0, 0);
1411 if (my $xdb = $s0->idx_acquire) {
1412 my $n = $xdb->get_metadata('has_threadid');
1413 $xdb->set_metadata('has_threadid', '1') if $n ne '1';
1416 $self->lock_release;
1419 # reindex does not pick up new changes, so we rerun w/o it:
1420 if ($opt->{reindex} && !$sync->{quit}) {
1423 delete @again{qw(rethread reindex -skip_lock)};
1424 index_sync($self, \%again);
1425 $opt->{quit} = $again{quit}; # propagate to caller
1427 warn <<EOF if $quit_warn;
1428 W: interrupted, --xapian-only --reindex required upon restart