1 # Copyright (C) 2018 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 package PublicInbox::V2Writable;
8 use Fcntl qw(:flock :DEFAULT);
9 use PublicInbox::SearchIdxPart;
10 use PublicInbox::SearchIdxSkeleton;
11 use PublicInbox::MIME;
13 use PublicInbox::Import;
14 use PublicInbox::MID qw(mids);
15 use PublicInbox::ContentId qw(content_id content_digest);
16 use PublicInbox::Inbox;
18 # an estimate of the post-packed size to the raw uncompressed size
19 my $PACKING_FACTOR = 0.4;
21 # assume 2 cores if GNU nproc(1) is not available
23 int($ENV{NPROC} || `nproc 2>/dev/null` || 2);
27 my ($class, $v2ibx, $creat) = @_;
28 my $dir = $v2ibx->{mainrepo} or die "no mainrepo in inbox\n";
29 my $lock_path = "$dir/inbox.lock";
33 File::Path::mkpath($dir);
34 open my $fh, '>>', $lock_path or
35 die "failed to open $lock_path: $!\n";
37 die "$dir does not exist\n";
42 my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION;
44 # always load existing partitions in case core count changes:
46 foreach my $part (<$xpfx/*>) {
47 -d $part && $part =~ m!/\d+\z! or next;
49 Search::Xapian::Database->new($part)->close;
54 $nparts = nproc() if ($nparts == 0);
58 im => undef, # PublicInbox::Import
59 xap_rw => undef, # PublicInbox::V2SearchIdx
61 partitions => $nparts,
63 lock_path => "$dir/inbox.lock",
64 # limit each repo to 1GB or so
65 rotate_bytes => int((1024 * 1024 * 1024) / $PACKING_FACTOR),
70 # returns undef on duplicate or spam
71 # mimics Import::add and wraps it for v2
73 my ($self, $mime, $check_cb) = @_;
77 $mime = $check_cb->($mime) or return;
80 # All pipes (> $^F) known to Perl 5.6+ have FD_CLOEXEC set,
81 # as does SQLite 3.4.1+ (released in 2007-07-20), and
82 # Xapian 1.3.2+ (released 2015-03-15).
83 # For the most part, we can spawn git-fast-import without
84 # leaking FDs to it...
88 my $num = num_for($self, $mime, \$mid0);
89 defined $num or return; # duplicate
90 defined $mid0 or die "BUG: $mid0 undefined\n";
91 my $im = $self->importer;
92 my $cmt = $im->add($mime);
93 $cmt = $im->get_mark($cmt);
94 my ($oid, $len, $msgref) = @{$im->{last_object}};
96 my $nparts = $self->{partitions};
97 my $part = $num % $nparts;
98 my $idx = $self->idx_part($part);
99 $idx->index_raw($len, $msgref, $num, $oid, $mid0);
100 my $n = $self->{transact_bytes} += $len;
101 if ($n > (PublicInbox::SearchIdx::BATCH_BYTES * $nparts)) {
109 my ($self, $mime, $mid0) = @_;
110 my $mids = mids($mime->header_obj);
112 my $mid = $mids->[0];
113 my $num = $self->{skel}->{mm}->mid_insert($mid);
114 if (defined $num) { # common case
119 # crap, Message-ID is already known, hope somebody just resent:
121 foreach my $m (@$mids) {
122 # read-only lookup now safe to do after above barrier
123 my $existing = $self->lookup_content($mime, $m);
125 warn "<$m> resent\n";
126 return; # easy, don't store duplicates
131 warn "<$mid> reused for mismatched content\n";
133 # try the rest of the mids
134 foreach my $i (1..$#$mids) {
136 $num = $self->{skel}->{mm}->mid_insert($m);
138 warn "alternative <$m> for <$mid> found\n";
144 # none of the existing Message-IDs are good, generate a new one:
145 num_for_harder($self, $mime, $mid0);
149 my ($self, $mime, $mid0) = @_;
151 my $hdr = $mime->header_obj;
152 my $dig = content_digest($mime);
153 $$mid0 = PublicInbox::Import::digest2mid($dig);
154 my $num = $self->{skel}->{mm}->mid_insert($$mid0);
155 unless (defined $num) {
156 # it's hard to spoof the last Received: header
157 my @recvd = $hdr->header_raw('Received');
158 $dig->add("Received: $_") foreach (@recvd);
159 $$mid0 = PublicInbox::Import::digest2mid($dig);
160 $num = $self->{skel}->{mm}->mid_insert($$mid0);
162 # fall back to a random Message-ID and give up determinism:
163 until (defined($num)) {
165 $$mid0 = PublicInbox::Import::digest2mid($dig);
166 warn "using random Message-ID <$$mid0> as fallback\n";
167 $num = $self->{skel}->{mm}->mid_insert($$mid0);
170 my @cur = $hdr->header_raw('Message-Id');
171 $hdr->header_set('Message-Id', "<$$mid0>", @cur);
176 my ($self, $part) = @_;
177 $self->{idx_parts}->[$part];
183 return if $self->{idx_parts};
184 my $ibx = $self->{-inbox};
186 # do not leak read-only FDs to child processes, we only have these
187 # FDs for duplicate detection so they should not be
188 # frequently activated.
189 delete $ibx->{$_} foreach (qw(git mm search));
191 # first time initialization, first we create the skeleton pipe:
192 my $skel = $self->{skel} = PublicInbox::SearchIdxSkeleton->new($self);
194 # need to create all parts before initializing msgmap FD
195 my $max = $self->{partitions} - 1;
196 my $idx = $self->{idx_parts} = [];
197 for my $i (0..$max) {
198 push @$idx, PublicInbox::SearchIdxPart->new($self, $i, $skel);
201 # Now that all subprocesses are up, we can open the FD for SQLite:
202 $skel->_msgmap_init->{dbh}->begin_work;
206 my ($self, $mime, $cmt_msg) = @_;
209 my $im = $self->importer;
210 my $ibx = $self->{-inbox};
211 my $srch = $ibx->search;
212 my $cid = content_id($mime);
213 my $skel = $self->{skel};
214 my $parts = $self->{idx_parts};
215 my $mm = $skel->{mm};
217 my $mids = mids($mime->header_obj);
218 foreach my $mid (@$mids) {
219 $srch->reopen->each_smsg_by_mid($mid, sub {
222 my $msg = $ibx->msg_by_smsg($smsg);
223 if (!defined($msg)) {
224 warn "broken smsg for $mid\n";
228 my $cur = PublicInbox::MIME->new($msg);
229 if (content_id($cur) eq $cid) {
230 $mm->num_delete($smsg->num);
231 # $removed should only be set once assuming
232 # no bugs in our deduplication code:
234 $removed->{mime} = $cur;
235 $im->remove(\$orig, $cmt_msg);
237 $removed->num; # memoize this for callers
239 my $oid = $smsg->{blob};
240 foreach my $idx (@$parts, $skel) {
241 $idx->remote_remove($oid, $mid);
253 my $im = delete $self->{im};
254 $im->done if $im; # PublicInbox::Import::done
255 $self->searchidx_checkpoint(0);
260 my $im = $self->{im};
261 $im->checkpoint if $im; # PublicInbox::Import::checkpoint
262 $self->searchidx_checkpoint(1);
265 # issue a write barrier to ensure all data is visible to other processes
266 # and read-only ops. Order of data importance is: git > SQLite > Xapian
270 if (my $im = $self->{im}) {
273 my $skel = $self->{skel};
274 my $parts = $self->{idx_parts};
275 if ($parts && $skel) {
276 my $dbh = $skel->{mm}->{dbh};
277 $dbh->commit; # SQLite data is second in importance
279 # Now deal with Xapian
280 $skel->barrier_init(scalar(@$parts));
281 # each partition needs to issue a barrier command to skel:
282 $_->barrier foreach @$parts;
284 $skel->barrier_wait; # wait for each Xapian partition
288 $self->{transact_bytes} = 0;
291 sub searchidx_checkpoint {
292 my ($self, $more) = @_;
294 # order matters, we can only close {skel} after all partitions
295 # are done because the partitions also write to {skel}
296 if (my $parts = $self->{idx_parts}) {
297 foreach my $idx (@$parts) {
298 $idx->remote_commit; # propagates commit to skel
299 $idx->remote_close unless $more;
301 delete $self->{idx_parts} unless $more;
304 if (my $skel = $self->{skel}) {
305 my $dbh = $skel->{mm}->{dbh};
311 delete $self->{skel};
314 $self->{transact_bytes} = 0;
318 my ($self, $new) = @_;
319 my $pfx = "$self->{-inbox}->{mainrepo}/git";
320 my $git_dir = "$pfx/$new.git";
321 die "$git_dir exists\n" if -e $git_dir;
322 my @cmd = (qw(git init --bare -q), $git_dir);
323 PublicInbox::Import::run_die(\@cmd);
325 my $all = "$self->{-inbox}->{mainrepo}/all.git";
327 @cmd = (qw(git init --bare -q), $all);
328 PublicInbox::Import::run_die(\@cmd);
329 @cmd = (qw/git config/, "--file=$all/config",
330 'repack.writeBitmaps', 'true');
331 PublicInbox::Import::run_die(\@cmd);
334 @cmd = (qw/git config/, "--file=$git_dir/config",
335 'include.path', '../../all.git/config');
336 PublicInbox::Import::run_die(\@cmd);
338 my $alt = "$all/objects/info/alternates";
339 my $new_obj_dir = "../../git/$new.git/objects";
342 open(my $fh, '<', $alt) or die "open < $alt: $!\n";
343 %alts = map { chomp; $_ => 1 } (<$fh>);
345 return $git_dir if $alts{$new_obj_dir};
346 open my $fh, '>>', $alt or die "open >> $alt: $!\n";
347 print $fh "$new_obj_dir\n" or die "print >> $alt: $!\n";
348 close $fh or die "close $alt: $!\n";
354 my $im = $self->{im};
356 if ($im->{bytes_added} < $self->{rotate_bytes}) {
361 $self->searchidx_checkpoint(1);
363 my $git_dir = $self->git_init(++$self->{max_git});
364 my $git = PublicInbox::Git->new($git_dir);
365 return $self->import_init($git, 0);
371 my $pfx = "$self->{-inbox}->{mainrepo}/git";
373 foreach my $git_dir (glob("$pfx/*.git")) {
374 $git_dir =~ m!/(\d+)\.git\z! or next;
382 if (defined $latest) {
383 my $git = PublicInbox::Git->new($latest);
384 my $packed_bytes = $git->packed_bytes;
385 if ($packed_bytes >= $self->{rotate_bytes}) {
388 $self->{max_git} = $max;
389 return $self->import_init($git, $packed_bytes);
392 $self->{max_git} = $new;
393 $latest = $self->git_init($new);
394 $self->import_init(PublicInbox::Git->new($latest), 0);
398 my ($self, $git, $packed_bytes) = @_;
399 my $im = PublicInbox::Import->new($git, undef, undef, $self->{-inbox});
400 $im->{bytes_added} = int($packed_bytes / $PACKING_FACTOR);
401 $im->{want_object_info} = 1;
402 $im->{lock_path} = $self->{lock_path};
403 $im->{path_type} = 'v2';
408 my ($self, $mime, $mid) = @_;
409 my $ibx = $self->{-inbox};
411 my $srch = $ibx->search->reopen;
412 my $cid = content_id($mime);
414 $srch->each_smsg_by_mid($mid, sub {
417 my $msg = $ibx->msg_by_smsg($smsg);
418 if (!defined($msg)) {
419 warn "broken smsg for $mid\n";
422 my $cur = PublicInbox::MIME->new($msg);
423 if (content_id($cur) eq $cid) {
424 $smsg->{mime} = $cur;
426 return 0; # break out of loop
435 if (my $parts = $self->{idx_parts}) {
436 $_->atfork_child foreach @$parts;
438 if (my $im = $self->{im}) {