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
22 my $NPROC = int($ENV{NPROC} || `nproc 2>/dev/null` || 2);
25 my ($class, $v2ibx, $creat) = @_;
26 my $dir = $v2ibx->{mainrepo} or die "no mainrepo in inbox\n";
30 File::Path::mkpath($dir);
32 die "$dir does not exist\n";
37 im => undef, # PublicInbox::Import
38 xap_rw => undef, # PublicInbox::V2SearchIdx
42 # limit each repo to 1GB or so
43 rotate_bytes => int((1024 * 1024 * 1024) / $PACKING_FACTOR),
48 # returns undef on duplicate or spam
49 # mimics Import::add and wraps it for v2
51 my ($self, $mime, $check_cb) = @_;
55 $mime = $check_cb->($mime) or return;
58 # All pipes (> $^F) known to Perl 5.6+ have FD_CLOEXEC set,
59 # as does SQLite 3.4.1+ (released in 2007-07-20), and
60 # Xapian 1.3.2+ (released 2015-03-15).
61 # For the most part, we can spawn git-fast-import without
62 # leaking FDs to it...
65 my $num = num_for($self, $mime);
66 defined $num or return; # duplicate
67 my $im = $self->importer;
68 my $cmt = $im->add($mime);
69 $cmt = $im->get_mark($cmt);
70 my $oid = $im->{last_object_id};
71 my ($len, $msgref) = @{$im->{last_object}};
73 my $nparts = $self->{partitions};
74 my $part = $num % $nparts;
75 my $idx = $self->idx_part($part);
76 $idx->index_raw($len, $msgref, $num, $oid);
77 my $n = $self->{transact_bytes} += $len;
78 if ($n > (PublicInbox::SearchIdx::BATCH_BYTES * $nparts)) {
86 my ($self, $mime) = @_;
87 my $mids = mids($mime->header_obj);
90 my $num = $self->{skel}->{mm}->mid_insert($mid);
91 return $num if defined($num); # common case
93 # crap, Message-ID is already known, hope somebody just resent:
94 $self->done; # write barrier, clears $self->{skel}
95 foreach my $m (@$mids) {
96 # read-only lookup now safe to do after above barrier
97 my $existing = $self->lookup_content($mime, $m);
100 return; # easy, don't store duplicates
105 warn "<$mid> reused for mismatched content\n";
108 # try the rest of the mids
109 foreach my $i (1..$#$mids) {
111 $num = $self->{skel}->{mm}->mid_insert($m);
113 warn "alternative <$m> for <$mid> found\n";
118 # none of the existing Message-IDs are good, generate a new one:
119 num_for_harder($self, $mime);
123 my ($self, $mime) = @_;
125 my $hdr = $mime->header_obj;
126 my $dig = content_digest($mime);
127 my $mid = $dig->clone->hexdigest . '@localhost';
128 my $num = $self->{skel}->{mm}->mid_insert($mid);
129 unless (defined $num) {
130 # it's hard to spoof the last Received: header
131 my @recvd = $hdr->header_raw('Received');
132 $dig->add("Received: $_") foreach (@recvd);
133 $mid = $dig->clone->hexdigest . '@localhost';
134 $num = $self->{skel}->{mm}->mid_insert($mid);
136 # fall back to a random Message-ID and give up determinism:
137 until (defined($num)) {
139 $mid = $dig->clone->hexdigest . '@localhost';
140 warn "using random Message-ID <$mid> as fallback\n";
141 $num = $self->{skel}->{mm}->mid_insert($mid);
144 my @cur = $hdr->header_raw('Message-Id');
145 $hdr->header_set('Message-Id', "<$mid>", @cur);
150 my ($self, $part) = @_;
151 $self->{idx_parts}->[$part];
157 return if $self->{idx_parts};
158 my $ibx = $self->{-inbox};
160 # do not leak read-only FDs to child processes, we only have these
161 # FDs for duplicate detection so they should not be
162 # frequently activated.
163 delete $ibx->{$_} foreach (qw(git mm search));
165 # first time initialization, first we create the skeleton pipe:
166 my $skel = $self->{skel} = PublicInbox::SearchIdxSkeleton->new($self);
168 # need to create all parts before initializing msgmap FD
169 my $max = $self->{partitions} - 1;
170 my $idx = $self->{idx_parts} = [];
171 for my $i (0..$max) {
172 push @$idx, PublicInbox::SearchIdxPart->new($self, $i, $skel);
175 # Now that all subprocesses are up, we can open the FD for SQLite:
176 $skel->_msgmap_init->{dbh}->begin_work;
180 my ($self, $mime, $msg) = @_;
181 my $existing = $self->lookup_content($mime) or return;
183 # don't touch ghosts or already junked messages
184 return unless $existing->type eq 'mail';
186 # always write removals to the current (latest) git repo since
187 # we process chronologically
188 my $im = $self->importer;
189 my ($cmt, undef) = $im->remove($mime, $msg);
190 $cmt = $im->get_mark($cmt);
191 $self->unindex_msg($existing, $cmt);
196 my $im = delete $self->{im};
197 $im->done if $im; # PublicInbox::Import::done
198 $self->searchidx_checkpoint(0);
203 my $im = $self->{im};
204 $im->checkpoint if $im; # PublicInbox::Import::checkpoint
205 $self->searchidx_checkpoint(1);
208 sub searchidx_checkpoint {
209 my ($self, $more) = @_;
211 # order matters, we can only close {skel} after all partitions
212 # are done because the partitions also write to {skel}
213 if (my $parts = $self->{idx_parts}) {
214 foreach my $idx (@$parts) {
215 $idx->remote_commit; # propagates commit to skel
216 $idx->remote_close unless $more;
218 delete $self->{idx_parts} unless $more;
221 if (my $skel = $self->{skel}) {
222 my $dbh = $skel->{mm}->{dbh};
227 $skel->remote_commit; # XXX should be unnecessary...
229 delete $self->{skel};
232 $self->{transact_bytes} = 0;
236 my ($self, $new) = @_;
237 my $pfx = "$self->{-inbox}->{mainrepo}/git";
238 my $git_dir = "$pfx/$new.git";
239 die "$git_dir exists\n" if -e $git_dir;
240 my @cmd = (qw(git init --bare -q), $git_dir);
241 PublicInbox::Import::run_die(\@cmd);
242 @cmd = (qw/git config/, "--file=$git_dir/config",
243 'repack.writeBitmaps', 'true');
244 PublicInbox::Import::run_die(\@cmd);
246 my $all = "$self->{-inbox}->{mainrepo}/all.git";
248 @cmd = (qw(git init --bare -q), $all);
249 PublicInbox::Import::run_die(\@cmd);
252 my $alt = "$all/objects/info/alternates";
253 my $new_obj_dir = "../../git/$new.git/objects";
256 open(my $fh, '<', $alt) or die "open < $alt: $!\n";
257 %alts = map { chomp; $_ => 1 } (<$fh>);
259 return $git_dir if $alts{$new_obj_dir};
260 open my $fh, '>>', $alt or die "open >> $alt: $!\n";
261 print $fh "$new_obj_dir\n" or die "print >> $alt: $!\n";
262 close $fh or die "close $alt: $!\n";
268 my $im = $self->{im};
270 if ($im->{bytes_added} < $self->{rotate_bytes}) {
275 $self->searchidx_checkpoint(1);
277 my $git_dir = $self->git_init(++$self->{max_git});
278 my $git = PublicInbox::Git->new($git_dir);
279 return $self->import_init($git, 0);
285 my $pfx = "$self->{-inbox}->{mainrepo}/git";
287 foreach my $git_dir (glob("$pfx/*.git")) {
288 $git_dir =~ m!/(\d+)\.git\z! or next;
296 if (defined $latest) {
297 my $git = PublicInbox::Git->new($latest);
298 my $packed_bytes = $git->packed_bytes;
299 if ($packed_bytes >= $self->{rotate_bytes}) {
302 $self->{max_git} = $max;
303 return $self->import_init($git, $packed_bytes);
306 $self->{max_git} = $new;
307 $latest = $self->git_init($new);
308 $self->import_init(PublicInbox::Git->new($latest), 0);
312 my ($self, $git, $packed_bytes) = @_;
313 my $im = PublicInbox::Import->new($git, undef, undef, $self->{-inbox});
314 $im->{bytes_added} = int($packed_bytes / $PACKING_FACTOR);
315 $im->{want_object_id} = 1;
316 $im->{ssoma_lock} = 0;
317 $im->{path_type} = 'v2';
322 my ($self, $mime, $mid) = @_;
323 my $ibx = $self->{-inbox};
325 my $srch = $ibx->search;
326 my $cid = content_id($mime);
328 $srch->each_smsg_by_mid($mid, sub {
331 my $msg = $ibx->msg_by_smsg($smsg);
332 if (!defined($msg)) {
333 warn "broken smsg for $mid\n";
336 my $cur = PublicInbox::MIME->new($msg);
337 if (content_id($cur) eq $cid) {
338 $smsg->{mime} = $cur;
340 return 0; # break out of loop
349 if (my $parts = $self->{idx_parts}) {
350 $_->atfork_child foreach @$parts;
352 if (my $im = $self->{im}) {