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 # for XOVER, OVER in NNTP, and feeds/homepage/threads in PSGI
5 # Unlike Msgmap, this is an _UNSTABLE_ cache which can be
6 # tweaked/updated over time and rebuilt.
8 # Ghost messages (messages which are only referenced in References/In-Reply-To)
9 # are denoted by a negative NNTP article number.
10 package PublicInbox::OverIdx;
13 use base qw(PublicInbox::Over);
15 use DBI qw(:sql_types); # SQL_BLOB
16 use PublicInbox::MID qw/id_compress mids_for_index references/;
17 use PublicInbox::Smsg qw(subject_normalized);
18 use Compress::Zlib qw(compress);
19 use PublicInbox::Search;
24 my $dbh = $self->SUPER::dbh_new($self->{-no_fsync} ? 2 : 1);
26 # TRUNCATE reduces I/O compared to the default (DELETE)
27 # We do not use WAL since we're optimized for read-only ops,
28 # (and read-only requires SQLite 3.22.0 (2018-01-22)).
29 $dbh->do('PRAGMA journal_mode = TRUNCATE');
31 # 80000 pages (80MiB on SQLite <3.12.0, 320MiB on 3.12.0+)
32 # was found to be good in 2018 during the large LKML import
33 # at the time. This ought to be configurable based on HW
34 # and inbox size; I suspect it's overkill for many inboxes.
35 $dbh->do('PRAGMA cache_size = 80000');
43 my $self = $class->SUPER::new($f);
48 sub get_counter ($$) {
50 my $sth = $dbh->prepare_cached(<<'', undef, 1);
51 SELECT val FROM counter WHERE key = ? LIMIT 1
57 sub adj_counter ($$$) {
58 my ($self, $key, $op) = @_;
59 my $dbh = $self->{dbh};
60 my $sth = $dbh->prepare_cached(<<"");
61 UPDATE counter SET val = val $op 1 WHERE key = ?
65 get_counter($dbh, $key);
68 sub next_tid { adj_counter($_[0], 'thread', '+') }
69 sub next_ghost_num { adj_counter($_[0], 'ghost', '-') }
72 my ($self, $tbl, $id_col, $val_col, $val) = @_;
73 my $dbh = $self->{dbh};
74 my $in = $dbh->prepare_cached(<<"")->execute($val);
75 INSERT OR IGNORE INTO $tbl ($val_col) VALUES (?)
78 my $sth = $dbh->prepare_cached(<<"", undef, 1);
79 SELECT $id_col FROM $tbl WHERE $val_col = ? LIMIT 1
84 $dbh->last_insert_id(undef, undef, $tbl, $id_col);
89 my ($self, $path) = @_;
90 return unless defined $path && $path ne '';
91 id_for($self, 'subject', 'sid', 'path' => $path);
95 my ($self, $mid) = @_;
96 id_for($self, 'msgid', 'id', 'mid' => $mid);
100 my ($self, $num, $tid_ref) = @_;
101 my $dbh = $self->{dbh};
103 my $sth = $dbh->prepare_cached(<<'', undef, 1);
104 SELECT tid FROM over WHERE num = ? LIMIT 1
107 $$tid_ref = $sth->fetchrow_array; # may be undef
109 foreach (qw(over id2num)) {
110 $dbh->prepare_cached(<<"")->execute($num);
111 DELETE FROM $_ WHERE num = ?
116 # this includes ghosts
118 my ($self, $mid, $cols, $cb, @arg) = @_;
119 my $dbh = $self->{dbh};
122 I originally wanted to stuff everything into a single query:
124 SELECT over.* FROM over
125 LEFT JOIN id2num ON over.num = id2num.num
126 LEFT JOIN msgid ON msgid.id = id2num.id
127 WHERE msgid.mid = ? AND over.num >= ?
128 ORDER BY over.num ASC
131 But it's faster broken out (and we're always in a
132 transaction for subroutines in this file)
135 my $sth = $dbh->prepare_cached(<<'', undef, 1);
136 SELECT id FROM msgid WHERE mid = ? LIMIT 1
139 my $id = $sth->fetchrow_array;
140 defined $id or return;
143 $cols = join(',', map { $_ } @$cols);
145 my $prev = get_counter($dbh, 'ghost');
147 $sth = $dbh->prepare_cached(<<"", undef, 1);
148 SELECT num FROM id2num WHERE id = ? AND num >= ?
152 $sth->execute($id, $prev);
153 my $nums = $sth->fetchall_arrayref;
154 my $nr = scalar(@$nums) or return;
155 $prev = $nums->[-1]->[0];
157 $sth = $dbh->prepare_cached(<<"", undef, 1);
158 SELECT $cols FROM over WHERE over.num = ? LIMIT 1
161 $sth->execute($_->[0]);
162 my $smsg = $sth->fetchrow_hashref;
163 $smsg = PublicInbox::Over::load_from_row($smsg);
164 $cb->($self, $smsg, @arg) or return;
166 return if $nr != $lim;
170 sub _resolve_mid_to_tid {
171 my ($self, $smsg, $tid) = @_;
172 my $cur_tid = $smsg->{tid};
174 merge_threads($self, $$tid, $cur_tid);
175 } elsif ($cur_tid > $self->{min_tid}) {
177 } else { # rethreading, queue up dead ghosts
178 $$tid = next_tid($self);
179 my $num = $smsg->{num};
180 push(@{$self->{-ghosts_to_delete}}, $num) if $num < 0;
185 # this will create a ghost as necessary
186 sub resolve_mid_to_tid {
187 my ($self, $mid) = @_;
189 each_by_mid($self, $mid, ['tid'], \&_resolve_mid_to_tid, \$tid);
190 if (my $del = delete $self->{-ghosts_to_delete}) {
191 delete_by_num($self, $_) for @$del;
193 $tid // create_ghost($self, $mid);
197 my ($self, $mid) = @_;
198 my $id = mid2id($self, $mid);
199 my $num = next_ghost_num($self);
200 $num < 0 or die "ghost num is non-negative: $num\n";
201 my $tid = next_tid($self);
202 my $dbh = $self->{dbh};
203 $dbh->prepare_cached(<<'')->execute($num, $tid);
204 INSERT INTO over (num, tid) VALUES (?,?)
206 $dbh->prepare_cached(<<'')->execute($id, $num);
207 INSERT INTO id2num (id, num) VALUES (?,?)
213 my ($self, $winner_tid, $loser_tid) = @_;
214 return if $winner_tid == $loser_tid;
215 my $dbh = $self->{dbh};
216 $dbh->prepare_cached(<<'')->execute($winner_tid, $loser_tid);
217 UPDATE over SET tid = ? WHERE tid = ?
222 my ($self, $refs, $old_tid) = @_;
226 # first ref *should* be the thread root,
227 # but we can never trust clients to do the right thing
228 my $ref = $refs->[0];
229 $tid = resolve_mid_to_tid($self, $ref);
230 merge_threads($self, $tid, $old_tid) if defined $old_tid;
232 # the rest of the refs should point to this tid:
233 foreach my $i (1..$#$refs) {
235 my $ptid = resolve_mid_to_tid($self, $ref);
236 merge_threads($self, $tid, $ptid);
239 $tid = $old_tid // next_tid($self);
244 sub parse_references ($$$) {
245 my ($smsg, $hdr, $mids) = @_;
246 my $refs = references($hdr);
247 push(@$refs, @$mids) if scalar(@$mids) > 1;
248 return $refs if scalar(@$refs) == 0;
250 # prevent circular references here:
251 my %seen = ( $smsg->{mid} => 1 );
253 foreach my $ref (@$refs) {
254 if (length($ref) > PublicInbox::MID::MAX_MID_SIZE) {
255 warn "References: <$ref> too long, ignoring\n";
258 push(@keep, $ref) unless $seen{$ref}++;
260 $smsg->{references} = '<'.join('> <', @keep).'>' if @keep;
264 # normalize subjects so they are suitable as pathnames for URLs
265 # XXX: consider for removal
266 sub subject_path ($) {
268 $subj = subject_normalized($subj);
269 $subj =~ s![^a-zA-Z0-9_\.~/\-]+!_!g;
274 my ($self, $eml, $smsg) = @_;
275 $smsg->{lines} = $eml->body_raw =~ tr!\n!\n!;
276 my $mids = mids_for_index($eml);
277 my $refs = parse_references($smsg, $eml, $mids);
278 my $subj = $smsg->{subject};
281 $xpath = subject_path($subj);
282 $xpath = id_compress($xpath);
284 my $dd = $smsg->to_doc_data;
287 add_over($self, [ @$smsg{qw(ts ds num)}, $mids, $refs, $xpath, $dd ]);
291 my ($self, $smsg, $mid, $refs, $old_tid, $v) = @_;
292 my $cur_tid = $smsg->{tid};
293 my $n = $smsg->{num};
294 die "num must not be zero for $mid" if !$n;
295 my $cur_valid = $cur_tid > $self->{min_tid};
297 if ($n > 0) { # regular mail
299 $$old_tid //= $cur_tid;
300 merge_threads($self, $$old_tid, $cur_tid);
302 $$old_tid //= next_tid($self);
304 } elsif ($n < 0) { # ghost
305 $$old_tid //= $cur_valid ? $cur_tid : next_tid($self);
306 link_refs($self, $refs, $$old_tid);
307 delete_by_num($self, $n);
314 my ($self, $values) = @_;
315 my ($ts, $ds, $num, $mids, $refs, $xpath, $ddd) = @$values;
320 delete_by_num($self, $num, \$old_tid);
321 $old_tid = undef if ($old_tid // 0) <= $self->{min_tid};
322 foreach my $mid (@$mids) {
324 each_by_mid($self, $mid, ['tid'], \&_add_over,
325 $mid, $refs, \$old_tid, \$v);
326 $v > 1 and warn "BUG: vivified multiple ($v) ghosts for $mid\n";
329 my $tid = $vivified ? $old_tid : link_refs($self, $refs, $old_tid);
330 my $sid = sid($self, $xpath);
331 my $dbh = $self->{dbh};
332 my $sth = $dbh->prepare_cached(<<'');
333 INSERT INTO over (num, tid, sid, ts, ds, ddd)
337 my @v = ($num, $tid, $sid, $ts, $ds);
338 foreach (@v) { $sth->bind_param(++$n, $_) }
339 $sth->bind_param(++$n, $ddd, SQL_BLOB);
341 $sth = $dbh->prepare_cached(<<'');
342 INSERT INTO id2num (id, num) VALUES (?,?)
344 foreach my $mid (@$mids) {
345 my $id = mid2id($self, $mid);
346 $sth->execute($id, $num);
351 my ($self, $smsg, $oid, $removed) = @_;
352 if (!defined($oid) || $smsg->{blob} eq $oid) {
353 delete_by_num($self, $smsg->{num});
354 push @$removed, $smsg->{num};
359 # returns number of removed messages in scalar context,
360 # array of removed article numbers in array context.
361 # $oid may be undef to match only on $mid
363 my ($self, $oid, $mid) = @_;
366 each_by_mid($self, $mid, ['ddd'], \&_remove_oid, $oid, $removed);
370 sub _num_mid0_for_oid {
371 my ($self, $smsg, $oid, $res) = @_;
372 my $blob = $smsg->{blob};
373 return 1 if (!defined($blob) || $blob ne $oid); # continue;
374 @$res = ($smsg->{num}, $smsg->{mid});
378 sub num_mid0_for_oid {
379 my ($self, $oid, $mid) = @_;
382 each_by_mid($self, $mid, ['ddd'], \&_num_mid0_for_oid, $oid, $res);
383 @$res, # ($num, $mid0);
390 CREATE TABLE IF NOT EXISTS over (
391 num INTEGER NOT NULL,
392 tid INTEGER NOT NULL,
396 ddd VARBINARY, /* doc-data-deflated */
400 $dbh->do('CREATE INDEX IF NOT EXISTS idx_tid ON over (tid)');
401 $dbh->do('CREATE INDEX IF NOT EXISTS idx_sid ON over (sid)');
402 $dbh->do('CREATE INDEX IF NOT EXISTS idx_ts ON over (ts)');
403 $dbh->do('CREATE INDEX IF NOT EXISTS idx_ds ON over (ds)');
406 CREATE TABLE IF NOT EXISTS counter (
407 key VARCHAR(8) PRIMARY KEY NOT NULL,
408 val INTEGER DEFAULT 0,
412 $dbh->do("INSERT OR IGNORE INTO counter (key) VALUES ('thread')");
413 $dbh->do("INSERT OR IGNORE INTO counter (key) VALUES ('ghost')");
416 CREATE TABLE IF NOT EXISTS subject (
417 sid INTEGER PRIMARY KEY AUTOINCREMENT,
418 path VARCHAR(40) NOT NULL,
423 CREATE TABLE IF NOT EXISTS id2num (
425 num INTEGER NOT NULL,
429 # performance critical:
430 $dbh->do('CREATE INDEX IF NOT EXISTS idx_inum ON id2num (num)');
431 $dbh->do('CREATE INDEX IF NOT EXISTS idx_id ON id2num (id)');
434 CREATE TABLE IF NOT EXISTS msgid (
435 id INTEGER PRIMARY KEY AUTOINCREMENT,
436 mid VARCHAR(244) NOT NULL,
444 delete $self->{txn} or return;
445 $self->{dbh}->commit;
450 return if $self->{txn};
451 my $dbh = $self->connect or return;
453 # $dbh->{Profile} = 2;
459 delete $self->{txn} or return;
460 $self->{dbh}->rollback;
465 die "in transaction" if $self->{txn};
466 $self->SUPER::disconnect;
471 unless (-r $self->{filename}) {
473 require File::Basename;
474 File::Path::mkpath(File::Basename::dirname($self->{filename}));
477 PublicInbox::Over::connect($self);
481 sub rethread_prepare {
482 my ($self, $opt) = @_;
483 return unless $opt->{rethread};
485 my $min = $self->{min_tid} = get_counter($self->{dbh}, 'thread') // 0;
486 my $pr = $opt->{-progress};
487 $pr->("rethread min THREADID ".($min + 1)."\n") if $pr && $min;
491 my ($self, $opt) = @_;
492 return unless $opt->{rethread} && $self->{txn};
493 defined(my $min = $self->{min_tid}) or croak('BUG: no min_tid');
494 my $dbh = $self->{dbh} or croak('BUG: no dbh');
495 my $rows = $dbh->selectall_arrayref(<<'', { Slice => {} }, $min);
496 SELECT num,tid FROM over WHERE num < 0 AND tid < ?
498 my $show_id = $dbh->prepare('SELECT id FROM id2num WHERE num = ?');
499 my $show_mid = $dbh->prepare('SELECT mid FROM msgid WHERE id = ?');
500 my $pr = $opt->{-progress};
504 $show_id->execute($r->{num});
505 while (defined(my $id = $show_id->fetchrow_array)) {
507 $show_mid->execute($id);
508 my $mid = $show_mid->fetchrow_array;
509 if (!defined($mid)) {
511 E: ghost NUM=$r->{num} ID=$id THREADID=$r->{tid} has no Message-ID
516 I: ghost $r->{num} <$mid> THREADID=$r->{tid} culled
519 delete_by_num($self, $r->{num});
521 $pr->("I: rethread culled $total ghosts\n") if $pr && $total;