1 # Copyright (C) 2020-2021 all contributors <meta@public-inbox.org>
2 # License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
4 # Detached/external index cross inbox search indexing support
5 # read-write counterpart to PublicInbox::ExtSearch
7 # It's based on the same ideas as public-inbox-v2-format(5) using
8 # over.sqlite3 for dedupe and sharded Xapian. msgmap.sqlite3 is
9 # missing, so there is no Message-ID conflict resolution, meaning
10 # no NNTP support for now.
12 # v2 has a 1:1 mapping of index:inbox or msgmap for NNTP support.
13 # This is intended to be an M:N index:inbox mapping, but it'll likely
14 # be 1:N in common practice (M==1)
16 package PublicInbox::ExtSearchIdx;
19 use parent qw(PublicInbox::ExtSearch PublicInbox::Lock);
20 use Carp qw(croak carp);
21 use Scalar::Util qw(blessed);
22 use Sys::Hostname qw(hostname);
23 use POSIX qw(strftime);
24 use File::Glob qw(bsd_glob GLOB_NOSORT);
25 use PublicInbox::MultiGit;
26 use PublicInbox::Search;
27 use PublicInbox::SearchIdx qw(prepare_stack is_ancestor is_bad_blob);
28 use PublicInbox::OverIdx;
29 use PublicInbox::MiscIdx;
30 use PublicInbox::MID qw(mids);
31 use PublicInbox::V2Writable;
32 use PublicInbox::InboxWritable;
33 use PublicInbox::ContentHash qw(content_hash);
35 use PublicInbox::DS qw(now add_timer);
36 use DBI qw(:sql_types); # SQL_BLOB
39 my (undef, $dir, $opt) = @_;
40 my $l = $opt->{indexlevel} // 'full';
41 $l !~ $PublicInbox::SearchIdx::INDEXLEVELS and
42 die "invalid indexlevel=$l\n";
43 $l eq 'basic' and die "E: indexlevel=basic not yet supported\n";
45 xpfx => "$dir/ei".PublicInbox::Search::SCHEMA_VERSION,
47 creat => $opt->{creat},
48 ibx_map => {}, # (newsgroup//inboxdir) => $ibx
49 ibx_active => [], # by config section order
50 ibx_known => [], # by config section order
56 lock_path => "$dir/ei.lock",
58 $self->{shards} = $self->count_shards ||
59 nproc_shards({ nproc => $opt->{jobs} });
60 my $oidx = PublicInbox::OverIdx->new("$self->{xpfx}/over.sqlite3");
61 $self->{-no_fsync} = $oidx->{-no_fsync} = 1 if !$opt->{fsync};
62 $self->{oidx} = $oidx;
67 my ($self, $ibx, $types) = @_;
68 $self->{ibx_map}->{$ibx->eidx_key} //= do {
69 delete $self->{-ibx_ary_known}; # invalidate cache
70 delete $self->{-ibx_ary_active}; # invalidate cache
71 $types //= [ qw(active known) ];
73 push @{$self->{"ibx_$t"}}, $ibx;
79 sub _ibx_attach { # each_inbox callback
80 my ($ibx, $self, $types) = @_;
81 attach_inbox($self, $ibx, $types);
85 my ($self, $cfg, $ibxs) = @_;
89 for my $ibx (@$ibxs) {
90 $self->{ibx_map}->{$ibx->eidx_key} //= do {
91 push @{$self->{ibx_active}}, $ibx;
92 push @{$self->{ibx_known}}, $ibx;
97 delete $self->{-ibx_ary_known};
98 delete $self->{-ibx_ary_active};
101 $types //= [ qw(known active) ];
102 $cfg->each_inbox(\&_ibx_attach, $self, $types);
105 sub check_batch_limit ($) {
107 my $self = $req->{self};
108 my $new_smsg = $req->{new_smsg};
109 my $n = $self->{transact_bytes} += $new_smsg->{bytes};
111 # set flag for PublicInbox::V2Writable::index_todo:
112 ${$req->{need_checkpoint}} = 1 if $n >= $self->{batch_bytes};
115 sub apply_boost ($$) {
116 my ($req, $smsg) = @_;
117 my $id2pos = $req->{id2pos}; # index in ibx_sorted
118 my $xr3 = $req->{self}->{oidx}->get_xref3($smsg->{num}, 1);
120 $id2pos->{$a->[0]} <=> $id2pos->{$b->[0]}
122 $a->[1] <=> $b->[1] # break ties with {xnum}
124 my $new_smsg = $req->{new_smsg};
125 return if $xr3->[0]->[2] ne $new_smsg->oidbin; # loser
127 # replace the old smsg with the more boosted one
128 $new_smsg->{num} = $smsg->{num};
129 $new_smsg->populate($req->{eml}, $req);
130 $req->{self}->{oidx}->add_overview($req->{eml}, $new_smsg);
133 sub remove_doc ($$) {
134 my ($self, $docid) = @_;
135 $self->{oidx}->delete_by_num($docid);
136 $self->{oidx}->eidxq_del($docid);
137 $self->idx_shard($docid)->ipc_do('xdb_remove', $docid);
140 sub _unref_doc ($$$$$;$) {
141 my ($sync, $docid, $ibx, $xnum, $oidbin, $eml) = @_;
145 $docid = $smsg->{num};
147 if (defined($oidbin) && defined($xnum) && blessed($ibx) && $ibx->over) {
148 my $smsg = $ibx->over->get_art($xnum);
149 if ($smsg && $smsg->oidbin eq $oidbin) {
150 carp("BUG: (non-fatal) ".$ibx->eidx_key.
151 " #$xnum $smsg->{blob} still valid");
155 my $s = 'DELETE FROM xref3 WHERE oidbin = ?';
156 $s .= ' AND ibx_id = ?' if defined($ibx);
157 $s .= ' AND xnum = ?' if defined($xnum);
158 my $del = $sync->{self}->{oidx}->dbh->prepare_cached($s);
160 $del->bind_param(++$col, $oidbin, SQL_BLOB);
161 $del->bind_param(++$col, $ibx->{-ibx_id}) if $ibx;
162 $del->bind_param(++$col, $xnum) if defined($xnum);
164 my $xr3 = $sync->{self}->{oidx}->get_xref3($docid);
165 if (scalar(@$xr3) == 0) { # all gone
166 remove_doc($sync->{self}, $docid);
167 } else { # enqueue for reindex of remaining messages
169 my $ekey = $ibx->{-gc_eidx_key} // $ibx->eidx_key;
170 my $idx = $sync->{self}->idx_shard($docid);
171 $idx->ipc_do('remove_eidx_info', $docid, $ekey, $eml);
172 } # else: we can't remove_eidx_info in reindex-only path
174 # replace invalidated blob ASAP with something which should be
175 # readable since we may commit the transaction on checkpoint.
176 # eidxq processing will re-apply boost
177 $smsg //= $sync->{self}->{oidx}->get_art($docid);
178 my $hex = unpack('H*', $oidbin);
179 if ($smsg && $smsg->{blob} eq $hex) {
180 $xr3->[0] =~ /:([a-f0-9]{40,}+)\z/ or
181 die "BUG: xref $xr3->[0] has no OID";
182 $sync->{self}->{oidx}->update_blob($smsg, $1);
184 # yes, add, we'll need to re-apply boost
185 $sync->{self}->{oidx}->eidxq_add($docid);
191 my ($req, $smsg) = @_;
192 my $self = $req->{self};
193 my $docid = $smsg->{num};
194 my $oid = $req->{oid};
195 my $xibx = $req->{ibx};
196 my $eml = $req->{eml};
197 if (my $new_smsg = $req->{new_smsg}) { # 'm' on cross-posted message
198 my $eidx_key = $xibx->eidx_key;
199 my $xnum = $req->{xnum};
200 $self->{oidx}->add_xref3($docid, $xnum, $oid, $eidx_key);
201 my $idx = $self->idx_shard($docid);
202 $idx->ipc_do('add_eidx_info', $docid, $eidx_key, $eml);
203 apply_boost($req, $smsg) if $req->{boost_in_use};
204 } else { # 'd' no {xnum}
205 $self->git->async_wait_all;
206 $oid = pack('H*', $oid);
207 _unref_doc($req, $docid, $xibx, undef, $oid, $eml);
211 # called by V2Writable::sync_prepare
212 sub artnum_max { $_[0]->{oidx}->eidx_max }
214 sub index_unseen ($) {
216 my $new_smsg = $req->{new_smsg} or die 'BUG: {new_smsg} unset';
217 my $eml = delete $req->{eml};
218 $new_smsg->populate($eml, $req);
219 my $self = $req->{self};
220 my $docid = $self->{oidx}->adj_counter('eidx_docid', '+');
221 $new_smsg->{num} = $docid;
222 my $idx = $self->idx_shard($docid);
223 $self->{oidx}->add_overview($eml, $new_smsg);
224 my $oid = $new_smsg->{blob};
225 my $ibx = delete $req->{ibx} or die 'BUG: {ibx} unset';
226 $self->{oidx}->add_xref3($docid, $req->{xnum}, $oid, $ibx->eidx_key);
227 $idx->index_eml($eml, $new_smsg, $ibx->eidx_key);
228 check_batch_limit($req);
231 sub do_finalize ($) {
233 if (my $indexed = $req->{indexed}) { # duplicated messages
234 do_xpost($req, $_) for @$indexed;
235 } elsif (exists $req->{new_smsg}) { # totally unseen messsage
238 # `d' message was already unindexed in the v1/v2 inboxes,
239 # so it's too noisy to warn, here.
241 # cur_cmt may be undef for unindex_oid, set by V2Writable::index_todo
242 if (defined(my $cur_cmt = $req->{cur_cmt})) {
243 ${$req->{latest_cmt}} = $cur_cmt;
247 sub do_step ($) { # main iterator for adding messages to the index
249 my $self = $req->{self} // die 'BUG: {self} missing';
251 if (my $next_arg = $req->{next_arg}) {
252 if (my $smsg = $self->{oidx}->next_by_mid(@$next_arg)) {
253 $req->{cur_smsg} = $smsg;
254 $self->git->cat_async($smsg->{blob},
255 \&ck_existing, $req);
256 return; # ck_existing calls do_step
258 delete $req->{next_arg};
260 die "BUG: {cur_smsg} still set" if $req->{cur_smsg};
261 my $mid = shift(@{$req->{mids}}) // last;
263 $req->{next_arg} = [ $mid, \$id, \$prev ];
269 sub _blob_missing ($$) { # called when a known $smsg->{blob} is gone
270 my ($req, $smsg) = @_;
271 # xnum and ibx are unknown, we only call this when an entry from
272 # /ei*/over.sqlite3 is bad, not on entries from xap*/over.sqlite3
273 $req->{self}->git->async_wait_all;
274 _unref_doc($req, $smsg, undef, undef, $smsg->oidbin);
277 sub ck_existing { # git->cat_async callback
278 my ($bref, $oid, $type, $size, $req) = @_;
279 my $smsg = delete $req->{cur_smsg} or die 'BUG: {cur_smsg} missing';
280 if ($type eq 'missing') {
281 _blob_missing($req, $smsg);
282 } elsif (!is_bad_blob($oid, $type, $size, $smsg->{blob})) {
283 my $self = $req->{self} // die 'BUG: {self} missing';
284 local $self->{current_info} = "$self->{current_info} $oid";
285 my $cur = PublicInbox::Eml->new($bref);
286 if (content_hash($cur) eq $req->{chash}) {
287 push @{$req->{indexed}}, $smsg; # for do_xpost
288 } # else { index_unseen later }
293 # is the messages visible in the inbox currently being indexed?
294 # return the number if so
295 sub cur_ibx_xnum ($$;$) {
296 my ($req, $bref, $mismatch) = @_;
297 my $ibx = $req->{ibx} or die 'BUG: current {ibx} missing';
299 $req->{eml} = PublicInbox::Eml->new($bref);
300 $req->{chash} = content_hash($req->{eml});
301 $req->{mids} = mids($req->{eml});
302 for my $mid (@{$req->{mids}}) {
304 while (my $x = $ibx->over->next_by_mid($mid, \$id, \$prev)) {
305 return $x->{num} if $x->{blob} eq $req->{oid};
306 push @$mismatch, $x if $mismatch;
312 sub index_oid { # git->cat_async callback for 'm'
313 my ($bref, $oid, $type, $size, $req) = @_;
314 my $self = $req->{self};
315 local $self->{current_info} = "$self->{current_info} $oid";
316 return if is_bad_blob($oid, $type, $size, $req->{oid});
317 my $new_smsg = $req->{new_smsg} = bless {
319 }, 'PublicInbox::Smsg';
320 $new_smsg->set_bytes($$bref, $size);
323 $req->{xnum} = cur_ibx_xnum($req, $bref, $mismatch) // do {
325 warn "# mismatch $_->{blob}\n" for @$mismatch;
326 ${$req->{latest_cmt}} = $req->{cur_cmt} //
327 die "BUG: {cur_cmt} unset ($oid)\n";
333 sub unindex_oid { # git->cat_async callback for 'd'
334 my ($bref, $oid, $type, $size, $req) = @_;
335 my $self = $req->{self};
336 local $self->{current_info} = "$self->{current_info} $oid";
337 return if is_bad_blob($oid, $type, $size, $req->{oid});
338 return if defined(cur_ibx_xnum($req, $bref)); # was re-added
342 # overrides V2Writable::last_commits, called by sync_ranges via sync_prepare
344 my ($self, $sync) = @_;
346 my $ekey = $sync->{ibx}->eidx_key;
347 my $uv = $sync->{ibx}->uidvalidity;
348 for my $i (0..$sync->{epoch_max}) {
349 $heads->[$i] = $self->{oidx}->eidx_meta("lc-v2:$ekey//$uv;$i");
354 sub _ibx_index_reject ($) {
356 $ibx->mm // return 'unindexed, no msgmap.sqlite3';
357 $ibx->uidvalidity // return 'no UIDVALIDITY';
358 $ibx->over // return 'unindexed, no over.sqlite3';
362 sub _sync_inbox ($$$) {
363 my ($self, $sync, $ibx) = @_;
364 my $ekey = $ibx->eidx_key;
365 if (defined(my $err = _ibx_index_reject($ibx))) {
366 return "W: skipping $ekey ($err)";
369 $sync->{nr} = \(my $nr = 0);
370 my $v = $ibx->version;
372 $sync->{epoch_max} = $ibx->max_git_epoch // return;
373 sync_prepare($self, $sync); # or return # TODO: once MiscIdx is stable
375 my $uv = $ibx->uidvalidity;
376 my $lc = $self->{oidx}->eidx_meta("lc-v1:$ekey//$uv");
377 my $head = $ibx->mm->last_commit //
378 return "E: $ibx->{inboxdir} is not indexed";
379 my $stk = prepare_stack($sync, $lc ? "$lc..$head" : $head);
380 my $unit = { stack => $stk, git => $ibx->git };
381 push @{$sync->{todo}}, $unit;
383 return "E: $ekey unsupported inbox version (v$v)";
385 for my $unit (@{delete($sync->{todo}) // []}) {
386 last if $sync->{quit};
387 index_todo($self, $sync, $unit);
389 $self->{midx}->index_ibx($ibx) unless $sync->{quit};
390 $ibx->git->cleanup; # done with this inbox, now
394 sub eidx_gc_scan_inboxes ($$) {
395 my ($self, $sync) = @_;
396 my ($x3_doc, $ibx_ck);
398 $x3_doc = $self->{oidx}->dbh->prepare(<<EOM);
399 SELECT docid,xnum,oidbin FROM xref3 WHERE ibx_id = ?
401 $ibx_ck = $self->{oidx}->dbh->prepare(<<EOM);
402 SELECT ibx_id,eidx_key FROM inboxes
405 while (my ($ibx_id, $eidx_key) = $ibx_ck->fetchrow_array) {
406 next if $self->{ibx_map}->{$eidx_key};
407 $self->{midx}->remove_eidx_key($eidx_key);
408 warn "I: deleting messages for $eidx_key...\n";
409 $x3_doc->execute($ibx_id);
410 my $ibx = { -ibx_id => $ibx_id, -gc_eidx_key => $eidx_key };
411 while (my ($docid, $xnum, $oid) = $x3_doc->fetchrow_array) {
412 my $r = _unref_doc($sync, $docid, $ibx, $xnum, $oid);
413 $oid = unpack('H*', $oid);
414 $r = $r ? 'unref' : 'remove';
415 warn "I: $r #$docid $eidx_key $oid\n";
416 if (checkpoint_due($sync)) {
417 $x3_doc = $ibx_ck = undef;
418 reindex_checkpoint($self, $sync);
422 $self->{oidx}->dbh->do(<<'', undef, $ibx_id);
423 DELETE FROM inboxes WHERE ibx_id = ?
425 # drop last_commit info
427 $pat =~ s/([_%\\])/\\$1/g;
428 $self->{oidx}->dbh->do('PRAGMA case_sensitive_like = ON');
429 my $lc_i = $self->{oidx}->dbh->prepare(<<'');
430 SELECT key FROM eidx_meta WHERE key LIKE ? ESCAPE ?
432 $lc_i->execute("lc-%:$pat//%", '\\');
433 while (my ($key) = $lc_i->fetchrow_array) {
434 next if $key !~ m!\Alc-v[1-9]+:\Q$eidx_key\E//!;
435 warn "I: removing $key\n";
436 $self->{oidx}->dbh->do(<<'', undef, $key);
437 DELETE FROM eidx_meta WHERE key = ?
440 warn "I: $eidx_key removed\n";
444 sub eidx_gc_scan_shards ($$) { # TODO: use for lei/store
445 my ($self, $sync) = @_;
446 my $nr = $self->{oidx}->dbh->do(<<'');
447 DELETE FROM xref3 WHERE docid NOT IN (SELECT num FROM over)
449 warn "I: eliminated $nr stale xref3 entries\n" if $nr != 0;
450 reindex_checkpoint($self, $sync) if checkpoint_due($sync);
452 # fixup from old bugs:
453 $nr = $self->{oidx}->dbh->do(<<'');
454 DELETE FROM over WHERE num > 0 AND num NOT IN (SELECT docid FROM xref3)
456 warn "I: eliminated $nr stale over entries\n" if $nr != 0;
457 reindex_checkpoint($self, $sync) if checkpoint_due($sync);
459 $nr = $self->{oidx}->dbh->do(<<'');
460 DELETE FROM eidxq WHERE docid NOT IN (SELECT num FROM over)
462 warn "I: eliminated $nr stale reindex queue entries\n" if $nr != 0;
463 reindex_checkpoint($self, $sync) if checkpoint_due($sync);
465 my ($cur) = $self->{oidx}->dbh->selectrow_array(<<EOM);
466 SELECT MIN(num) FROM over WHERE num > 0
468 $cur // return; # empty
469 my ($r, $n, %active_shards);
472 $r = $self->{oidx}->dbh->selectcol_arrayref(<<"", undef, $cur);
473 SELECT num FROM over WHERE num >= ? ORDER BY num ASC LIMIT 10000
475 last unless scalar(@$r);
476 while (defined($n = shift @$r)) {
477 for my $i ($cur..($n - 1)) {
478 my $idx = idx_shard($self, $i);
479 $idx->ipc_do('xdb_remove_quiet', $i);
480 $active_shards{$idx} = $idx;
484 if (checkpoint_due($sync)) {
485 for my $idx (values %active_shards) {
486 $nr += $idx->ipc_do('nr_quiet_rm')
489 reindex_checkpoint($self, $sync);
492 warn "I: eliminated $nr stale Xapian documents\n" if $nr != 0;
496 my ($self, $opt) = @_;
497 $self->{cfg} or die "E: GC requires ->attach_config\n";
500 need_checkpoint => \(my $need_checkpoint = 0),
502 next_check => now() + 10,
503 checkpoint_unlocks => 1,
507 $self->idx_init($opt); # acquire lock via V2Writable::_idx_init
508 eidx_gc_scan_inboxes($self, $sync);
509 eidx_gc_scan_shards($self, $sync);
514 my ($self, $sync, $smsg) = @_;
515 my $ibx_id = delete($smsg->{ibx_id}) // die '{ibx_id} unset';
516 my $pos = $sync->{id2pos}->{$ibx_id} // die "$ibx_id no pos";
517 $self->{-ibx_ary_known}->[$pos] //
518 die "BUG: ibx for $smsg->{blob} not mapped"
521 sub _fd_constrained ($) {
523 $self->{-fd_constrained} //= do {
525 if (eval { require BSD::Resource; 1 }) {
526 my $NOFILE = BSD::Resource::RLIMIT_NOFILE();
527 ($soft, undef) = BSD::Resource::getrlimit($NOFILE);
529 chomp($soft = `sh -c 'ulimit -n'`);
531 if (defined($soft)) {
532 # $want is an estimate
533 my $want = scalar(@{$self->{ibx_active}}) + 64;
534 my $ret = $want > $soft;
537 RLIMIT_NOFILE=$soft insufficient (want: $want), will close DB handles early
542 warn "Unable to determine RLIMIT_NOFILE: $@\n";
548 sub _reindex_finalize ($$$) {
549 my ($req, $smsg, $eml) = @_;
550 my $sync = $req->{sync};
551 my $self = $sync->{self};
552 my $by_chash = delete $req->{by_chash} or die 'BUG: no {by_chash}';
553 my $nr = scalar(keys(%$by_chash)) or die 'BUG: no content hashes';
554 my $orig_smsg = $req->{orig_smsg} // die 'BUG: no {orig_smsg}';
555 my $docid = $smsg->{num} = $orig_smsg->{num};
556 $self->{oidx}->add_overview($eml, $smsg); # may rethread
557 check_batch_limit({ %$sync, new_smsg => $smsg });
558 my $chash0 = $smsg->{chash} // die "BUG: $smsg->{blob} no {chash}";
559 my $stable = delete($by_chash->{$chash0}) //
560 die "BUG: $smsg->{blob} chash missing";
561 my $idx = $self->idx_shard($docid);
562 my $top_smsg = pop @$stable;
563 $top_smsg == $smsg or die 'BUG: top_smsg != smsg';
564 my $ibx = _ibx_for($self, $sync, $smsg);
565 $idx->index_eml($eml, $smsg, $ibx->eidx_key);
566 for my $x (reverse @$stable) {
567 $ibx = _ibx_for($self, $sync, $x);
568 my $hdr = delete $x->{hdr} // die 'BUG: no {hdr}';
569 $idx->ipc_do('add_eidx_info', $docid, $ibx->eidx_key, $hdr);
571 return if $nr == 1; # likely, all good
573 $self->git->async_wait_all;
574 warn "W: #$docid split into $nr due to deduplication change\n";
576 for my $ary (values %$by_chash) {
577 for my $x (reverse @$ary) {
578 warn "removing #$docid xref3 $x->{blob}\n";
579 my $bin = $x->oidbin;
580 my $n = _unref_doc($sync, $docid, undef, undef, $bin);
581 die "BUG: $x->{blob} invalidated #$docid" if $n == 0;
583 my $x = pop(@$ary) // die "BUG: #$docid {by_chash} empty";
584 $x->{num} = delete($x->{xnum}) // die '{xnum} unset';
585 $ibx = _ibx_for($self, $sync, $x);
586 if (my $over = $ibx->over) {
587 my $e = $over->get_art($x->{num});
588 $e->{blob} eq $x->{blob} or die <<EOF;
589 $x->{blob} != $e->{blob} (${\$ibx->eidx_key}:$e->{num});
591 push @todo, $ibx, $e;
592 $over->dbh_close if _fd_constrained($self);
594 die "$ibx->{inboxdir}: over.sqlite3 unusable: $!\n";
598 while (my ($ibx, $e) = splice(@todo, 0, 2)) {
599 reindex_unseen($self, $sync, $ibx, $e);
603 sub _reindex_oid { # git->cat_async callback
604 my ($bref, $oid, $type, $size, $req) = @_;
605 my $sync = $req->{sync};
606 my $self = $sync->{self};
607 my $orig_smsg = $req->{orig_smsg} // die 'BUG: no {orig_smsg}';
608 my $expect_oid = $req->{xr3r}->[$req->{ix}]->[2];
609 my $docid = $orig_smsg->{num};
610 if (is_bad_blob($oid, $type, $size, $expect_oid)) {
611 my $oidbin = pack('H*', $expect_oid);
612 my $remain = _unref_doc($sync, $docid, undef, undef, $oidbin);
614 warn "W: #$docid ($oid) gone or corrupt\n";
615 } elsif (my $next_oid = $req->{xr3r}->[++$req->{ix}]->[2]) {
616 $self->git->cat_async($next_oid, \&_reindex_oid, $req);
618 warn "BUG: #$docid ($oid) gone (UNEXPECTED)\n";
622 my $ci = $self->{current_info};
623 local $self->{current_info} = "$ci #$docid $oid";
624 my $re_smsg = bless { blob => $oid }, 'PublicInbox::Smsg';
625 $re_smsg->set_bytes($$bref, $size);
626 my $eml = PublicInbox::Eml->new($bref);
627 $re_smsg->populate($eml, { autime => $orig_smsg->{ds},
628 cotime => $orig_smsg->{ts} });
629 my $chash = content_hash($eml);
630 $re_smsg->{chash} = $chash;
631 $re_smsg->{xnum} = $req->{xr3r}->[$req->{ix}]->[1];
632 $re_smsg->{ibx_id} = $req->{xr3r}->[$req->{ix}]->[0];
633 $re_smsg->{hdr} = $eml->header_obj;
634 push @{$req->{by_chash}->{$chash}}, $re_smsg;
635 if (my $next_oid = $req->{xr3r}->[++$req->{ix}]->[2]) {
636 $self->git->cat_async($next_oid, \&_reindex_oid, $req);
637 } else { # last $re_smsg is the highest priority xref3
638 local $self->{current_info} = "$ci #$docid";
639 _reindex_finalize($req, $re_smsg, $eml);
643 sub _reindex_smsg ($$$) {
644 my ($self, $sync, $smsg) = @_;
645 my $docid = $smsg->{num};
646 my $xr3 = $self->{oidx}->get_xref3($docid, 1);
647 if (scalar(@$xr3) == 0) { # _reindex_check_stale should've covered this
649 BUG? #$docid $smsg->{blob} is not referenced by inboxes during reindex
651 remove_doc($self, $docid);
655 # we sort {xr3r} in the reverse order of ibx_sorted so we can
656 # hit the common case in _reindex_finalize without rereading
657 # from git (or holding multiple messages in memory).
658 my $id2pos = $sync->{id2pos}; # index in ibx_sorted
660 $id2pos->{$b->[0]} <=> $id2pos->{$a->[0]}
662 $b->[1] <=> $a->[1] # break ties with {xnum}
664 @$xr3 = map { [ $_->[0], $_->[1], unpack('H*', $_->[2]) ] } @$xr3;
665 my $req = { orig_smsg => $smsg, sync => $sync, xr3r => $xr3, ix => 0 };
666 $self->git->cat_async($xr3->[$req->{ix}]->[2], \&_reindex_oid, $req);
669 sub checkpoint_due ($) {
671 ${$sync->{need_checkpoint}} || (now() > $sync->{next_check});
675 # I've copied FS images and only changed the hostname before,
676 # so prepend hostname. Use `state' since these a BOFH can change
677 # these while this process is running and we always want to be
678 # able to release locks taken by this process.
679 state $retval = hostname . '-' . do {
680 my $m; # machine-id(5) is systemd
681 if (open(my $fh, '<', '/etc/machine-id')) { $m = <$fh> }
682 # (g)hostid(1) is in GNU coreutils, kern.hostid is most BSDs
683 chomp($m ||= `{ sysctl -n kern.hostid ||
684 hostid || ghostid; } 2>/dev/null`
685 || "no-machine-id-or-hostid-on-$^O");
692 my $expect = delete($self->{-eidxq_locked}) or return;
693 my ($owner_pid, undef) = split(/-/, $expect);
694 return if $owner_pid != $$; # shards may fork
695 my $oidx = $self->{oidx};
697 my $cur = $oidx->eidx_meta('eidxq_lock') // '';
698 if ($cur eq $expect) {
699 $oidx->eidx_meta('eidxq_lock', '');
701 } elsif ($cur ne '') {
702 warn "E: eidxq_lock($expect) stolen by $cur\n";
704 warn "E: eidxq_lock($expect) released by another process\n";
711 eidxq_release($self) and $self->{oidx}->commit_lazy;
714 sub _eidxq_take ($) {
716 my $val = "$$-${\time}-$>-".host_ident;
717 $self->{oidx}->eidx_meta('eidxq_lock', $val);
718 $self->{-eidxq_locked} = $val;
721 sub eidxq_lock_acquire ($) {
723 my $oidx = $self->{oidx};
725 my $cur = $oidx->eidx_meta('eidxq_lock') || return _eidxq_take($self);
726 if (my $locked = $self->{-eidxq_locked}) { # be lazy
727 return $locked if $locked eq $cur;
729 my ($pid, $time, $euid, $ident) = split(/-/, $cur, 4);
730 my $t = strftime('%Y-%m-%d %k:%M %z', localtime($time));
731 local $self->{current_info} = 'eidxq';
732 if ($euid == $> && $ident eq host_ident) {
735 I: PID:$pid (re)indexing since $t, it will continue our work
739 warn "I: eidxq_lock is stale ($cur), clobbering\n";
740 return _eidxq_take($self);
742 warn "E: kill(0, $pid) failed: $!\n"; # fall-through:
744 my $fn = $oidx->dbh->sqlite_db_filename;
746 W: PID:$pid, UID:$euid on $ident is indexing Xapian since $t
747 W: If this is unexpected, delete `eidxq_lock' from the `eidx_meta' table:
748 W: sqlite3 $fn 'DELETE FROM eidx_meta WHERE key = "eidxq_lock"'
753 sub ibx_sorted ($$) {
754 my ($self, $type) = @_;
755 $self->{"-ibx_ary_$type"} //= do {
756 # highest boost first, stable for config-ordering tiebreaker
759 ($b->{boost} // 0) <=> ($a->{boost} // 0)
760 } @{$self->{'ibx_'.$type} // die "BUG: $type unknown"} ];
764 sub prep_id2pos ($) {
768 $id2pos{$_->{-ibx_id}} = $pos++ for (@{ibx_sorted($self, 'known')});
772 sub eidxq_process ($$) { # for reindexing
773 my ($self, $sync) = @_;
774 local $self->{current_info} = 'eidxq process';
775 return unless ($self->{cfg} && eidxq_lock_acquire($self));
776 my $dbh = $self->{oidx}->dbh;
777 my $tot = $dbh->selectrow_array('SELECT COUNT(*) FROM eidxq') or return;
779 local $sync->{-regen_fmt} = "%u/$tot\n";
780 my $pr = $sync->{-opt}->{-progress};
782 my $min = $dbh->selectrow_array('SELECT MIN(docid) FROM eidxq');
783 my $max = $dbh->selectrow_array('SELECT MAX(docid) FROM eidxq');
784 $pr->("Xapian indexing $min..$max (total=$tot)\n");
786 $sync->{id2pos} //= prep_id2pos($self);
789 $del = $dbh->prepare('DELETE FROM eidxq WHERE docid = ?');
790 $iter = $dbh->prepare('SELECT docid FROM eidxq ORDER BY docid ASC');
792 while (defined(my $docid = $iter->fetchrow_array)) {
793 last if $sync->{quit};
794 if (my $smsg = $self->{oidx}->get_art($docid)) {
795 _reindex_smsg($self, $sync, $smsg);
797 warn "E: #$docid does not exist in over\n";
799 $del->execute($docid);
802 if (checkpoint_due($sync)) {
803 $dbh = $del = $iter = undef;
804 reindex_checkpoint($self, $sync); # release lock
805 $dbh = $self->{oidx}->dbh;
809 $self->git->async_wait_all;
810 $pr->("reindexed ${$sync->{nr}}/$tot\n") if $pr;
813 sub _reindex_unseen { # git->cat_async callback
814 my ($bref, $oid, $type, $size, $req) = @_;
815 return if is_bad_blob($oid, $type, $size, $req->{oid});
816 my $self = $req->{self} // die 'BUG: {self} unset';
817 local $self->{current_info} = "$self->{current_info} $oid";
818 my $new_smsg = bless { blob => $oid, }, 'PublicInbox::Smsg';
819 $new_smsg->set_bytes($$bref, $size);
820 my $eml = $req->{eml} = PublicInbox::Eml->new($bref);
821 $req->{new_smsg} = $new_smsg;
822 $req->{chash} = content_hash($eml);
823 $req->{mids} = mids($eml); # do_step iterates through this
824 do_step($req); # enter the normal indexing flow
827 # --reindex may catch totally unseen messages, this handles them
828 sub reindex_unseen ($$$$) {
829 my ($self, $sync, $ibx, $xsmsg) = @_;
832 autime => $xsmsg->{ds},
833 cotime => $xsmsg->{ts},
834 oid => $xsmsg->{blob},
836 xnum => $xsmsg->{num},
837 # {mids} and {chash} will be filled in at _reindex_unseen
839 warn "I: reindex_unseen ${\$ibx->eidx_key}:$req->{xnum}:$req->{oid}\n";
840 $self->git->cat_async($xsmsg->{blob}, \&_reindex_unseen, $req);
843 sub _unref_stale_range ($$$) {
844 my ($sync, $ibx, $lt_or_gt) = @_;
848 $r = $sync->{self}->{oidx}->dbh->selectall_arrayref(
849 <<EOS, undef, $ibx->{-ibx_id});
850 SELECT docid,xnum,oidbin FROM xref3
851 WHERE ibx_id = ? AND $lt_or_gt LIMIT $lim
853 return if $sync->{quit};
854 for (@$r) { # hopefully rare, not worth optimizing:
855 my ($docid, $xnum, $oidbin) = @$_;
856 my $hex = unpack('H*', $oidbin);
857 warn("# $xnum:$hex (#$docid): stale\n");
858 _unref_doc($sync, $docid, $ibx, $xnum, $oidbin);
860 } while (scalar(@$r) == $lim);
864 sub _reindex_check_ibx ($$$) {
865 my ($self, $sync, $ibx) = @_;
866 my $ibx_id = $ibx->{-ibx_id};
868 my $opt = { limit => $slice };
869 my ($beg, $end) = (1, $slice);
870 my $ekey = $ibx->eidx_key;
873 $max0 = $ibx->mm->num_highwater;
874 sync_inbox($self, $sync, $ibx) and return; # warned
875 $max = $ibx->mm->num_highwater;
876 return if $sync->{quit};
877 } while ($max > $max0 &&
878 warn("# $ekey moved $max0..$max, resyncing..\n"));
879 $end = $max if $end > $max;
881 # first, check if we missed any messages in target $ibx
883 my $pr = $sync->{-opt}->{-progress};
884 local $sync->{-regen_fmt} = "$ekey checking %u/$max\n";
886 my $fast = $sync->{-opt}->{fast};
887 my $usr; # _unref_stale_range (< $lo) called
889 while (scalar(@{$msgs = $ibx->over->query_xover($beg, $end, $opt)})) {
890 ${$sync->{nr}} = $beg;
891 $beg = $msgs->[-1]->{num} + 1;
892 $end = $beg + $slice;
893 $end = $max if $end > $max;
894 if (checkpoint_due($sync)) {
895 reindex_checkpoint($self, $sync); # release lock
897 ($lo, $hi) = ($msgs->[0]->{num}, $msgs->[-1]->{num});
898 $usr //= _unref_stale_range($sync, $ibx, "xnum < $lo");
899 my $x3a = $self->{oidx}->dbh->selectall_arrayref(
900 <<"", undef, $ibx_id, $lo, $hi);
901 SELECT xnum,oidbin,docid FROM xref3 WHERE
902 ibx_id = ? AND xnum >= ? AND xnum <= ?
906 my $k = pack('J', $_->[0]) . $_->[1];
907 push @{$x3m{$k}}, $_->[2];
910 for my $xsmsg (@$msgs) {
911 my $k = pack('JH*', $xsmsg->{num}, $xsmsg->{blob});
912 my $docids = delete($x3m{$k});
913 if (!defined($docids)) {
914 reindex_unseen($self, $sync, $ibx, $xsmsg);
916 for my $num (@$docids) {
917 $self->{oidx}->eidxq_add($num);
920 return if $sync->{quit};
922 next unless scalar keys %x3m;
923 $self->git->async_wait_all; # wait for reindex_unseen
925 # eliminate stale/mismatched entries
926 my %mismatch = map { $_->{num} => $_->{blob} } @$msgs;
927 while (my ($k, $docids) = each %x3m) {
928 my ($xnum, $hex) = unpack('JH*', $k);
929 my $bin = pack('H*', $hex);
930 my $exp = $mismatch{$xnum};
932 my $smsg = $ibx->over->get_art($xnum) // next;
933 # $xnum may be expired by another process
934 if ($smsg->{blob} eq $hex) {
936 BUG: (non-fatal) $ekey #$xnum $smsg->{blob} still matches (old exp: $exp)
939 } # else: continue to unref
941 my $m = defined($exp) ? "mismatch (!= $exp)" : 'stale';
942 warn("# $xnum:$hex (#@$docids): $m\n");
943 for my $i (@$docids) {
944 _unref_doc($sync, $i, $ibx, $xnum, $bin);
946 return if $sync->{quit};
949 defined($hi) and ($hi < $max) and
950 _unref_stale_range($sync, $ibx, "xnum > $hi AND xnum <= $max");
953 sub _reindex_inbox ($$$) {
954 my ($self, $sync, $ibx) = @_;
955 my $ekey = $ibx->eidx_key;
956 local $self->{current_info} = $ekey;
957 if (defined(my $err = _ibx_index_reject($ibx))) {
958 warn "W: cannot reindex $ekey ($err)\n";
960 _reindex_check_ibx($self, $sync, $ibx);
962 delete @$ibx{qw(over mm search git)}; # won't need these for a bit
966 my ($self, $sync) = @_;
967 return unless $self->{cfg};
969 # acquire eidxq_lock early because full reindex takes forever
970 # and incremental -extindex processes can run during our checkpoints
971 if (!eidxq_lock_acquire($self)) {
972 warn "E: aborting --reindex\n";
975 for my $ibx (@{ibx_sorted($self, 'active')}) {
976 _reindex_inbox($self, $sync, $ibx);
977 last if $sync->{quit};
979 $self->git->async_wait_all; # ensure eidxq gets filled completely
980 eidxq_process($self, $sync) unless $sync->{quit};
984 my ($self, $sync, $ibx) = @_;
985 my $err = _sync_inbox($self, $sync, $ibx);
986 delete @$ibx{qw(mm over)};
987 warn $err, "\n" if defined($err);
991 sub dd_smsg { # git->cat_async callback
992 my ($bref, $oid, $type, $size, $dd) = @_;
993 my $smsg = $dd->{smsg} // die 'BUG: dd->{smsg} missing';
994 my $self = $dd->{self} // die 'BUG: {self} missing';
995 my $per_mid = $dd->{per_mid} // die 'BUG: {per_mid} missing';
996 if ($type eq 'missing') {
997 _blob_missing($dd, $smsg);
998 } elsif (!is_bad_blob($oid, $type, $size, $smsg->{blob})) {
999 local $self->{current_info} = "$self->{current_info} $oid";
1000 my $chash = content_hash(PublicInbox::Eml->new($bref));
1001 push(@{$per_mid->{dd_chash}->{$chash}}, $smsg);
1003 return if $per_mid->{last_smsg} != $smsg;
1004 while (my ($chash, $ary) = each %{$per_mid->{dd_chash}}) {
1005 my $keep = shift @$ary;
1006 next if !scalar(@$ary);
1007 $per_mid->{sync}->{dedupe_cull} += scalar(@$ary);
1009 "# <$keep->{mid}> keeping #$keep->{num}, dropping ",
1010 join(', ', map { "#$_->{num}" } @$ary),"\n";
1011 next if $per_mid->{sync}->{-opt}->{'dry-run'};
1012 my $oidx = $self->{oidx};
1013 for my $smsg (@$ary) {
1014 my $gone = $smsg->{num};
1015 $oidx->merge_xref3($keep->{num}, $gone, $smsg->oidbin);
1016 remove_doc($self, $gone);
1021 sub eidx_dedupe ($$$) {
1022 my ($self, $sync, $msgids) = @_;
1023 $sync->{dedupe_cull} = 0;
1026 return unless eidxq_lock_acquire($self);
1027 my ($iter, $cur_mid);
1030 my ($max_id) = $self->{oidx}->dbh->selectrow_array(<<EOS);
1031 SELECT MAX(id) FROM msgid
1033 local $sync->{-regen_fmt} = "dedupe %u/$max_id\n";
1035 # note: we could write this query more intelligently,
1036 # but that causes lock contention with read-only processes
1038 $cur_mid = $msgids->[$idx];
1039 if ($cur_mid eq '') { # all Message-IDs
1040 $iter = $self->{oidx}->dbh->prepare(<<EOS);
1041 SELECT mid,id FROM msgid WHERE id > ? ORDER BY id ASC
1043 $iter->execute($min_id);
1045 $iter = $self->{oidx}->dbh->prepare(<<EOS);
1046 SELECT mid,id FROM msgid WHERE mid = ? AND id > ? ORDER BY id ASC
1048 $iter->execute($cur_mid, $min_id);
1050 while (my ($mid, $id) = $iter->fetchrow_array) {
1051 last if $sync->{quit};
1052 $self->{current_info} = "dedupe $mid";
1053 ${$sync->{nr}} = $min_id = $id;
1055 while (my $x = $self->{oidx}->next_by_mid($mid, \$id, \$prv)) {
1058 next if scalar(@smsg) < 2;
1060 dd_chash => {}, # chash => [ary of smsgs]
1061 last_smsg => $smsg[-1],
1065 $candidates += scalar(@smsg) - 1;
1066 for my $smsg (@smsg) {
1068 per_mid => $per_mid,
1072 $self->git->cat_async($smsg->{blob}, \&dd_smsg, $dd);
1074 # need to wait on every single one @smsg contents can get
1075 # invalidated inside dd_smsg for messages with multiple
1077 $self->git->async_wait_all;
1079 if (checkpoint_due($sync)) {
1081 reindex_checkpoint($self, $sync);
1082 goto dedupe_restart;
1085 goto dedupe_restart if defined($msgids->[++$idx]);
1087 my $n = delete $sync->{dedupe_cull};
1088 if (my $pr = $sync->{-opt}->{-progress}) {
1089 $pr->("culled $n/$candidates candidates ($nr_mid msgids)\n");
1094 sub eidx_sync { # main entry point
1095 my ($self, $opt) = @_;
1097 my $warn_cb = $SIG{__WARN__} || \&CORE::warn;
1098 local $self->{current_info} = '';
1099 local $SIG{__WARN__} = sub {
1100 return if PublicInbox::Eml::warn_ignore(@_);
1101 $warn_cb->($self->{current_info}, ': ', @_);
1103 $self->idx_init($opt); # acquire lock via V2Writable::_idx_init
1104 $self->{oidx}->rethread_prepare($opt);
1106 need_checkpoint => \(my $need_checkpoint = 0),
1108 next_check => now() + 10,
1110 # DO NOT SET {reindex} here, it's incompatible with reused
1111 # V2Writable code, reindex is totally different here
1112 # compared to v1/v2 inboxes because we have multiple histories
1114 -regen_fmt => "%u/?\n",
1116 local $SIG{USR1} = sub { $need_checkpoint = 1 };
1117 my $quit = PublicInbox::SearchIdx::quit_cb($sync);
1118 local $SIG{QUIT} = $quit;
1119 local $SIG{INT} = $quit;
1120 local $SIG{TERM} = $quit;
1121 for my $ibx (@{ibx_sorted($self, 'known')}) {
1122 $ibx->{-ibx_id} //= $self->{oidx}->ibx_id($ibx->eidx_key);
1125 if (scalar(grep { defined($_->{boost}) } @{$self->{ibx_known}})) {
1126 $sync->{id2pos} //= prep_id2pos($self);
1127 $sync->{boost_in_use} = 1;
1130 if (my $msgids = delete($opt->{dedupe})) {
1131 local $sync->{checkpoint_unlocks} = 1;
1132 eidx_dedupe($self, $sync, $msgids);
1134 if (delete($opt->{reindex})) {
1135 local $sync->{checkpoint_unlocks} = 1;
1136 eidx_reindex($self, $sync);
1139 # don't use $_ here, it'll get clobbered by reindex_checkpoint
1140 if ($opt->{scan} // 1) {
1141 for my $ibx (@{ibx_sorted($self, 'active')}) {
1142 last if $sync->{quit};
1143 sync_inbox($self, $sync, $ibx);
1146 $self->{oidx}->rethread_done($opt) unless $sync->{quit};
1147 eidxq_process($self, $sync) unless $sync->{quit};
1149 eidxq_release($self);
1151 $sync; # for eidx_watch
1154 sub update_last_commit { # overrides V2Writable
1155 my ($self, $sync, $stk) = @_;
1156 my $unit = $sync->{unit} // return;
1157 my $latest_cmt = $stk ? $stk->{latest_cmt} : ${$sync->{latest_cmt}};
1158 defined($latest_cmt) or return;
1159 my $ibx = $sync->{ibx} or die 'BUG: {ibx} missing';
1160 my $ekey = $ibx->eidx_key;
1161 my $uv = $ibx->uidvalidity;
1162 my $epoch = $unit->{epoch};
1164 my $v = $ibx->version;
1166 die 'No {epoch} for v2 unit' unless defined $epoch;
1167 $meta_key = "lc-v2:$ekey//$uv;$epoch";
1169 die 'Unexpected {epoch} for v1 unit' if defined $epoch;
1170 $meta_key = "lc-v1:$ekey//$uv";
1172 die "Unsupported inbox version: $v";
1174 my $last = $self->{oidx}->eidx_meta($meta_key);
1175 if (defined $last && is_ancestor($self->git, $last, $latest_cmt)) {
1176 my @cmd = (qw(rev-list --count), "$last..$latest_cmt");
1177 chomp(my $n = $unit->{git}->qx(@cmd));
1178 return if $n ne '' && $n == 0;
1180 $self->{oidx}->eidx_meta($meta_key, $latest_cmt);
1183 sub _idx_init { # with_umask callback
1184 my ($self, $opt) = @_;
1185 PublicInbox::V2Writable::_idx_init($self, $opt); # acquires ei.lock
1186 $self->{midx} = PublicInbox::MiscIdx->new($self);
1189 sub symlink_packs ($$) {
1190 my ($ibx, $pd) = @_;
1192 my $glob = "$ibx->{inboxdir}/git/*.git/objects/pack/*.idx";
1193 for my $idx (bsd_glob($glob, GLOB_NOSORT)) {
1194 my $src = substr($idx, 0, -length('.idx'));
1195 my $dst = $pd . substr($src, rindex($src, '/'));
1196 if (-f "$src.pack" and
1197 symlink("$src.pack", "$dst.pack") and
1198 symlink($idx, "$dst.idx") and
1201 # .promisor, .bitmap, .rev and .keep are optional
1202 # XXX should we symlink .keep here?
1203 for my $s (qw(promisor bitmap rev)) {
1204 symlink("$src.$s", "$dst.$s") if -f "$src.$s";
1206 } elsif (!$!{EEXIST}) {
1207 warn "W: ln -s $src.{pack,idx} => $dst.*: $!\n";
1208 unlink "$dst.pack", "$dst.idx";
1214 sub idx_init { # similar to V2Writable
1215 my ($self, $opt) = @_;
1216 return if $self->{idx_shards};
1218 $self->git->cleanup;
1220 my $ALL = $self->git->{git_dir}; # topdir/ALL.git
1221 my ($has_new, $alt, $seen, $prune, $prune_nr);
1222 if ($opt->{-private}) { # LeiStore
1223 my $local = "$self->{topdir}/local"; # lei/store
1224 $self->{mg} //= PublicInbox::MultiGit->new($self->{topdir},
1225 'ALL.git', 'local');
1228 umask 077; # don't bother restoring for lei
1229 PublicInbox::Import::init_bare($ALL);
1230 $self->git->qx(qw(config core.sharedRepository 0600));
1232 ($alt, $seen) = $self->{mg}->read_alternates(\$mode);
1233 $has_new = $self->{mg}->merge_epochs($alt, $seen);
1234 } else { # extindex has no epochs
1235 $self->{mg} //= PublicInbox::MultiGit->new($self->{topdir},
1237 $prune = $opt->{-idx_gc} ? \$prune_nr : undef;
1238 ($alt, $seen) = $self->{mg}->read_alternates(\$mode, $prune);
1239 PublicInbox::Import::init_bare($ALL);
1242 # git-multi-pack-index(1) can speed up "git cat-file" startup slightly
1244 my $pd = "$ALL/objects/pack";
1245 if (opendir(my $dh, $pd)) { # drop stale symlinks
1246 while (defined(my $dn = readdir($dh))) {
1247 if ($dn =~ /\.(?:idx|pack|promisor|bitmap|rev)\z/) {
1249 unlink($f) if -l $f && !-e $f;
1252 } elsif ($!{ENOENT}) {
1253 mkdir($pd) or die "mkdir($pd): $!";
1255 die "opendir($pd): $!";
1258 for my $ibx (@{ibx_sorted($self, 'active')}) {
1259 # create symlinks for multi-pack-index
1260 $git_midx += symlink_packs($ibx, $pd);
1261 # add new lines to our alternates file
1262 my $d = $ibx->git->{git_dir} . '/objects';
1263 next if exists $alt->{$d};
1264 if (my @st = stat($d)) {
1265 next if $seen->{"$st[0]\0$st[1]"}++;
1267 warn "W: stat($d) failed (from $ibx->{inboxdir}): $!\n";
1268 next if $opt->{-idx_gc};
1272 ($has_new || $prune_nr || $new ne '') and
1273 $self->{mg}->write_alternates($mode, $alt, $new);
1274 $git_midx and $self->with_umask(sub {
1275 my @cmd = ('multi-pack-index');
1276 push @cmd, '--no-progress' if ($opt->{quiet}//0) > 1;
1277 my $lk = $self->lock_for_scope;
1278 system('git', "--git-dir=$ALL", @cmd, 'write');
1279 # ignore errors, fairly new command, may not exist
1281 $self->parallel_init($self->{indexlevel});
1282 $self->with_umask(\&_idx_init, $self, $opt);
1283 $self->{oidx}->begin_lazy;
1284 $self->{oidx}->eidx_prep;
1285 $self->{midx}->create_xdb if $new ne '';
1288 sub _watch_commit { # PublicInbox::DS::add_timer callback
1290 delete $self->{-commit_timer};
1291 eidxq_process($self, $self->{-watch_sync});
1292 eidxq_release($self);
1293 my $fmt = delete $self->{-watch_sync}->{-regen_fmt};
1294 reindex_checkpoint($self, $self->{-watch_sync});
1295 $self->{-watch_sync}->{-regen_fmt} = $fmt;
1297 # call event_step => done unless commit_timer is armed
1298 PublicInbox::DS::requeue($self);
1301 sub on_inbox_unlock { # called by PublicInbox::InboxIdle
1302 my ($self, $ibx) = @_;
1303 my $opt = $self->{-watch_sync}->{-opt};
1304 my $pr = $opt->{-progress};
1305 my $ekey = $ibx->eidx_key;
1306 local $0 = "sync $ekey";
1307 $pr->("indexing $ekey\n") if $pr;
1308 $self->idx_init($opt);
1309 sync_inbox($self, $self->{-watch_sync}, $ibx);
1310 $self->{-commit_timer} //= add_timer($opt->{'commit-interval'} // 10,
1311 \&_watch_commit, $self);
1314 sub eidx_reload { # -extindex --watch SIGHUP handler
1315 my ($self, $idler) = @_;
1317 my $pr = $self->{-watch_sync}->{-opt}->{-progress};
1318 $pr->('reloading ...') if $pr;
1319 delete $self->{-resync_queue};
1320 delete $self->{-ibx_ary_known};
1321 delete $self->{-ibx_ary_active};
1322 $self->{ibx_known} = [];
1323 $self->{ibx_active} = [];
1324 %{$self->{ibx_map}} = ();
1325 delete $self->{-watch_sync}->{id2pos};
1326 my $cfg = PublicInbox::Config->new;
1327 attach_config($self, $cfg);
1328 $idler->refresh($cfg);
1329 $pr->(" done\n") if $pr;
1331 warn "reload not supported without --all\n";
1335 sub eidx_resync_start ($) { # -extindex --watch SIGUSR1 handler
1337 $self->{-resync_queue} //= [ @{ibx_sorted($self, 'active')} ];
1338 PublicInbox::DS::requeue($self); # trigger our ->event_step
1341 sub event_step { # PublicInbox::DS::requeue callback
1343 if (my $resync_queue = $self->{-resync_queue}) {
1344 if (my $ibx = shift(@$resync_queue)) {
1345 on_inbox_unlock($self, $ibx);
1346 PublicInbox::DS::requeue($self);
1348 delete $self->{-resync_queue};
1349 _watch_commit($self);
1352 done($self) unless $self->{-commit_timer};
1356 sub eidx_watch { # public-inbox-extindex --watch main loop
1357 my ($self, $opt) = @_;
1358 local @SIG{keys %SIG} = values %SIG;
1359 for my $sig (qw(HUP USR1 TSTP QUIT INT TERM)) {
1360 $SIG{$sig} = sub { warn "SIG$sig ignored while scanning\n" };
1362 require PublicInbox::InboxIdle;
1363 require PublicInbox::DS;
1364 require PublicInbox::Syscall;
1365 require PublicInbox::Sigfd;
1366 my $idler = PublicInbox::InboxIdle->new($self->{cfg});
1367 if (!$self->{cfg}) {
1368 $idler->watch_inbox($_) for (@{ibx_sorted($self, 'active')});
1370 for my $ibx (@{ibx_sorted($self, 'active')}) {
1371 $ibx->subscribe_unlock(__PACKAGE__, $self)
1373 my $pr = $opt->{-progress};
1374 $pr->("performing initial scan ...\n") if $pr;
1375 my $sync = eidx_sync($self, $opt); # initial sync
1376 return if $sync->{quit};
1377 my $oldset = PublicInbox::DS::block_signals();
1378 local $self->{current_info} = '';
1379 my $cb = $SIG{__WARN__} || \&CORE::warn;
1380 local $SIG{__WARN__} = sub {
1381 return if PublicInbox::Eml::warn_ignore(@_);
1382 $cb->($self->{current_info}, ': ', @_);
1385 HUP => sub { eidx_reload($self, $idler) },
1386 USR1 => sub { eidx_resync_start($self) },
1387 TSTP => sub { kill('STOP', $$) },
1389 my $quit = PublicInbox::SearchIdx::quit_cb($sync);
1390 $sig->{QUIT} = $sig->{INT} = $sig->{TERM} = $quit;
1391 local $self->{-watch_sync} = $sync; # for ->on_inbox_unlock
1392 PublicInbox::DS->SetPostLoopCallback(sub { !$sync->{quit} });
1393 $pr->("initial scan complete, entering event loop\n") if $pr;
1394 # calls InboxIdle->event_step:
1395 PublicInbox::DS::event_loop($sig, $oldset);
1400 *done = \&PublicInbox::V2Writable::done;
1401 *with_umask = \&PublicInbox::InboxWritable::with_umask;
1402 *parallel_init = \&PublicInbox::V2Writable::parallel_init;
1403 *nproc_shards = \&PublicInbox::V2Writable::nproc_shards;
1404 *sync_prepare = \&PublicInbox::V2Writable::sync_prepare;
1405 *index_todo = \&PublicInbox::V2Writable::index_todo;
1406 *count_shards = \&PublicInbox::V2Writable::count_shards;
1407 *atfork_child = \&PublicInbox::V2Writable::atfork_child;
1408 *idx_shard = \&PublicInbox::V2Writable::idx_shard;
1409 *reindex_checkpoint = \&PublicInbox::V2Writable::reindex_checkpoint;
1410 *checkpoint = \&PublicInbox::V2Writable::checkpoint;