X-Git-Url: http://www.git.stargrave.org/?a=blobdiff_plain;f=lib%2FPublicInbox%2FExtSearchIdx.pm;h=c358162889760cde1f4ecf6089918b325ba58329;hb=af0b0fb7a454470a32c452119d0392e0dedb3fe1;hp=9d576adb844a770855ab69ecf9dd966224ea7908;hpb=7ce1a0d97294c5603cf8c6cf89cab491cc5840d3;p=public-inbox.git diff --git a/lib/PublicInbox/ExtSearchIdx.pm b/lib/PublicInbox/ExtSearchIdx.pm index 9d576adb..c3581628 100644 --- a/lib/PublicInbox/ExtSearchIdx.pm +++ b/lib/PublicInbox/ExtSearchIdx.pm @@ -1,4 +1,4 @@ -# Copyright (C) 2020 all contributors +# Copyright (C) 2020-2021 all contributors # License: AGPL-3.0+ # Detached/external index cross inbox search indexing support @@ -18,15 +18,20 @@ use strict; use v5.10.1; use parent qw(PublicInbox::ExtSearch PublicInbox::Lock); use Carp qw(croak carp); +use Sys::Hostname qw(hostname); +use POSIX qw(strftime); use PublicInbox::Search; -use PublicInbox::SearchIdx qw(crlf_adjust prepare_stack is_ancestor); +use PublicInbox::SearchIdx qw(crlf_adjust prepare_stack is_ancestor + is_bad_blob); use PublicInbox::OverIdx; +use PublicInbox::MiscIdx; use PublicInbox::MID qw(mids); use PublicInbox::V2Writable; use PublicInbox::InboxWritable; use PublicInbox::ContentHash qw(content_hash); use PublicInbox::Eml; -use File::Spec; +use PublicInbox::DS qw(now); +use DBI qw(:sql_types); # SQL_BLOB sub new { my (undef, $dir, $opt) = @_; @@ -49,30 +54,14 @@ sub new { }, __PACKAGE__; $self->{shards} = $self->count_shards || nproc_shards($opt->{creat}); my $oidx = PublicInbox::OverIdx->new("$self->{xpfx}/over.sqlite3"); - $oidx->{-no_fsync} = 1 if $opt->{-no_fsync}; + $self->{-no_fsync} = $oidx->{-no_fsync} = 1 if !$opt->{fsync}; $self->{oidx} = $oidx; $self } sub attach_inbox { my ($self, $ibx) = @_; - $ibx = PublicInbox::InboxWritable->new($ibx); - my $key = $ibx->eidx_key; - if (!$ibx->over || !$ibx->mm) { - warn "W: skipping $key (unindexed)\n"; - return; - } - if (!defined($ibx->uidvalidity)) { - warn "W: skipping $key (no UIDVALIDITY)\n"; - return; - } - my $ibxdir = File::Spec->canonpath($ibx->{inboxdir}); - if ($ibxdir ne $ibx->{inboxdir}) { - warn "W: `$ibx->{inboxdir}' canonicalized to `$ibxdir'\n"; - $ibx->{inboxdir} = $ibxdir; - } - $ibx = PublicInbox::InboxWritable->new($ibx); - $self->{ibx_map}->{$key} //= do { + $self->{ibx_map}->{$ibx->eidx_key} //= do { push @{$self->{ibx_list}}, $ibx; $ibx; } @@ -85,27 +74,10 @@ sub _ibx_attach { # each_inbox callback sub attach_config { my ($self, $cfg) = @_; + $self->{cfg} = $cfg; $cfg->each_inbox(\&_ibx_attach, $self); } -sub git_blob_digest ($) { - my ($bref) = @_; - my $dig = Digest::SHA->new(1); # XXX SHA256 later - $dig->add('blob '.length($$bref)."\0"); - $dig->add($$bref); - $dig; -} - -sub is_bad_blob ($$$$) { - my ($oid, $type, $size, $expect_oid) = @_; - if ($type ne 'blob') { - carp "W: $expect_oid is not a blob (type=$type)"; - return 1; - } - croak "BUG: $oid != $expect_oid" if $oid ne $expect_oid; - $size == 0 ? 1 : 0; # size == 0 means purged -} - sub check_batch_limit ($) { my ($req) = @_; my $self = $req->{self}; @@ -126,14 +98,23 @@ sub do_xpost ($$) { my $oid = $req->{oid}; my $xibx = $req->{ibx}; my $eml = $req->{eml}; + my $eidx_key = $xibx->eidx_key; if (my $new_smsg = $req->{new_smsg}) { # 'm' on cross-posted message my $xnum = $req->{xnum}; - $self->{oidx}->add_xref3($docid, $xnum, $oid, $xibx->eidx_key); - $idx->shard_add_eidx_info($docid, $oid, $xibx, $eml); + $self->{oidx}->add_xref3($docid, $xnum, $oid, $eidx_key); + $idx->shard_add_eidx_info($docid, $eidx_key, $eml); check_batch_limit($req); } else { # 'd' - $self->{oidx}->remove_xref3($docid, $oid, $xibx->eidx_key); - $idx->shard_remove_eidx_info($docid, $oid, $xibx, $eml); + my $rm_eidx_info; + my $nr = $self->{oidx}->remove_xref3($docid, $oid, $eidx_key, + \$rm_eidx_info); + if ($nr == 0) { + $self->{oidx}->eidxq_del($docid); + $idx->shard_remove($docid); + } elsif ($rm_eidx_info) { + $idx->shard_remove_eidx_info($docid, $eidx_key, $eml); + $self->{oidx}->eidxq_add($docid); # yes, add + } } } @@ -153,7 +134,7 @@ sub index_unseen ($) { my $oid = $new_smsg->{blob}; my $ibx = delete $req->{ibx} or die 'BUG: {ibx} unset'; $self->{oidx}->add_xref3($docid, $req->{xnum}, $oid, $ibx->eidx_key); - $idx->index_raw(undef, $eml, $new_smsg, $ibx); + $idx->index_raw(undef, $eml, $new_smsg, $ibx->eidx_key); check_batch_limit($req); } @@ -164,13 +145,18 @@ sub do_finalize ($) { } elsif (exists $req->{new_smsg}) { # totally unseen messsage index_unseen($req); } else { - warn "W: ignoring delete $req->{oid} (not found)\n"; + # `d' message was already unindexed in the v1/v2 inboxes, + # so it's too noisy to warn, here. + } + # cur_cmt may be undef for unindex_oid, set by V2Writable::index_todo + if (defined(my $cur_cmt = $req->{cur_cmt})) { + ${$req->{latest_cmt}} = $cur_cmt; } } sub do_step ($) { # main iterator for adding messages to the index my ($req) = @_; - my $self = $req->{self}; + my $self = $req->{self} // die 'BUG: {self} missing'; while (1) { if (my $next_arg = $req->{next_arg}) { if (my $smsg = $self->{oidx}->next_by_mid(@$next_arg)) { @@ -215,6 +201,8 @@ sub ck_existing { # git->cat_async callback if ($type eq 'missing') { _blob_missing($req); } elsif (!is_bad_blob($oid, $type, $size, $smsg->{blob})) { + my $self = $req->{self} // die 'BUG: {self} missing'; + local $self->{current_info} = "$self->{current_info} $oid"; my $cur = PublicInbox::Eml->new($bref); if (content_hash($cur) eq $req->{chash}) { push @{$req->{indexed}}, $smsg; # for do_xpost @@ -229,10 +217,6 @@ sub cur_ibx_xnum ($$) { my ($req, $bref) = @_; my $ibx = $req->{ibx} or die 'BUG: current {ibx} missing'; - # XXX overkill? - git_blob_digest($bref)->hexdigest eq $req->{oid} or die - "BUG: blob mismatch $req->{oid}"; - $req->{eml} = PublicInbox::Eml->new($bref); $req->{chash} = content_hash($req->{eml}); $req->{mids} = mids($req->{eml}); @@ -248,17 +232,22 @@ sub cur_ibx_xnum ($$) { sub index_oid { # git->cat_async callback for 'm' my ($bref, $oid, $type, $size, $req) = @_; + my $self = $req->{self}; + local $self->{current_info} = "$self->{current_info} $oid"; return if is_bad_blob($oid, $type, $size, $req->{oid}); my $new_smsg = $req->{new_smsg} = bless { blob => $oid, }, 'PublicInbox::Smsg'; $new_smsg->{bytes} = $size + crlf_adjust($$bref); defined($req->{xnum} = cur_ibx_xnum($req, $bref)) or return; + ++${$req->{nr}}; do_step($req); } sub unindex_oid { # git->cat_async callback for 'd' my ($bref, $oid, $type, $size, $req) = @_; + my $self = $req->{self}; + local $self->{current_info} = "$self->{current_info} $oid"; return if is_bad_blob($oid, $type, $size, $req->{oid}); return if defined(cur_ibx_xnum($req, $bref)); # was re-added do_step($req); @@ -276,61 +265,641 @@ sub last_commits { $heads; } +sub _ibx_index_reject ($) { + my ($ibx) = @_; + $ibx->mm // return 'unindexed, no msgmap.sqlite3'; + $ibx->uidvalidity // return 'no UIDVALIDITY'; + $ibx->over // return 'unindexed, no over.sqlite3'; + undef; +} + sub _sync_inbox ($$$) { - my ($self, $opt, $ibx) = @_; - my $sync = { - need_checkpoint => \(my $bool = 0), - reindex => $opt->{reindex}, - -opt => $opt, - self => $self, - ibx => $ibx, - }; - my $v = $ibx->version; + my ($self, $sync, $ibx) = @_; my $ekey = $ibx->eidx_key; + if (defined(my $err = _ibx_index_reject($ibx))) { + return "W: skipping $ekey ($err)"; + } + $sync->{ibx} = $ibx; + $sync->{nr} = \(my $nr = 0); + my $v = $ibx->version; if ($v == 2) { - my $epoch_max; - defined($ibx->git_dir_latest(\$epoch_max)) or return; - $sync->{epoch_max} = $epoch_max; - sync_prepare($self, $sync) or return; # fills $sync->{todo} + $sync->{epoch_max} = $ibx->max_git_epoch // return; + sync_prepare($self, $sync); # or return # TODO: once MiscIdx is stable } elsif ($v == 1) { my $uv = $ibx->uidvalidity; my $lc = $self->{oidx}->eidx_meta("lc-v1:$ekey//$uv"); - my $stk = prepare_stack($sync, $lc ? "$lc..HEAD" : 'HEAD'); + my $head = $ibx->mm->last_commit // + return "E: $ibx->{inboxdir} is not indexed"; + my $stk = prepare_stack($sync, $lc ? "$lc..$head" : $head); my $unit = { stack => $stk, git => $ibx->git }; push @{$sync->{todo}}, $unit; } else { - warn "E: $ekey unsupported inbox version (v$v)\n"; + return "E: $ekey unsupported inbox version (v$v)"; + } + for my $unit (@{delete($sync->{todo}) // []}) { + last if $sync->{quit}; + index_todo($self, $sync, $unit); + } + $self->{midx}->index_ibx($ibx) unless $sync->{quit}; + $ibx->git->cleanup; # done with this inbox, now + undef; +} + +sub gc_unref_doc ($$$$) { + my ($self, $ibx_id, $eidx_key, $docid) = @_; + my $dbh = $self->{oidx}->dbh; + + # for debug/info purposes, oids may no longer be accessible + my $sth = $dbh->prepare_cached(<<'', undef, 1); +SELECT oidbin FROM xref3 WHERE docid = ? AND ibx_id = ? + + $sth->execute($docid, $ibx_id); + my @oid = map { unpack('H*', $_->[0]) } @{$sth->fetchall_arrayref}; + + $dbh->prepare_cached(<<'')->execute($docid, $ibx_id); +DELETE FROM xref3 WHERE docid = ? AND ibx_id = ? + + my $remain = $self->{oidx}->get_xref3($docid); + if (scalar(@$remain)) { + $self->{oidx}->eidxq_add($docid); # enqueue for reindex + for my $oid (@oid) { + warn "I: unref #$docid $eidx_key $oid\n"; + } + } else { + warn "I: remove #$docid $eidx_key @oid\n"; + $self->idx_shard($docid)->shard_remove($docid); + } +} + +sub eidx_gc { + my ($self, $opt) = @_; + $self->{cfg} or die "E: GC requires ->attach_config\n"; + $opt->{-idx_gc} = 1; + $self->idx_init($opt); # acquire lock via V2Writable::_idx_init + + my $dbh = $self->{oidx}->dbh; + my $x3_doc = $dbh->prepare('SELECT docid FROM xref3 WHERE ibx_id = ?'); + my $ibx_ck = $dbh->prepare('SELECT ibx_id,eidx_key FROM inboxes'); + my $lc_i = $dbh->prepare('SELECT key FROM eidx_meta WHERE key LIKE ?'); + + $ibx_ck->execute; + while (my ($ibx_id, $eidx_key) = $ibx_ck->fetchrow_array) { + next if $self->{ibx_map}->{$eidx_key}; + $self->{midx}->remove_eidx_key($eidx_key); + warn "I: deleting messages for $eidx_key...\n"; + $x3_doc->execute($ibx_id); + while (defined(my $docid = $x3_doc->fetchrow_array)) { + gc_unref_doc($self, $ibx_id, $eidx_key, $docid); + } + $dbh->prepare_cached(<<'')->execute($ibx_id); +DELETE FROM inboxes WHERE ibx_id = ? + + # drop last_commit info + my $pat = $eidx_key; + $pat =~ s/([_%])/\\$1/g; + $lc_i->execute("lc-%:$pat//%"); + while (my ($key) = $lc_i->fetchrow_array) { + next if $key !~ m!\Alc-v[1-9]+:\Q$eidx_key\E//!; + warn "I: removing $key\n"; + $dbh->prepare_cached(<<'')->execute($key); +DELETE FROM eidx_meta WHERE key = ? + + } + + warn "I: $eidx_key removed\n"; + } + + # it's not real unless it's in `over', we use parallelism here, + # shards will be reading directly from over, so commit + $self->{oidx}->commit_lazy; + $self->{oidx}->begin_lazy; + + for my $idx (@{$self->{idx_shards}}) { + warn "I: cleaning up shard #$idx->{shard}\n"; + $idx->shard_over_check($self->{oidx}); + } + my $nr = $dbh->do(<<''); +DELETE FROM xref3 WHERE docid NOT IN (SELECT num FROM over) + + warn "I: eliminated $nr stale xref3 entries\n" if $nr != 0; + + done($self); +} + +sub _ibx_for ($$$) { + my ($self, $sync, $smsg) = @_; + my $ibx_id = delete($smsg->{ibx_id}) // die '{ibx_id} unset'; + my $pos = $sync->{id2pos}->{$ibx_id} // die "$ibx_id no pos"; + $self->{ibx_list}->[$pos] // die "BUG: ibx for $smsg->{blob} not mapped" +} + +sub _fd_constrained ($) { + my ($self) = @_; + $self->{-fd_constrained} //= do { + my $soft; + if (eval { require BSD::Resource; 1 }) { + my $NOFILE = BSD::Resource::RLIMIT_NOFILE(); + ($soft, undef) = BSD::Resource::getrlimit($NOFILE); + } else { + chomp($soft = `sh -c 'ulimit -n'`); + } + if (defined($soft)) { + my $want = scalar(@{$self->{ibx_list}}) + 64; # estimate + my $ret = $want > $soft; + if ($ret) { + warn <{sync}; + my $self = $sync->{self}; + my $by_chash = delete $req->{by_chash} or die 'BUG: no {by_chash}'; + my $nr = scalar(keys(%$by_chash)) or die 'BUG: no content hashes'; + my $orig_smsg = $req->{orig_smsg} // die 'BUG: no {orig_smsg}'; + my $docid = $smsg->{num} = $orig_smsg->{num}; + $self->{oidx}->add_overview($eml, $smsg); # may rethread + check_batch_limit({ %$sync, new_smsg => $smsg }); + my $chash0 = $smsg->{chash} // die "BUG: $smsg->{blob} no {chash}"; + my $stable = delete($by_chash->{$chash0}) // + die "BUG: $smsg->{blob} chash missing"; + my $idx = $self->idx_shard($docid); + my $top_smsg = pop @$stable; + $top_smsg == $smsg or die 'BUG: top_smsg != smsg'; + my $ibx = _ibx_for($self, $sync, $smsg); + $idx->index_raw(undef, $eml, $smsg, $ibx->eidx_key); + for my $x (reverse @$stable) { + $ibx = _ibx_for($self, $sync, $x); + my $hdr = delete $x->{hdr} // die 'BUG: no {hdr}'; + $idx->shard_add_eidx_info($docid, $ibx->eidx_key, $hdr); + } + return if $nr == 1; # likely, all good + + warn "W: #$docid split into $nr due to deduplication change\n"; + my @todo; + for my $ary (values %$by_chash) { + for my $x (reverse @$ary) { + warn "removing #$docid xref3 $x->{blob}\n"; + my $n = $self->{oidx}->remove_xref3($docid, $x->{blob}); + die "BUG: $x->{blob} invalidated #$docid" if $n == 0; + } + my $x = pop(@$ary) // die "BUG: #$docid {by_chash} empty"; + $x->{num} = delete($x->{xnum}) // die '{xnum} unset'; + $ibx = _ibx_for($self, $sync, $x); + if (my $over = $ibx->over) { + my $e = $over->get_art($x->{num}); + $e->{blob} eq $x->{blob} or die <{blob} != $e->{blob} (${\$ibx->eidx_key}:$e->{num}); +EOF + push @todo, $ibx, $e; + $over->dbh_close if _fd_constrained($self); + } else { + die "$ibx->{inboxdir}: over.sqlite3 unusable: $!\n"; + } + } + undef $by_chash; + while (my ($ibx, $e) = splice(@todo, 0, 2)) { + reindex_unseen($self, $sync, $ibx, $e); + } +} + +sub _reindex_oid { # git->cat_async callback + my ($bref, $oid, $type, $size, $req) = @_; + my $sync = $req->{sync}; + my $self = $sync->{self}; + my $orig_smsg = $req->{orig_smsg} // die 'BUG: no {orig_smsg}'; + my $expect_oid = $req->{xr3r}->[$req->{ix}]->[2]; + my $docid = $orig_smsg->{num}; + if (is_bad_blob($oid, $type, $size, $expect_oid)) { + my $remain = $self->{oidx}->remove_xref3($docid, $expect_oid); + if ($remain == 0) { + warn "W: #$docid gone or corrupted\n"; + $self->idx_shard($docid)->shard_remove($docid); + } elsif (my $next_oid = $req->{xr3r}->[++$req->{ix}]->[2]) { + $self->git->cat_async($next_oid, \&_reindex_oid, $req); + } else { + warn "BUG: #$docid gone (UNEXPECTED)\n"; + $self->idx_shard($docid)->shard_remove($docid); + } + return; + } + my $ci = $self->{current_info}; + local $self->{current_info} = "$ci #$docid $oid"; + my $re_smsg = bless { blob => $oid }, 'PublicInbox::Smsg'; + $re_smsg->{bytes} = $size + crlf_adjust($$bref); + my $eml = PublicInbox::Eml->new($bref); + $re_smsg->populate($eml, { autime => $orig_smsg->{ds}, + cotime => $orig_smsg->{ts} }); + my $chash = content_hash($eml); + $re_smsg->{chash} = $chash; + $re_smsg->{xnum} = $req->{xr3r}->[$req->{ix}]->[1]; + $re_smsg->{ibx_id} = $req->{xr3r}->[$req->{ix}]->[0]; + $re_smsg->{hdr} = $eml->header_obj; + push @{$req->{by_chash}->{$chash}}, $re_smsg; + if (my $next_oid = $req->{xr3r}->[++$req->{ix}]->[2]) { + $self->git->cat_async($next_oid, \&_reindex_oid, $req); + } else { # last $re_smsg is the highest priority xref3 + local $self->{current_info} = "$ci #$docid"; + _reindex_finalize($req, $re_smsg, $eml); + } +} + +sub _reindex_smsg ($$$) { + my ($self, $sync, $smsg) = @_; + my $docid = $smsg->{num}; + my $xr3 = $self->{oidx}->get_xref3($docid, 1); + if (scalar(@$xr3) == 0) { # _reindex_check_stale should've covered this + warn <<""; +BUG? #$docid $smsg->{blob} is not referenced by inboxes during reindex + + $self->{oidx}->delete_by_num($docid); + $self->idx_shard($docid)->shard_remove($docid); + return; + } + + # we sort {xr3r} in the reverse order of {ibx_list} so we can + # hit the common case in _reindex_finalize without rereading + # from git (or holding multiple messages in memory). + my $id2pos = $sync->{id2pos}; # index in {ibx_list} + @$xr3 = sort { + $id2pos->{$b->[0]} <=> $id2pos->{$a->[0]} + || + $b->[1] <=> $a->[1] # break ties with {xnum} + } @$xr3; + @$xr3 = map { [ $_->[0], $_->[1], unpack('H*', $_->[2]) ] } @$xr3; + my $req = { orig_smsg => $smsg, sync => $sync, xr3r => $xr3, ix => 0 }; + $self->git->cat_async($xr3->[$req->{ix}]->[2], \&_reindex_oid, $req); +} + +sub checkpoint_due ($) { + my ($sync) = @_; + ${$sync->{need_checkpoint}} || (now() > $sync->{next_check}); +} + +sub host_ident () { + # I've copied FS images and only changed the hostname before, + # so prepend hostname. Use `state' since these a BOFH can change + # these while this process is running and we always want to be + # able to release locks taken by this process. + state $retval = hostname . '-' . do { + my $m; # machine-id(5) is systemd + if (open(my $fh, '<', '/etc/machine-id')) { $m = <$fh> } + # (g)hostid(1) is in GNU coreutils, kern.hostid is most BSDs + chomp($m ||= `{ sysctl -n kern.hostid || + hostid || ghostid; } 2>/dev/null` + || "no-machine-id-or-hostid-on-$^O"); + $m; + }; +} + +sub eidxq_release { + my ($self) = @_; + my $expect = delete($self->{-eidxq_locked}) or return; + my ($owner_pid, undef) = split(/-/, $expect); + return if $owner_pid != $$; # shards may fork + my $oidx = $self->{oidx}; + $oidx->begin_lazy; + my $cur = $oidx->eidx_meta('eidxq_lock') // ''; + if ($cur eq $expect) { + $oidx->eidx_meta('eidxq_lock', ''); + return 1; + } elsif ($cur ne '') { + warn "E: eidxq_lock($expect) stolen by $cur\n"; + } else { + warn "E: eidxq_lock($expect) released by another process\n"; + } + undef; +} + +sub DESTROY { + my ($self) = @_; + eidxq_release($self) and $self->{oidx}->commit_lazy; +} + +sub _eidxq_take ($) { + my ($self) = @_; + my $val = "$$-${\time}-$>-".host_ident; + $self->{oidx}->eidx_meta('eidxq_lock', $val); + $self->{-eidxq_locked} = $val; +} + +sub eidxq_lock_acquire ($) { + my ($self) = @_; + my $oidx = $self->{oidx}; + $oidx->begin_lazy; + my $cur = $oidx->eidx_meta('eidxq_lock') || return _eidxq_take($self); + if (my $locked = $self->{-eidxq_locked}) { # be lazy + return $locked if $locked eq $cur; + } + my ($pid, $time, $euid, $ident) = split(/-/, $cur, 4); + my $t = strftime('%Y-%m-%d %k:%M:%S', gmtime($time)); + if ($euid == $> && $ident eq host_ident) { + if (kill(0, $pid)) { + warn <dbh->sqlite_db_filename; + warn <{oidx}->dbh; + my $tot = $dbh->selectrow_array('SELECT COUNT(*) FROM eidxq') or return; + ${$sync->{nr}} = 0; + local $sync->{-regen_fmt} = "%u/$tot\n"; + my $pr = $sync->{-opt}->{-progress}; + if ($pr) { + my $min = $dbh->selectrow_array('SELECT MIN(docid) FROM eidxq'); + my $max = $dbh->selectrow_array('SELECT MAX(docid) FROM eidxq'); + $pr->("Xapian indexing $min..$max (total=$tot)\n"); + } + $sync->{id2pos} //= do { + my %id2pos; + my $pos = 0; + $id2pos{$_->{-ibx_id}} = $pos++ for @{$self->{ibx_list}}; + \%id2pos; + }; + my ($del, $iter); +restart: + $del = $dbh->prepare('DELETE FROM eidxq WHERE docid = ?'); + $iter = $dbh->prepare('SELECT docid FROM eidxq ORDER BY docid ASC'); + $iter->execute; + while (defined(my $docid = $iter->fetchrow_array)) { + last if $sync->{quit}; + if (my $smsg = $self->{oidx}->get_art($docid)) { + _reindex_smsg($self, $sync, $smsg); + } else { + warn "E: #$docid does not exist in over\n"; + } + $del->execute($docid); + ++${$sync->{nr}}; + + if (checkpoint_due($sync)) { + $dbh = $del = $iter = undef; + reindex_checkpoint($self, $sync); # release lock + $dbh = $self->{oidx}->dbh; + goto restart; + } + } + $self->git->async_wait_all; + $pr->("reindexed ${$sync->{nr}}/$tot\n") if $pr; +} + +sub _reindex_unseen { # git->cat_async callback + my ($bref, $oid, $type, $size, $req) = @_; + return if is_bad_blob($oid, $type, $size, $req->{oid}); + my $self = $req->{self} // die 'BUG: {self} unset'; + local $self->{current_info} = "$self->{current_info} $oid"; + my $new_smsg = bless { blob => $oid, }, 'PublicInbox::Smsg'; + $new_smsg->{bytes} = $size + crlf_adjust($$bref); + my $eml = $req->{eml} = PublicInbox::Eml->new($bref); + $req->{new_smsg} = $new_smsg; + $req->{chash} = content_hash($eml); + $req->{mids} = mids($eml); # do_step iterates through this + do_step($req); # enter the normal indexing flow +} + +# --reindex may catch totally unseen messages, this handles them +sub reindex_unseen ($$$$) { + my ($self, $sync, $ibx, $xsmsg) = @_; + my $req = { + %$sync, # has {self} + autime => $xsmsg->{ds}, + cotime => $xsmsg->{ts}, + oid => $xsmsg->{blob}, + ibx => $ibx, + xnum => $xsmsg->{num}, + # {mids} and {chash} will be filled in at _reindex_unseen + }; + warn "I: reindex_unseen ${\$ibx->eidx_key}:$req->{xnum}:$req->{oid}\n"; + $self->git->cat_async($xsmsg->{blob}, \&_reindex_unseen, $req); +} + +sub _reindex_check_unseen ($$$) { + my ($self, $sync, $ibx) = @_; + my $ibx_id = $ibx->{-ibx_id}; + my $slice = 1000; + my ($beg, $end) = (1, $slice); + + # first, check if we missed any messages in target $ibx + my $msgs; + my $pr = $sync->{-opt}->{-progress}; + my $ekey = $ibx->eidx_key; + local $sync->{-regen_fmt} = + "$ekey checking unseen %u/".$ibx->over->max."\n"; + ${$sync->{nr}} = 0; + + while (scalar(@{$msgs = $ibx->over->query_xover($beg, $end)})) { + ${$sync->{nr}} = $beg; + $beg = $msgs->[-1]->{num} + 1; + $end = $beg + $slice; + if (checkpoint_due($sync)) { + reindex_checkpoint($self, $sync); # release lock + } + + my $inx3 = $self->{oidx}->dbh->prepare_cached(<<'', undef, 1); +SELECT DISTINCT(docid) FROM xref3 WHERE +ibx_id = ? AND xnum = ? AND oidbin = ? + + for my $xsmsg (@$msgs) { + my $oidbin = pack('H*', $xsmsg->{blob}); + $inx3->bind_param(1, $ibx_id); + $inx3->bind_param(2, $xsmsg->{num}); + $inx3->bind_param(3, $oidbin, SQL_BLOB); + $inx3->execute; + my $docids = $inx3->fetchall_arrayref; + # index messages which were totally missed + # the first time around ASAP: + if (scalar(@$docids) == 0) { + reindex_unseen($self, $sync, $ibx, $xsmsg); + } else { # already seen, reindex later + for my $r (@$docids) { + $self->{oidx}->eidxq_add($r->[0]); + } + } + last if $sync->{quit}; + } + last if $sync->{quit}; + } +} + +sub _reindex_check_stale ($$$) { + my ($self, $sync, $ibx) = @_; + my $min = 0; + my $pr = $sync->{-opt}->{-progress}; + my $fetching; + my $ekey = $ibx->eidx_key; + local $sync->{-regen_fmt} = + "$ekey check stale/missing %u/".$ibx->over->max."\n"; + ${$sync->{nr}} = 0; + do { + if (checkpoint_due($sync)) { + reindex_checkpoint($self, $sync); # release lock + } + # now, check if there's stale xrefs + my $iter = $self->{oidx}->dbh->prepare_cached(<<'', undef, 1); +SELECT docid,xnum,oidbin FROM xref3 WHERE ibx_id = ? AND docid > ? +ORDER BY docid,xnum ASC LIMIT 10000 + + $iter->execute($ibx->{-ibx_id}, $min); + $fetching = undef; + + while (my ($docid, $xnum, $oidbin) = $iter->fetchrow_array) { + return if $sync->{quit}; + ${$sync->{nr}} = $xnum; + + $fetching = $min = $docid; + my $smsg = $ibx->over->get_art($xnum); + my $oidhex = unpack('H*', $oidbin); + my $err; + if (!$smsg) { + $err = 'stale'; + } elsif ($smsg->{blob} ne $oidhex) { + $err = "mismatch (!= $smsg->{blob})"; + } else { + next; # likely, all good + } + # current_info already has eidx_key + warn "$xnum:$oidhex (#$docid): $err\n"; + my $del = $self->{oidx}->dbh->prepare_cached(<<''); +DELETE FROM xref3 WHERE ibx_id = ? AND xnum = ? AND oidbin = ? + + $del->bind_param(1, $ibx->{-ibx_id}); + $del->bind_param(2, $xnum); + $del->bind_param(3, $oidbin, SQL_BLOB); + $del->execute; + + # get_xref3 over-fetches, but this is a rare path: + my $xr3 = $self->{oidx}->get_xref3($docid); + my $idx = $self->idx_shard($docid); + if (scalar(@$xr3) == 0) { # all gone + $self->{oidx}->delete_by_num($docid); + $self->{oidx}->eidxq_del($docid); + $idx->shard_remove($docid); + } else { # enqueue for reindex of remaining messages + $idx->shard_remove_eidx_info($docid, + $ibx->eidx_key); + $self->{oidx}->eidxq_add($docid); # yes, add + } + } + } while (defined $fetching); +} + +sub _reindex_inbox ($$$) { + my ($self, $sync, $ibx) = @_; + my $ekey = $ibx->eidx_key; + local $self->{current_info} = $ekey; + if (defined(my $err = _ibx_index_reject($ibx))) { + warn "W: cannot reindex $ekey ($err)\n"; + } else { + _reindex_check_unseen($self, $sync, $ibx); + _reindex_check_stale($self, $sync, $ibx) unless $sync->{quit}; + } + delete @$ibx{qw(over mm search git)}; # won't need these for a bit +} + +sub eidx_reindex { + my ($self, $sync) = @_; + + # acquire eidxq_lock early because full reindex takes forever + # and incremental -extindex processes can run during our checkpoints + if (!eidxq_lock_acquire($self)) { + warn "E: aborting --reindex\n"; return; } - index_todo($self, $sync, $_) for @{$sync->{todo}}; + for my $ibx (@{$self->{ibx_list}}) { + _reindex_inbox($self, $sync, $ibx); + last if $sync->{quit}; + } + $self->git->async_wait_all; # ensure eidxq gets filled completely + eidxq_process($self, $sync) unless $sync->{quit}; +} + +sub sync_inbox { + my ($self, $sync, $ibx) = @_; + my $err = _sync_inbox($self, $sync, $ibx); + delete @$ibx{qw(mm over)}; + warn $err, "\n" if defined($err); } sub eidx_sync { # main entry point my ($self, $opt) = @_; - $self->idx_init($opt); # acquire lock via V2Writable::_idx_init - $self->{oidx}->rethread_prepare($opt); - my $warn_cb = $SIG{__WARN__} || sub { print STDERR @_ }; + my $warn_cb = $SIG{__WARN__} || \&CORE::warn; local $self->{current_info} = ''; local $SIG{__WARN__} = sub { $warn_cb->($self->{current_info}, ': ', @_); }; - - # don't use $_ here, it'll get clobbered by reindex_checkpoint + $self->idx_init($opt); # acquire lock via V2Writable::_idx_init + $self->{oidx}->rethread_prepare($opt); + my $sync = { + need_checkpoint => \(my $need_checkpoint = 0), + check_intvl => 10, + next_check => now() + 10, + -opt => $opt, + # DO NOT SET {reindex} here, it's incompatible with reused + # V2Writable code, reindex is totally different here + # compared to v1/v2 inboxes because we have multiple histories + self => $self, + -regen_fmt => "%u/?\n", + }; + local $SIG{USR1} = sub { $need_checkpoint = 1 }; + my $quit = PublicInbox::SearchIdx::quit_cb($sync); + local $SIG{QUIT} = $quit; + local $SIG{INT} = $quit; + local $SIG{TERM} = $quit; for my $ibx (@{$self->{ibx_list}}) { - _sync_inbox($self, $opt, $ibx); + $ibx->{-ibx_id} //= $self->{oidx}->ibx_id($ibx->eidx_key); + } + if (delete($opt->{reindex})) { + local $sync->{checkpoint_unlocks} = 1; + eidx_reindex($self, $sync); } - $self->{oidx}->rethread_done($opt); + # don't use $_ here, it'll get clobbered by reindex_checkpoint + if ($opt->{scan} // 1) { + for my $ibx (@{$self->{ibx_list}}) { + last if $sync->{quit}; + sync_inbox($self, $sync, $ibx); + } + } + $self->{oidx}->rethread_done($opt) unless $sync->{quit}; + eidxq_process($self, $sync) unless $sync->{quit}; - PublicInbox::V2Writable::done($self); + eidxq_release($self); + done($self); + $sync; # for eidx_watch } sub update_last_commit { # overrides V2Writable - my ($self, $sync, $unit, $latest_cmt) = @_; - return unless defined $latest_cmt; - - $self->git->async_wait_all; + my ($self, $sync, $stk) = @_; + my $unit = $sync->{unit} // return; + my $latest_cmt = $stk ? $stk->{latest_cmt} : ${$sync->{latest_cmt}}; + defined($latest_cmt) or return; my $ibx = $sync->{ibx} or die 'BUG: {ibx} missing'; my $ekey = $ibx->eidx_key; my $uv = $ibx->uidvalidity; @@ -347,7 +916,7 @@ sub update_last_commit { # overrides V2Writable die "Unsupported inbox version: $v"; } my $last = $self->{oidx}->eidx_meta($meta_key); - if (defined $last && is_ancestor($unit->{git}, $last, $latest_cmt)) { + if (defined $last && is_ancestor($self->git, $last, $latest_cmt)) { my @cmd = (qw(rev-list --count), "$last..$latest_cmt"); chomp(my $n = $unit->{git}->qx(@cmd)); return if $n ne '' && $n == 0; @@ -355,43 +924,203 @@ sub update_last_commit { # overrides V2Writable $self->{oidx}->eidx_meta($meta_key, $latest_cmt); } +sub _idx_init { # with_umask callback + my ($self, $opt) = @_; + PublicInbox::V2Writable::_idx_init($self, $opt); + $self->{midx} = PublicInbox::MiscIdx->new($self); +} + sub idx_init { # similar to V2Writable my ($self, $opt) = @_; return if $self->{idx_shards}; $self->git->cleanup; - + my $mode = 0644; my $ALL = $self->git->{git_dir}; # ALL.git - PublicInbox::Import::init_bare($ALL) unless -d $ALL; + my $old = -d $ALL; + if ($opt->{-private}) { # LeiStore + $mode = 0600; + if (!$old) { + umask 077; # don't bother restoring + PublicInbox::Import::init_bare($ALL); + $self->git->qx(qw(config core.sharedRepository 0600)); + } + } else { + PublicInbox::Import::init_bare($ALL) unless $old; + } my $info_dir = "$ALL/objects/info"; my $alt = "$info_dir/alternates"; - my $mode = 0644; - my (%old, @old, %new, @new); + my (@old, @new, %seen); # seen: st_dev + st_ino if (-e $alt) { open(my $fh, '<', $alt) or die "open $alt: $!"; $mode = (stat($fh))[2] & 07777; - while (<$fh>) { - push @old, $_ if !$old{$_}++; + while (my $line = <$fh>) { + chomp(my $d = $line); + + # expand relative path (/local/ stuff) + substr($d, 0, 3) eq '../' and + $d = "$ALL/objects/$d"; + if (my @st = stat($d)) { + next if $seen{"$st[0]\0$st[1]"}++; + } else { + warn "W: stat($d) failed (from $alt): $!\n"; + next if $opt->{-idx_gc}; + } + push @old, $line; + } + } + + # for LeiStore, and possibly some mirror-only state + if (opendir(my $dh, my $local = "$self->{topdir}/local")) { + # highest numbered epoch first + for my $n (sort { $b <=> $a } map { substr($_, 0, -4) + 0 } + grep(/\A[0-9]+\.git\z/, readdir($dh))) { + my $d = "$local/$n.git/objects"; # absolute path + if (my @st = stat($d)) { + next if $seen{"$st[0]\0$st[1]"}++; + # favor relative paths for rename-friendliness + push @new, "../../local/$n.git/objects\n"; + } else { + warn "W: stat($d) failed: $!\n"; + } } } for my $ibx (@{$self->{ibx_list}}) { my $line = $ibx->git->{git_dir} . "/objects\n"; - next if $old{$line}; - $new{$line} = 1; + chomp(my $d = $line); + if (my @st = stat($d)) { + next if $seen{"$st[0]\0$st[1]"}++; + } else { + warn "W: stat($d) failed (from $ibx->{inboxdir}): $!\n"; + next if $opt->{-idx_gc}; + } push @new, $line; } - push @old, @new; - PublicInbox::V2Writable::write_alternates($info_dir, $mode, \@old); + if (scalar @new) { + push @old, @new; + my $o = \@old; + PublicInbox::V2Writable::write_alternates($info_dir, $mode, $o); + } $self->parallel_init($self->{indexlevel}); - $self->umask_prepare; - $self->with_umask(\&PublicInbox::V2Writable::_idx_init, $self, $opt); + $self->with_umask(\&_idx_init, $self, $opt); $self->{oidx}->begin_lazy; $self->{oidx}->eidx_prep; + $self->{midx}->begin_txn; +} + +sub _watch_commit { # PublicInbox::DS::add_timer callback + my ($self) = @_; + delete $self->{-commit_timer}; + eidxq_process($self, $self->{-watch_sync}); + eidxq_release($self); + delete local $self->{-watch_sync}->{-regen_fmt}; + reindex_checkpoint($self, $self->{-watch_sync}); + + # call event_step => done unless commit_timer is armed + PublicInbox::DS::requeue($self); +} + +sub on_inbox_unlock { # called by PublicInbox::InboxIdle + my ($self, $ibx) = @_; + my $opt = $self->{-watch_sync}->{-opt}; + my $pr = $opt->{-progress}; + my $ekey = $ibx->eidx_key; + local $0 = "sync $ekey"; + $pr->("indexing $ekey\n") if $pr; + $self->idx_init($opt); + sync_inbox($self, $self->{-watch_sync}, $ibx); + $self->{-commit_timer} //= PublicInbox::DS::add_timer( + $opt->{'commit-interval'} // 10, + \&_watch_commit, $self); +} + +sub eidx_reload { # -extindex --watch SIGHUP handler + my ($self, $idler) = @_; + if ($self->{cfg}) { + my $pr = $self->{-watch_sync}->{-opt}->{-progress}; + $pr->('reloading ...') if $pr; + delete $self->{-resync_queue}; + @{$self->{ibx_list}} = (); + %{$self->{ibx_map}} = (); + delete $self->{-watch_sync}->{id2pos}; + my $cfg = PublicInbox::Config->new; + attach_config($self, $cfg); + $idler->refresh($cfg); + $pr->(" done\n") if $pr; + } else { + warn "reload not supported without --all\n"; + } +} + +sub eidx_resync_start ($) { # -extindex --watch SIGUSR1 handler + my ($self) = @_; + $self->{-resync_queue} //= [ @{$self->{ibx_list}} ]; + PublicInbox::DS::requeue($self); # trigger our ->event_step +} + +sub event_step { # PublicInbox::DS::requeue callback + my ($self) = @_; + if (my $resync_queue = $self->{-resync_queue}) { + if (my $ibx = shift(@$resync_queue)) { + on_inbox_unlock($self, $ibx); + PublicInbox::DS::requeue($self); + } else { + delete $self->{-resync_queue}; + _watch_commit($self); + } + } else { + done($self) unless $self->{-commit_timer}; + } +} + +sub eidx_watch { # public-inbox-extindex --watch main loop + my ($self, $opt) = @_; + local %SIG = %SIG; + for my $sig (qw(HUP USR1 TSTP QUIT INT TERM)) { + $SIG{$sig} = sub { warn "SIG$sig ignored while scanning\n" }; + } + require PublicInbox::InboxIdle; + require PublicInbox::DS; + require PublicInbox::Syscall; + require PublicInbox::Sigfd; + my $idler = PublicInbox::InboxIdle->new($self->{cfg}); + if (!$self->{cfg}) { + $idler->watch_inbox($_) for @{$self->{ibx_list}}; + } + $_->subscribe_unlock(__PACKAGE__, $self) for @{$self->{ibx_list}}; + my $pr = $opt->{-progress}; + $pr->("performing initial scan ...\n") if $pr; + my $sync = eidx_sync($self, $opt); # initial sync + return if $sync->{quit}; + my $oldset = PublicInbox::Sigfd::block_signals(); + local $self->{current_info} = ''; + my $cb = $SIG{__WARN__} || \&CORE::warn; + local $SIG{__WARN__} = sub { $cb->($self->{current_info}, ': ', @_) }; + my $sig = { + HUP => sub { eidx_reload($self, $idler) }, + USR1 => sub { eidx_resync_start($self) }, + TSTP => sub { kill('STOP', $$) }, + }; + my $quit = PublicInbox::SearchIdx::quit_cb($sync); + $sig->{QUIT} = $sig->{INT} = $sig->{TERM} = $quit; + my $sigfd = PublicInbox::Sigfd->new($sig, + $PublicInbox::Syscall::SFD_NONBLOCK); + %SIG = (%SIG, %$sig) if !$sigfd; + local $self->{-watch_sync} = $sync; # for ->on_inbox_unlock + if (!$sigfd) { + # wake up every second to accept signals if we don't + # have signalfd or IO::KQueue: + PublicInbox::Sigfd::sig_setmask($oldset); + PublicInbox::DS->SetLoopTimeout(1000); + } + PublicInbox::DS->SetPostLoopCallback(sub { !$sync->{quit} }); + $pr->("initial scan complete, entering event loop\n") if $pr; + PublicInbox::DS->EventLoop; # calls InboxIdle->event_step + done($self); } no warnings 'once'; *done = \&PublicInbox::V2Writable::done; -*umask_prepare = \&PublicInbox::InboxWritable::umask_prepare; *with_umask = \&PublicInbox::InboxWritable::with_umask; *parallel_init = \&PublicInbox::V2Writable::parallel_init; *nproc_shards = \&PublicInbox::V2Writable::nproc_shards; @@ -400,5 +1129,6 @@ no warnings 'once'; *count_shards = \&PublicInbox::V2Writable::count_shards; *atfork_child = \&PublicInbox::V2Writable::atfork_child; *idx_shard = \&PublicInbox::V2Writable::idx_shard; +*reindex_checkpoint = \&PublicInbox::V2Writable::reindex_checkpoint; 1;