X-Git-Url: http://www.git.stargrave.org/?a=blobdiff_plain;f=lib%2FPublicInbox%2FExtSearchIdx.pm;h=7c44a1a406308604c1319b91eaa7130a0be645d2;hb=23af251dd607c4e75ab1e68063f2c885c48cc035;hp=a14f0652651c4b9a2e7b3a0be139bec4ccebbcc0;hpb=991731f1084b99230d1f2a1f2bb8ce7c69bc252b;p=public-inbox.git diff --git a/lib/PublicInbox/ExtSearchIdx.pm b/lib/PublicInbox/ExtSearchIdx.pm index a14f0652..7c44a1a4 100644 --- a/lib/PublicInbox/ExtSearchIdx.pm +++ b/lib/PublicInbox/ExtSearchIdx.pm @@ -1,4 +1,4 @@ -# Copyright (C) 2020-2021 all contributors +# Copyright (C) all contributors # License: AGPL-3.0+ # Detached/external index cross inbox search indexing support @@ -18,9 +18,11 @@ use strict; use v5.10.1; use parent qw(PublicInbox::ExtSearch PublicInbox::Lock); use Carp qw(croak carp); +use Scalar::Util qw(blessed); use Sys::Hostname qw(hostname); use POSIX qw(strftime); use File::Glob qw(bsd_glob GLOB_NOSORT); +use PublicInbox::MultiGit; use PublicInbox::Search; use PublicInbox::SearchIdx qw(prepare_stack is_ancestor is_bad_blob); use PublicInbox::OverIdx; @@ -44,7 +46,8 @@ sub new { topdir => $dir, creat => $opt->{creat}, ibx_map => {}, # (newsgroup//inboxdir) => $ibx - ibx_list => [], + ibx_active => [], # by config section order + ibx_known => [], # by config section order indexlevel => $l, transact_bytes => 0, total_bytes => 0, @@ -52,30 +55,52 @@ sub new { parallel => 1, lock_path => "$dir/ei.lock", }, __PACKAGE__; - $self->{shards} = $self->count_shards || nproc_shards($opt->{creat}); + $self->{shards} = $self->count_shards || + nproc_shards({ nproc => $opt->{jobs} }); my $oidx = PublicInbox::OverIdx->new("$self->{xpfx}/over.sqlite3"); $self->{-no_fsync} = $oidx->{-no_fsync} = 1 if !$opt->{fsync}; + $self->{-dangerous} = 1 if $opt->{dangerous}; $self->{oidx} = $oidx; $self } sub attach_inbox { - my ($self, $ibx) = @_; + my ($self, $ibx, $types) = @_; $self->{ibx_map}->{$ibx->eidx_key} //= do { - push @{$self->{ibx_list}}, $ibx; + delete $self->{-ibx_ary_known}; # invalidate cache + delete $self->{-ibx_ary_active}; # invalidate cache + $types //= [ qw(active known) ]; + for my $t (@$types) { + push @{$self->{"ibx_$t"}}, $ibx; + } $ibx; } } sub _ibx_attach { # each_inbox callback - my ($ibx, $self) = @_; - attach_inbox($self, $ibx); + my ($ibx, $self, $types) = @_; + attach_inbox($self, $ibx, $types); } sub attach_config { - my ($self, $cfg) = @_; + my ($self, $cfg, $ibxs) = @_; $self->{cfg} = $cfg; - $cfg->each_inbox(\&_ibx_attach, $self); + my $types; + if ($ibxs) { + for my $ibx (@$ibxs) { + $self->{ibx_map}->{$ibx->eidx_key} //= do { + push @{$self->{ibx_active}}, $ibx; + push @{$self->{ibx_known}}, $ibx; + $ibx; + } + } + # invalidate cache + delete $self->{-ibx_ary_known}; + delete $self->{-ibx_ary_active}; + $types = [ 'known' ]; + } + $types //= [ qw(known active) ]; + $cfg->each_inbox(\&_ibx_attach, $self, $types); } sub check_batch_limit ($) { @@ -88,32 +113,99 @@ sub check_batch_limit ($) { ${$req->{need_checkpoint}} = 1 if $n >= $self->{batch_bytes}; } +sub apply_boost ($$) { + my ($req, $smsg) = @_; + my $id2pos = $req->{id2pos}; # index in ibx_sorted + my $xr3 = $req->{self}->{oidx}->get_xref3($smsg->{num}, 1); + @$xr3 = sort { + $id2pos->{$a->[0]} <=> $id2pos->{$b->[0]} + || + $a->[1] <=> $b->[1] # break ties with {xnum} + } @$xr3; + my $new_smsg = $req->{new_smsg}; + return if $xr3->[0]->[2] ne $new_smsg->oidbin; # loser + + # replace the old smsg with the more boosted one + $new_smsg->{num} = $smsg->{num}; + $new_smsg->populate($req->{eml}, $req); + $req->{self}->{oidx}->add_overview($req->{eml}, $new_smsg); +} + +sub remove_doc ($$) { + my ($self, $docid) = @_; + $self->{oidx}->delete_by_num($docid); + $self->{oidx}->eidxq_del($docid); + $self->idx_shard($docid)->ipc_do('xdb_remove', $docid); +} + +sub _unref_doc ($$$$$;$) { + my ($sync, $docid, $ibx, $xnum, $oidbin, $eml) = @_; + my $smsg; + if (ref($docid)) { + $smsg = $docid; + $docid = $smsg->{num}; + } + if (defined($oidbin) && defined($xnum) && blessed($ibx) && $ibx->over) { + my $smsg = $ibx->over->get_art($xnum); + if ($smsg && $smsg->oidbin eq $oidbin) { + carp("BUG: (non-fatal) ".$ibx->eidx_key. + " #$xnum $smsg->{blob} still valid"); + return; + } + } + my $s = 'DELETE FROM xref3 WHERE oidbin = ?'; + $s .= ' AND ibx_id = ?' if defined($ibx); + $s .= ' AND xnum = ?' if defined($xnum); + my $del = $sync->{self}->{oidx}->dbh->prepare_cached($s); + my $col = 0; + $del->bind_param(++$col, $oidbin, SQL_BLOB); + $del->bind_param(++$col, $ibx->{-ibx_id}) if $ibx; + $del->bind_param(++$col, $xnum) if defined($xnum); + $del->execute; + my $xr3 = $sync->{self}->{oidx}->get_xref3($docid); + if (scalar(@$xr3) == 0) { # all gone + remove_doc($sync->{self}, $docid); + } else { # enqueue for reindex of remaining messages + if ($ibx) { + my $ekey = $ibx->{-gc_eidx_key} // $ibx->eidx_key; + my $idx = $sync->{self}->idx_shard($docid); + $idx->ipc_do('remove_eidx_info', $docid, $ekey, $eml); + } # else: we can't remove_eidx_info in reindex-only path + + # replace invalidated blob ASAP with something which should be + # readable since we may commit the transaction on checkpoint. + # eidxq processing will re-apply boost + $smsg //= $sync->{self}->{oidx}->get_art($docid); + my $hex = unpack('H*', $oidbin); + if ($smsg && $smsg->{blob} eq $hex) { + $xr3->[0] =~ /:([a-f0-9]{40,}+)\z/ or + die "BUG: xref $xr3->[0] has no OID"; + $sync->{self}->{oidx}->update_blob($smsg, $1); + } + # yes, add, we'll need to re-apply boost + $sync->{self}->{oidx}->eidxq_add($docid); + } + @$xr3 +} + sub do_xpost ($$) { my ($req, $smsg) = @_; my $self = $req->{self}; my $docid = $smsg->{num}; - my $idx = $self->idx_shard($docid); 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 $eidx_key = $xibx->eidx_key; my $xnum = $req->{xnum}; $self->{oidx}->add_xref3($docid, $xnum, $oid, $eidx_key); + my $idx = $self->idx_shard($docid); $idx->ipc_do('add_eidx_info', $docid, $eidx_key, $eml); - check_batch_limit($req); - } else { # 'd' - 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->ipc_do('xdb_remove', $docid); - } elsif ($rm_eidx_info) { - $idx->ipc_do('remove_eidx_info', - $docid, $eidx_key, $eml); - $self->{oidx}->eidxq_add($docid); # yes, add - } + apply_boost($req, $smsg) if $req->{boost_in_use}; + } else { # 'd' no {xnum} + $self->git->async_wait_all; + $oid = pack('H*', $oid); + _unref_doc($req, $docid, $xibx, undef, $oid, $eml); } } @@ -139,7 +231,7 @@ sub index_unseen ($) { sub do_finalize ($) { my ($req) = @_; - if (my $indexed = $req->{indexed}) { + if (my $indexed = $req->{indexed}) { # duplicated messages do_xpost($req, $_) for @$indexed; } elsif (exists $req->{new_smsg}) { # totally unseen messsage index_unseen($req); @@ -164,11 +256,10 @@ sub do_step ($) { # main iterator for adding messages to the index \&ck_existing, $req); return; # ck_existing calls do_step } - delete $req->{cur_smsg}; delete $req->{next_arg}; } - my $mid = shift(@{$req->{mids}}); - last unless defined $mid; + die "BUG: {cur_smsg} still set" if $req->{cur_smsg}; + my $mid = shift(@{$req->{mids}}) // last; my ($id, $prev); $req->{next_arg} = [ $mid, \$id, \$prev ]; # loop again @@ -176,29 +267,19 @@ sub do_step ($) { # main iterator for adding messages to the index do_finalize($req); } -sub _blob_missing ($) { # called when req->{cur_smsg}->{blob} is bad - my ($req) = @_; - my $smsg = $req->{cur_smsg} or die 'BUG: {cur_smsg} missing'; - my $self = $req->{self}; - my $xref3 = $self->{oidx}->get_xref3($smsg->{num}); - my @keep = grep(!/:$smsg->{blob}\z/, @$xref3); - if (@keep) { - $keep[0] =~ /:([a-f0-9]{40,}+)\z/ or - die "BUG: xref $keep[0] has no OID"; - my $oidhex = $1; - $self->{oidx}->remove_xref3($smsg->{num}, $smsg->{blob}); - my $upd = $self->{oidx}->update_blob($smsg, $oidhex); - my $saved = $self->{oidx}->get_art($smsg->{num}); - } else { - $self->{oidx}->delete_by_num($smsg->{num}); - } +sub _blob_missing ($$) { # called when a known $smsg->{blob} is gone + my ($req, $smsg) = @_; + # xnum and ibx are unknown, we only call this when an entry from + # /ei*/over.sqlite3 is bad, not on entries from xap*/over.sqlite3 + $req->{self}->git->async_wait_all; + _unref_doc($req, $smsg, undef, undef, $smsg->oidbin); } sub ck_existing { # git->cat_async callback my ($bref, $oid, $type, $size, $req) = @_; - my $smsg = $req->{cur_smsg} or die 'BUG: {cur_smsg} missing'; + my $smsg = delete $req->{cur_smsg} or die 'BUG: {cur_smsg} missing'; if ($type eq 'missing') { - _blob_missing($req); + _blob_missing($req, $smsg); } elsif (!is_bad_blob($oid, $type, $size, $smsg->{blob})) { my $self = $req->{self} // die 'BUG: {self} missing'; local $self->{current_info} = "$self->{current_info} $oid"; @@ -212,18 +293,18 @@ sub ck_existing { # git->cat_async callback # is the messages visible in the inbox currently being indexed? # return the number if so -sub cur_ibx_xnum ($$) { - my ($req, $bref) = @_; +sub cur_ibx_xnum ($$;$) { + my ($req, $bref, $mismatch) = @_; my $ibx = $req->{ibx} or die 'BUG: current {ibx} missing'; $req->{eml} = PublicInbox::Eml->new($bref); $req->{chash} = content_hash($req->{eml}); $req->{mids} = mids($req->{eml}); - my @q = @{$req->{mids}}; # copy - while (defined(my $mid = shift @q)) { + for my $mid (@{$req->{mids}}) { my ($id, $prev); while (my $x = $ibx->over->next_by_mid($mid, \$id, \$prev)) { return $x->{num} if $x->{blob} eq $req->{oid}; + push @$mismatch, $x if $mismatch; } } undef; @@ -238,8 +319,15 @@ sub index_oid { # git->cat_async callback for 'm' blob => $oid, }, 'PublicInbox::Smsg'; $new_smsg->set_bytes($$bref, $size); - defined($req->{xnum} = cur_ibx_xnum($req, $bref)) or return; ++${$req->{nr}}; + my $mismatch = []; + $req->{xnum} = cur_ibx_xnum($req, $bref, $mismatch) // do { + warn "# deleted\n"; + warn "# mismatch $_->{blob}\n" for @$mismatch; + ${$req->{latest_cmt}} = $req->{cur_cmt} // + die "BUG: {cur_cmt} unset ($oid)\n"; + return; + }; do_step($req); } @@ -304,86 +392,122 @@ sub _sync_inbox ($$$) { undef; } -sub gc_unref_doc ($$$$) { - my ($self, $ibx_id, $eidx_key, $docid) = @_; - my $dbh = $self->{oidx}->dbh; - - # for debug/info purposes, oids may no longer be accessible - my $sth = $dbh->prepare_cached(<<'', undef, 1); -SELECT oidbin FROM xref3 WHERE docid = ? AND ibx_id = ? - - $sth->execute($docid, $ibx_id); - my @oid = map { unpack('H*', $_->[0]) } @{$sth->fetchall_arrayref}; - - $dbh->prepare_cached(<<'')->execute($docid, $ibx_id); -DELETE FROM xref3 WHERE docid = ? AND ibx_id = ? - - my $remain = $self->{oidx}->get_xref3($docid); - if (scalar(@$remain)) { - $self->{oidx}->eidxq_add($docid); # enqueue for reindex - for my $oid (@oid) { - warn "I: unref #$docid $eidx_key $oid\n"; - } - } else { - warn "I: remove #$docid $eidx_key @oid\n"; - $self->idx_shard($docid)->ipc_do('xdb_remove', $docid); - } -} - -sub eidx_gc { - my ($self, $opt) = @_; - $self->{cfg} or die "E: GC requires ->attach_config\n"; - $opt->{-idx_gc} = 1; - $self->idx_init($opt); # acquire lock via V2Writable::_idx_init - - my $dbh = $self->{oidx}->dbh; - $dbh->do('PRAGMA case_sensitive_like = ON'); # only place we use LIKE - my $x3_doc = $dbh->prepare('SELECT docid FROM xref3 WHERE ibx_id = ?'); - my $ibx_ck = $dbh->prepare('SELECT ibx_id,eidx_key FROM inboxes'); - my $lc_i = $dbh->prepare(<<''); -SELECT key FROM eidx_meta WHERE key LIKE ? ESCAPE ? - +sub eidx_gc_scan_inboxes ($$) { + my ($self, $sync) = @_; + my ($x3_doc, $ibx_ck); +restart: + $x3_doc = $self->{oidx}->dbh->prepare(<{oidx}->dbh->prepare(<execute; while (my ($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); + my $ibx = { -ibx_id => $ibx_id, -gc_eidx_key => $eidx_key }; + while (my ($docid, $xnum, $oid) = $x3_doc->fetchrow_array) { + my $r = _unref_doc($sync, $docid, $ibx, $xnum, $oid); + $oid = unpack('H*', $oid); + $r = $r ? 'unref' : 'remove'; + warn "I: $r #$docid $eidx_key $oid\n"; + if (checkpoint_due($sync)) { + $x3_doc = $ibx_ck = undef; + reindex_checkpoint($self, $sync); + goto restart; + } } - $dbh->prepare_cached(<<'')->execute($ibx_id); + $self->{oidx}->dbh->do(<<'', undef, $ibx_id); DELETE FROM inboxes WHERE ibx_id = ? # drop last_commit info my $pat = $eidx_key; $pat =~ s/([_%\\])/\\$1/g; + $self->{oidx}->dbh->do('PRAGMA case_sensitive_like = ON'); + my $lc_i = $self->{oidx}->dbh->prepare(<<''); +SELECT key FROM eidx_meta WHERE key LIKE ? ESCAPE ? + $lc_i->execute("lc-%:$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); + $self->{oidx}->dbh->do(<<'', undef, $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(<<''); +sub eidx_gc_scan_shards ($$) { # TODO: use for lei/store + my ($self, $sync) = @_; + my $nr = $self->{oidx}->dbh->do(<<''); DELETE FROM xref3 WHERE docid NOT IN (SELECT num FROM over) warn "I: eliminated $nr stale xref3 entries\n" if $nr != 0; + reindex_checkpoint($self, $sync) if checkpoint_due($sync); + + # fixup from old bugs: + $nr = $self->{oidx}->dbh->do(<<''); +DELETE FROM over WHERE num > 0 AND num NOT IN (SELECT docid FROM xref3) + warn "I: eliminated $nr stale over entries\n" if $nr != 0; + reindex_checkpoint($self, $sync) if checkpoint_due($sync); + + $nr = $self->{oidx}->dbh->do(<<''); +DELETE FROM eidxq WHERE docid NOT IN (SELECT num FROM over) + + warn "I: eliminated $nr stale reindex queue entries\n" if $nr != 0; + reindex_checkpoint($self, $sync) if checkpoint_due($sync); + + my ($cur) = $self->{oidx}->dbh->selectrow_array(< 0 +EOM + $cur // return; # empty + my ($r, $n, %active_shards); + $nr = 0; + while (1) { + $r = $self->{oidx}->dbh->selectcol_arrayref(<<"", undef, $cur); +SELECT num FROM over WHERE num >= ? ORDER BY num ASC LIMIT 10000 + + last unless scalar(@$r); + while (defined($n = shift @$r)) { + for my $i ($cur..($n - 1)) { + my $idx = idx_shard($self, $i); + $idx->ipc_do('xdb_remove_quiet', $i); + $active_shards{$idx} = $idx; + } + $cur = $n + 1; + } + if (checkpoint_due($sync)) { + for my $idx (values %active_shards) { + $nr += $idx->ipc_do('nr_quiet_rm') + } + %active_shards = (); + reindex_checkpoint($self, $sync); + } + } + warn "I: eliminated $nr stale Xapian documents\n" if $nr != 0; +} + +sub eidx_gc { + my ($self, $opt) = @_; + $self->{cfg} or die "E: GC requires ->attach_config\n"; + $opt->{-idx_gc} = 1; + my $sync = { + need_checkpoint => \(my $need_checkpoint = 0), + check_intvl => 10, + next_check => now() + 10, + checkpoint_unlocks => 1, + -opt => $opt, + self => $self, + }; + $self->idx_init($opt); # acquire lock via V2Writable::_idx_init + eidx_gc_scan_inboxes($self, $sync); + eidx_gc_scan_shards($self, $sync); done($self); } @@ -391,7 +515,8 @@ sub _ibx_for ($$$) { my ($self, $sync, $smsg) = @_; my $ibx_id = delete($smsg->{ibx_id}) // die '{ibx_id} unset'; my $pos = $sync->{id2pos}->{$ibx_id} // die "$ibx_id no pos"; - $self->{ibx_list}->[$pos] // die "BUG: ibx for $smsg->{blob} not mapped" + $self->{-ibx_ary_known}->[$pos] // + die "BUG: ibx for $smsg->{blob} not mapped" } sub _fd_constrained ($) { @@ -405,7 +530,8 @@ sub _fd_constrained ($) { chomp($soft = `sh -c 'ulimit -n'`); } if (defined($soft)) { - my $want = scalar(@{$self->{ibx_list}}) + 64; # estimate + # $want is an estimate + my $want = scalar(@{$self->{ibx_active}}) + 64; my $ret = $want > $soft; if ($ret) { warn <git->async_wait_all; warn "W: #$docid split into $nr due to deduplication change\n"; my @todo; for my $ary (values %$by_chash) { for my $x (reverse @$ary) { warn "removing #$docid xref3 $x->{blob}\n"; - my $n = $self->{oidx}->remove_xref3($docid, $x->{blob}); + my $bin = $x->oidbin; + my $n = _unref_doc($sync, $docid, undef, undef, $bin); die "BUG: $x->{blob} invalidated #$docid" if $n == 0; } my $x = pop(@$ary) // die "BUG: #$docid {by_chash} empty"; @@ -481,15 +609,14 @@ sub _reindex_oid { # git->cat_async callback my $expect_oid = $req->{xr3r}->[$req->{ix}]->[2]; my $docid = $orig_smsg->{num}; if (is_bad_blob($oid, $type, $size, $expect_oid)) { - my $remain = $self->{oidx}->remove_xref3($docid, $expect_oid); + my $oidbin = pack('H*', $expect_oid); + my $remain = _unref_doc($sync, $docid, undef, undef, $oidbin); if ($remain == 0) { - warn "W: #$docid gone or corrupted\n"; - $self->idx_shard($docid)->ipc_do('xdb_remove', $docid); + warn "W: #$docid ($oid) gone or corrupt\n"; } elsif (my $next_oid = $req->{xr3r}->[++$req->{ix}]->[2]) { $self->git->cat_async($next_oid, \&_reindex_oid, $req); } else { - warn "BUG: #$docid gone (UNEXPECTED)\n"; - $self->idx_shard($docid)->ipc_do('xdb_remove', $docid); + warn "BUG: #$docid ($oid) gone (UNEXPECTED)\n"; } return; } @@ -522,15 +649,14 @@ sub _reindex_smsg ($$$) { warn <<""; BUG? #$docid $smsg->{blob} is not referenced by inboxes during reindex - $self->{oidx}->delete_by_num($docid); - $self->idx_shard($docid)->ipc_do('xdb_remove', $docid); + remove_doc($self, $docid); return; } - # we sort {xr3r} in the reverse order of {ibx_list} so we can + # we sort {xr3r} in the reverse order of ibx_sorted 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} + my $id2pos = $sync->{id2pos}; # index in ibx_sorted @$xr3 = sort { $id2pos->{$b->[0]} <=> $id2pos->{$a->[0]} || @@ -602,11 +728,12 @@ sub eidxq_lock_acquire ($) { 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)); + my $t = strftime('%Y-%m-%d %k:%M %z', localtime($time)); + local $self->{current_info} = 'eidxq'; if ($euid == $> && $ident eq host_ident) { if (kill(0, $pid)) { warn <{"-ibx_ary_$type"} //= do { + # highest boost first, stable for config-ordering tiebreaker + use sort 'stable'; + [ sort { + ($b->{boost} // 0) <=> ($a->{boost} // 0) + } @{$self->{'ibx_'.$type} // die "BUG: $type unknown"} ]; + } +} + +sub prep_id2pos ($) { + my ($self) = @_; + my %id2pos; + my $pos = 0; + $id2pos{$_->{-ibx_id}} = $pos++ for (@{ibx_sorted($self, 'known')}); + \%id2pos; +} + sub eidxq_process ($$) { # for reindexing my ($self, $sync) = @_; - - return unless eidxq_lock_acquire($self); + local $self->{current_info} = 'eidxq process'; + return unless ($self->{cfg} && eidxq_lock_acquire($self)); my $dbh = $self->{oidx}->dbh; my $tot = $dbh->selectrow_array('SELECT COUNT(*) FROM eidxq') or return; ${$sync->{nr}} = 0; @@ -638,12 +784,7 @@ sub eidxq_process ($$) { # for reindexing 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; - }; + $sync->{id2pos} //= prep_id2pos($self); my ($del, $iter); restart: $del = $dbh->prepare('DELETE FROM eidxq WHERE docid = ?'); @@ -700,114 +841,114 @@ sub reindex_unseen ($$$$) { $self->git->cat_async($xsmsg->{blob}, \&_reindex_unseen, $req); } -sub _reindex_check_unseen ($$$) { +sub _unref_stale_range ($$$) { + my ($sync, $ibx, $lt_or_gt) = @_; + my $r; + my $lim = 10000; + do { + $r = $sync->{self}->{oidx}->dbh->selectall_arrayref( + <{-ibx_id}); +SELECT docid,xnum,oidbin FROM xref3 +WHERE ibx_id = ? AND $lt_or_gt LIMIT $lim +EOS + return if $sync->{quit}; + for (@$r) { # hopefully rare, not worth optimizing: + my ($docid, $xnum, $oidbin) = @$_; + my $hex = unpack('H*', $oidbin); + warn("# $xnum:$hex (#$docid): stale\n"); + _unref_doc($sync, $docid, $ibx, $xnum, $oidbin); + } + } while (scalar(@$r) == $lim); + 1; +} + +sub _reindex_check_ibx ($$$) { my ($self, $sync, $ibx) = @_; my $ibx_id = $ibx->{-ibx_id}; - my $slice = 1000; + my $slice = 10000; + my $opt = { limit => $slice }; my ($beg, $end) = (1, $slice); + my $ekey = $ibx->eidx_key; + my ($max, $max0); + do { + $max0 = $ibx->mm->num_highwater; + sync_inbox($self, $sync, $ibx) and return; # warned + $max = $ibx->mm->num_highwater; + return if $sync->{quit}; + } while ($max > $max0 && + warn("# $ekey moved $max0..$max, resyncing..\n")); + $end = $max if $end > $max; # 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"; + local $sync->{-regen_fmt} = "$ekey checking %u/$max\n"; ${$sync->{nr}} = 0; - - while (scalar(@{$msgs = $ibx->over->query_xover($beg, $end)})) { + my $fast = $sync->{-opt}->{fast}; + my $usr; # _unref_stale_range (< $lo) called + my ($lo, $hi); + while (scalar(@{$msgs = $ibx->over->query_xover($beg, $end, $opt)})) { ${$sync->{nr}} = $beg; $beg = $msgs->[-1]->{num} + 1; $end = $beg + $slice; + $end = $max if $end > $max; if (checkpoint_due($sync)) { reindex_checkpoint($self, $sync); # release lock } - - my $inx3 = $self->{oidx}->dbh->prepare_cached(<<'', undef, 1); -SELECT DISTINCT(docid) FROM xref3 WHERE -ibx_id = ? AND xnum = ? AND oidbin = ? - + ($lo, $hi) = ($msgs->[0]->{num}, $msgs->[-1]->{num}); + $usr //= _unref_stale_range($sync, $ibx, "xnum < $lo"); + my $x3a = $self->{oidx}->dbh->selectall_arrayref( + <<"", undef, $ibx_id, $lo, $hi); +SELECT xnum,oidbin,docid FROM xref3 WHERE +ibx_id = ? AND xnum >= ? AND xnum <= ? + + my %x3m; + for (@$x3a) { + my $k = pack('J', $_->[0]) . $_->[1]; + push @{$x3m{$k}}, $_->[2]; + } + undef $x3a; 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) { + my $k = pack('JH*', $xsmsg->{num}, $xsmsg->{blob}); + my $docids = delete($x3m{$k}); + if (!defined($docids)) { reindex_unseen($self, $sync, $ibx, $xsmsg); - } else { # already seen, reindex later - for my $r (@$docids) { - $self->{oidx}->eidxq_add($r->[0]); + } elsif (!$fast) { + for my $num (@$docids) { + $self->{oidx}->eidxq_add($num); } } - 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 + } + next unless scalar keys %x3m; + $self->git->async_wait_all; # wait for reindex_unseen + + # eliminate stale/mismatched entries + my %mismatch = map { $_->{num} => $_->{blob} } @$msgs; + while (my ($k, $docids) = each %x3m) { + my ($xnum, $hex) = unpack('JH*', $k); + my $bin = pack('H*', $hex); + my $exp = $mismatch{$xnum}; + if (defined $exp) { + my $smsg = $ibx->over->get_art($xnum) // next; + # $xnum may be expired by another process + if ($smsg->{blob} eq $hex) { + warn <<""; +BUG: (non-fatal) $ekey #$xnum $smsg->{blob} still matches (old exp: $exp) + + next; + } # else: continue to unref } - # 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->ipc_do('xdb_remove', $docid); - } else { # enqueue for reindex of remaining messages - $idx->ipc_do('remove_eidx_info', - $docid, $ibx->eidx_key); - $self->{oidx}->eidxq_add($docid); # yes, add + my $m = defined($exp) ? "mismatch (!= $exp)" : 'stale'; + warn("# $xnum:$hex (#@$docids): $m\n"); + for my $i (@$docids) { + _unref_doc($sync, $i, $ibx, $xnum, $bin); } + return if $sync->{quit}; } - } while (defined $fetching); + } + defined($hi) and ($hi < $max) and + _unref_stale_range($sync, $ibx, "xnum > $hi AND xnum <= $max"); } sub _reindex_inbox ($$$) { @@ -817,14 +958,14 @@ sub _reindex_inbox ($$$) { 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}; + _reindex_check_ibx($self, $sync, $ibx); } delete @$ibx{qw(over mm search git)}; # won't need these for a bit } sub eidx_reindex { my ($self, $sync) = @_; + return unless $self->{cfg}; # acquire eidxq_lock early because full reindex takes forever # and incremental -extindex processes can run during our checkpoints @@ -832,7 +973,7 @@ sub eidx_reindex { warn "E: aborting --reindex\n"; return; } - for my $ibx (@{$self->{ibx_list}}) { + for my $ibx (@{ibx_sorted($self, 'active')}) { _reindex_inbox($self, $sync, $ibx); last if $sync->{quit}; } @@ -845,6 +986,110 @@ sub sync_inbox { my $err = _sync_inbox($self, $sync, $ibx); delete @$ibx{qw(mm over)}; warn $err, "\n" if defined($err); + $err; +} + +sub dd_smsg { # git->cat_async callback + my ($bref, $oid, $type, $size, $dd) = @_; + my $smsg = $dd->{smsg} // die 'BUG: dd->{smsg} missing'; + my $self = $dd->{self} // die 'BUG: {self} missing'; + my $per_mid = $dd->{per_mid} // die 'BUG: {per_mid} missing'; + if ($type eq 'missing') { + _blob_missing($dd, $smsg); + } elsif (!is_bad_blob($oid, $type, $size, $smsg->{blob})) { + local $self->{current_info} = "$self->{current_info} $oid"; + my $chash = content_hash(PublicInbox::Eml->new($bref)); + push(@{$per_mid->{dd_chash}->{$chash}}, $smsg); + } + return if $per_mid->{last_smsg} != $smsg; + while (my ($chash, $ary) = each %{$per_mid->{dd_chash}}) { + my $keep = shift @$ary; + next if !scalar(@$ary); + $per_mid->{sync}->{dedupe_cull} += scalar(@$ary); + print STDERR + "# <$keep->{mid}> keeping #$keep->{num}, dropping ", + join(', ', map { "#$_->{num}" } @$ary),"\n"; + next if $per_mid->{sync}->{-opt}->{'dry-run'}; + my $oidx = $self->{oidx}; + for my $smsg (@$ary) { + my $gone = $smsg->{num}; + $oidx->merge_xref3($keep->{num}, $gone, $smsg->oidbin); + remove_doc($self, $gone); + } + } +} + +sub eidx_dedupe ($$$) { + my ($self, $sync, $msgids) = @_; + $sync->{dedupe_cull} = 0; + my $candidates = 0; + my $nr_mid = 0; + return unless eidxq_lock_acquire($self); + my ($iter, $cur_mid); + my $min_id = 0; + my $idx = 0; + my ($max_id) = $self->{oidx}->dbh->selectrow_array(<{-regen_fmt} = "dedupe %u/$max_id\n"; + + # note: we could write this query more intelligently, + # but that causes lock contention with read-only processes +dedupe_restart: + $cur_mid = $msgids->[$idx]; + if ($cur_mid eq '') { # all Message-IDs + $iter = $self->{oidx}->dbh->prepare(< ? ORDER BY id ASC +EOS + $iter->execute($min_id); + } else { + $iter = $self->{oidx}->dbh->prepare(< ? ORDER BY id ASC +EOS + $iter->execute($cur_mid, $min_id); + } + while (my ($mid, $id) = $iter->fetchrow_array) { + last if $sync->{quit}; + $self->{current_info} = "dedupe $mid"; + ${$sync->{nr}} = $min_id = $id; + my ($prv, @smsg); + while (my $x = $self->{oidx}->next_by_mid($mid, \$id, \$prv)) { + push @smsg, $x; + } + next if scalar(@smsg) < 2; + my $per_mid = { + dd_chash => {}, # chash => [ary of smsgs] + last_smsg => $smsg[-1], + sync => $sync + }; + $nr_mid++; + $candidates += scalar(@smsg) - 1; + for my $smsg (@smsg) { + my $dd = { + per_mid => $per_mid, + smsg => $smsg, + self => $self, + }; + $self->git->cat_async($smsg->{blob}, \&dd_smsg, $dd); + } + # need to wait on every single one @smsg contents can get + # invalidated inside dd_smsg for messages with multiple + # Message-IDs. + $self->git->async_wait_all; + + if (checkpoint_due($sync)) { + undef $iter; + reindex_checkpoint($self, $sync); + goto dedupe_restart; + } + } + goto dedupe_restart if defined($msgids->[++$idx]); + + my $n = delete $sync->{dedupe_cull}; + if (my $pr = $sync->{-opt}->{-progress}) { + $pr->("culled $n/$candidates candidates ($nr_mid msgids)\n"); + } + ${$sync->{nr}} = 0; } sub eidx_sync { # main entry point @@ -853,6 +1098,7 @@ sub eidx_sync { # main entry point my $warn_cb = $SIG{__WARN__} || \&CORE::warn; local $self->{current_info} = ''; local $SIG{__WARN__} = sub { + return if PublicInbox::Eml::warn_ignore(@_); $warn_cb->($self->{current_info}, ': ', @_); }; $self->idx_init($opt); # acquire lock via V2Writable::_idx_init @@ -873,9 +1119,19 @@ sub eidx_sync { # main entry point local $SIG{QUIT} = $quit; local $SIG{INT} = $quit; local $SIG{TERM} = $quit; - for my $ibx (@{$self->{ibx_list}}) { + for my $ibx (@{ibx_sorted($self, 'known')}) { $ibx->{-ibx_id} //= $self->{oidx}->ibx_id($ibx->eidx_key); } + + if (scalar(grep { defined($_->{boost}) } @{$self->{ibx_known}})) { + $sync->{id2pos} //= prep_id2pos($self); + $sync->{boost_in_use} = 1; + } + + if (my $msgids = delete($opt->{dedupe})) { + local $sync->{checkpoint_unlocks} = 1; + eidx_dedupe($self, $sync, $msgids); + } if (delete($opt->{reindex})) { local $sync->{checkpoint_unlocks} = 1; eidx_reindex($self, $sync); @@ -883,7 +1139,7 @@ sub eidx_sync { # main entry point # don't use $_ here, it'll get clobbered by reindex_checkpoint if ($opt->{scan} // 1) { - for my $ibx (@{$self->{ibx_list}}) { + for my $ibx (@{ibx_sorted($self, 'active')}) { last if $sync->{quit}; sync_inbox($self, $sync, $ibx); } @@ -943,9 +1199,9 @@ sub symlink_packs ($$) { symlink($idx, "$dst.idx") and -f $idx) { ++$ret; - # .promisor and .keep are optional + # .promisor, .bitmap, .rev and .keep are optional # XXX should we symlink .keep here? - for my $s (qw(promisor)) { + for my $s (qw(promisor bitmap rev)) { symlink("$src.$s", "$dst.$s") if -f "$src.$s"; } } elsif (!$!{EEXIST}) { @@ -962,91 +1218,64 @@ sub idx_init { # similar to V2Writable $self->git->cleanup; my $mode = 0644; - my $ALL = $self->git->{git_dir}; # ALL.git - my $old = -d $ALL; + my $ALL = $self->git->{git_dir}; # topdir/ALL.git + my ($has_new, $alt, $seen, $prune, $prune_nr); if ($opt->{-private}) { # LeiStore + my $local = "$self->{topdir}/local"; # lei/store + $self->{mg} //= PublicInbox::MultiGit->new($self->{topdir}, + 'ALL.git', 'local'); $mode = 0600; - if (!$old) { - umask 077; # don't bother restoring + unless (-d $ALL) { + umask 077; # don't bother restoring for lei PublicInbox::Import::init_bare($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 (@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 (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, $line; - } + ($alt, $seen) = $self->{mg}->read_alternates(\$mode); + $has_new = $self->{mg}->merge_epochs($alt, $seen); + } else { # extindex has no epochs + $self->{mg} //= PublicInbox::MultiGit->new($self->{topdir}, + 'ALL.git'); + $prune = $opt->{-idx_gc} ? \$prune_nr : undef; + ($alt, $seen) = $self->{mg}->read_alternates(\$mode, $prune); + PublicInbox::Import::init_bare($ALL); } - # 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"; - } - } - } # git-multi-pack-index(1) can speed up "git cat-file" startup slightly - my $dh; my $git_midx = 0; my $pd = "$ALL/objects/pack"; - if (!mkdir($pd) && $!{EEXIST} && opendir($dh, $pd)) { - # drop stale symlinks + if (opendir(my $dh, $pd)) { # drop stale symlinks while (defined(my $dn = readdir($dh))) { - if ($dn =~ /\.(?:idx|pack|promisor)\z/) { + if ($dn =~ /\.(?:idx|pack|promisor|bitmap|rev)\z/) { my $f = "$pd/$dn"; unlink($f) if -l $f && !-e $f; } } - undef $dh; + } elsif ($!{ENOENT}) { + mkdir($pd) or die "mkdir($pd): $!"; + } else { + die "opendir($pd): $!"; } - for my $ibx (@{$self->{ibx_list}}) { + my $new = ''; + for my $ibx (@{ibx_sorted($self, 'active')}) { # create symlinks for multi-pack-index $git_midx += symlink_packs($ibx, $pd); # add new lines to our alternates file - my $line = $ibx->git->{git_dir} . "/objects\n"; - chomp(my $d = $line); + my $d = $ibx->git->{git_dir} . '/objects'; + next if exists $alt->{$d}; if (my @st = stat($d)) { - next if $seen{"$st[0]\0$st[1]"}++; + 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; - } - if (scalar @new) { - push @old, @new; - my $o = \@old; - PublicInbox::V2Writable::write_alternates($info_dir, $mode, $o); + $new .= "$d\n"; } + ($has_new || $prune_nr || $new ne '') and + $self->{mg}->write_alternates($mode, $alt, $new); $git_midx and $self->with_umask(sub { my @cmd = ('multi-pack-index'); push @cmd, '--no-progress' if ($opt->{quiet}//0) > 1; + my $lk = $self->lock_for_scope; system('git', "--git-dir=$ALL", @cmd, 'write'); # ignore errors, fairly new command, may not exist }); @@ -1054,7 +1283,7 @@ sub idx_init { # similar to V2Writable $self->with_umask(\&_idx_init, $self, $opt); $self->{oidx}->begin_lazy; $self->{oidx}->eidx_prep; - $self->{midx}->create_xdb if @new; + $self->{midx}->create_xdb if $new ne ''; } sub _watch_commit { # PublicInbox::DS::add_timer callback @@ -1089,7 +1318,10 @@ sub eidx_reload { # -extindex --watch SIGHUP handler my $pr = $self->{-watch_sync}->{-opt}->{-progress}; $pr->('reloading ...') if $pr; delete $self->{-resync_queue}; - @{$self->{ibx_list}} = (); + delete $self->{-ibx_ary_known}; + delete $self->{-ibx_ary_active}; + $self->{ibx_known} = []; + $self->{ibx_active} = []; %{$self->{ibx_map}} = (); delete $self->{-watch_sync}->{id2pos}; my $cfg = PublicInbox::Config->new; @@ -1103,7 +1335,7 @@ sub eidx_reload { # -extindex --watch SIGHUP handler sub eidx_resync_start ($) { # -extindex --watch SIGUSR1 handler my ($self) = @_; - $self->{-resync_queue} //= [ @{$self->{ibx_list}} ]; + $self->{-resync_queue} //= [ @{ibx_sorted($self, 'active')} ]; PublicInbox::DS::requeue($self); # trigger our ->event_step } @@ -1124,7 +1356,7 @@ sub event_step { # PublicInbox::DS::requeue callback sub eidx_watch { # public-inbox-extindex --watch main loop my ($self, $opt) = @_; - local %SIG = %SIG; + local @SIG{keys %SIG} = values %SIG; for my $sig (qw(HUP USR1 TSTP QUIT INT TERM)) { $SIG{$sig} = sub { warn "SIG$sig ignored while scanning\n" }; } @@ -1134,9 +1366,11 @@ sub eidx_watch { # public-inbox-extindex --watch main loop require PublicInbox::Sigfd; my $idler = PublicInbox::InboxIdle->new($self->{cfg}); if (!$self->{cfg}) { - $idler->watch_inbox($_) for @{$self->{ibx_list}}; + $idler->watch_inbox($_) for (@{ibx_sorted($self, 'active')}); + } + for my $ibx (@{ibx_sorted($self, 'active')}) { + $ibx->subscribe_unlock(__PACKAGE__, $self) } - $_->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 @@ -1144,7 +1378,10 @@ sub eidx_watch { # public-inbox-extindex --watch main loop my $oldset = PublicInbox::DS::block_signals(); local $self->{current_info} = ''; my $cb = $SIG{__WARN__} || \&CORE::warn; - local $SIG{__WARN__} = sub { $cb->($self->{current_info}, ': ', @_) }; + local $SIG{__WARN__} = sub { + return if PublicInbox::Eml::warn_ignore(@_); + $cb->($self->{current_info}, ': ', @_); + }; my $sig = { HUP => sub { eidx_reload($self, $idler) }, USR1 => sub { eidx_resync_start($self) }, @@ -1152,19 +1389,11 @@ sub eidx_watch { # public-inbox-extindex --watch main loop }; 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::DS::sig_setmask($oldset); - PublicInbox::DS->SetLoopTimeout(1000); - } PublicInbox::DS->SetPostLoopCallback(sub { !$sync->{quit} }); $pr->("initial scan complete, entering event loop\n") if $pr; - PublicInbox::DS->EventLoop; # calls InboxIdle->event_step + # calls InboxIdle->event_step: + PublicInbox::DS::event_loop($sig, $oldset); done($self); }