X-Git-Url: http://www.git.stargrave.org/?a=blobdiff_plain;f=lib%2FPublicInbox%2FExtSearchIdx.pm;h=a2d70205c9cdb2a7faac5b977ecd3246de9974e2;hb=0c8106d44f317175e122744b43407bf067183175;hp=f29a84e3d1d3fa6db23779e739cc5da2ddb2cd26;hpb=35cca56ecb10649f7935d601ef513162e38cb3b8;p=public-inbox.git diff --git a/lib/PublicInbox/ExtSearchIdx.pm b/lib/PublicInbox/ExtSearchIdx.pm index f29a84e3..a2d70205 100644 --- a/lib/PublicInbox/ExtSearchIdx.pm +++ b/lib/PublicInbox/ExtSearchIdx.pm @@ -18,6 +18,8 @@ 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 is_bad_blob); @@ -28,12 +30,11 @@ 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) = @_; - $dir = File::Spec->canonpath($dir); my $l = $opt->{indexlevel} // 'full'; $l !~ $PublicInbox::SearchIdx::INDEXLEVELS and die "invalid indexlevel=$l\n"; @@ -53,30 +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; $ibx; } @@ -149,7 +134,7 @@ sub index_unseen ($) { 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); } @@ -280,31 +265,36 @@ 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, $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; - my $ekey = $ibx->eidx_key; if ($v == 2) { - my $epoch_max; - defined($ibx->git_dir_latest(\$epoch_max)) or return; - $sync->{epoch_max} = $epoch_max; + $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 $head = $ibx->mm->last_commit; - unless (defined $head) { - warn "E: $ibx->{inboxdir} is not indexed\n"; - return; - } + 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; + return "E: $ekey unsupported inbox version (v$v)"; } for my $unit (@{delete($sync->{todo}) // []}) { last if $sync->{quit}; @@ -312,6 +302,7 @@ sub _sync_inbox ($$$) { } $self->{midx}->index_ibx($ibx) unless $sync->{quit}; $ibx->git->cleanup; # done with this inbox, now + undef; } sub gc_unref_doc ($$$$) { @@ -395,23 +386,65 @@ DELETE FROM xref3 WHERE docid NOT IN (SELECT num FROM over) done($self); } +sub _ibx_for ($$$) { + 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 = $req->{by_chash}; + 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 - $self->{transact_bytes} += $smsg->{bytes}; - if ($nr == 1) { # likely, all good - $self->idx_shard($docid)->shard_reindex_docid($docid); - return; + check_batch_limit({ %$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 $chash0 = $smsg->{chash} // die "BUG: $smsg->{blob} no {chash}"; - delete($by_chash->{$chash0}) // die "BUG: $smsg->{blob} chash missing"; my @todo; for my $ary (values %$by_chash) { for my $x (reverse @$ary) { @@ -419,21 +452,21 @@ sub _reindex_finalize ($$$) { my $n = $self->{oidx}->remove_xref3($docid, $x->{blob}); die "BUG: $x->{blob} invalidated #$docid" if $n == 0; } - my $x = $ary->[-1] // die "BUG: #$docid {by_chash} empty"; + my $x = pop(@$ary) // die "BUG: #$docid {by_chash} empty"; $x->{num} = delete($x->{xnum}) // die '{xnum} unset'; - my $ibx_id = delete($x->{ibx_id}) // die '{ibx_id} unset'; - my $pos = $sync->{id2pos}->{$ibx_id} // die "$ibx_id no pos"; - my $ibx = $self->{ibx_list}->[$pos] // - die "BUG: ibx for $x->{blob} not mapped"; - my $e = $ibx->over->get_art($x->{num}); - $e->{blob} eq $x->{blob} or die <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; + push @todo, $ibx, $e; + $over->dbh_close if _fd_constrained($self); + } else { + die "$ibx->{inboxdir}: over.sqlite3 unusable: $!\n"; + } } - $self->{oidx}->commit_lazy; # ensure shard workers can see xref removals - $self->{oidx}->begin_lazy; - $self->idx_shard($docid)->shard_reindex_docid($docid); + undef $by_chash; while (my ($ibx, $e) = splice(@todo, 0, 2)) { reindex_unseen($self, $sync, $ibx, $e); } @@ -444,14 +477,14 @@ sub _reindex_oid { # git->cat_async callback my $sync = $req->{sync}; my $self = $sync->{self}; my $orig_smsg = $req->{orig_smsg} // die 'BUG: no {orig_smsg}'; - my $expect_oid = $req->{xr3}->[$req->{ix}]->[2]; + 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->{xr3}->[++$req->{ix}]->[2]) { + } 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"; @@ -468,10 +501,11 @@ sub _reindex_oid { # git->cat_async callback cotime => $orig_smsg->{ts} }); my $chash = content_hash($eml); $re_smsg->{chash} = $chash; - $re_smsg->{xnum} = $req->{xr3}->[$req->{ix}]->[1]; - $re_smsg->{ibx_id} = $req->{xr3}->[$req->{ix}]->[0]; + $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->{xr3}->[++$req->{ix}]->[2]) { + 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"; @@ -492,7 +526,7 @@ BUG? #$docid $smsg->{blob} is not referenced by inboxes during reindex return; } - # we sort {xr3} in the reverse order of {ibx_list} so we can + # 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} @@ -502,32 +536,117 @@ BUG? #$docid $smsg->{blob} is not referenced by inboxes during reindex $b->[1] <=> $a->[1] # break ties with {xnum} } @$xr3; @$xr3 = map { [ $_->[0], $_->[1], unpack('H*', $_->[2]) ] } @$xr3; - my $req = { orig_smsg => $smsg, sync => $sync, xr3 => $xr3, ix => 0 }; + 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 = "$$-${\time}-$>-".host_ident; + $self->{oidx}->eidx_meta('eidxq_lock', $val); + $self->{-eidxq_locked} = $val; +} + +sub eidxq_lock_acquire ($) { + my ($self) = @_; + my $oidx = $self->{oidx}; + $oidx->begin_lazy; + my $cur = $oidx->eidx_meta('eidxq_lock') || return _eidxq_take($self); + if (my $locked = $self->{-eidxq_locked}) { # be lazy + return $locked if $locked eq $cur; + } + my ($pid, $time, $euid, $ident) = split(/-/, $cur, 4); + my $t = strftime('%Y-%m-%d %k:%M:%S', gmtime($time)); + if ($euid == $> && $ident eq host_ident) { + if (kill(0, $pid)) { + warn <dbh->sqlite_db_filename; + warn <{oidx}->commit_lazy; # ensure shard workers can see it - $self->{oidx}->begin_lazy; + return unless eidxq_lock_acquire($self); my $dbh = $self->{oidx}->dbh; my $tot = $dbh->selectrow_array('SELECT COUNT(*) FROM eidxq') or return; ${$sync->{nr}} = 0; - $sync->{-regen_fmt} = "%u/$tot\n"; + local $sync->{-regen_fmt} = "%u/$tot\n"; my $pr = $sync->{-opt}->{-progress}; if ($pr) { my $min = $dbh->selectrow_array('SELECT MIN(docid) FROM eidxq'); my $max = $dbh->selectrow_array('SELECT MAX(docid) FROM eidxq'); $pr->("Xapian indexing $min..$max (total=$tot)\n"); } - my %id2pos; - my $pos = 0; - $id2pos{$_->{-ibx_id}} = $pos++ for @{$self->{ibx_list}}; - $sync->{id2pos} = \%id2pos; - - my $del = $dbh->prepare('DELETE FROM eidxq WHERE docid = ?'); - my $iter = $dbh->prepare('SELECT docid FROM eidxq ORDER BY docid ASC'); + $sync->{id2pos} //= do { + my %id2pos; + my $pos = 0; + $id2pos{$_->{-ibx_id}} = $pos++ for @{$self->{ibx_list}}; + \%id2pos; + }; + my ($del, $iter); +restart: + $del = $dbh->prepare('DELETE FROM eidxq WHERE docid = ?'); + $iter = $dbh->prepare('SELECT docid FROM eidxq ORDER BY docid ASC'); $iter->execute; while (defined(my $docid = $iter->fetchrow_array)) { last if $sync->{quit}; @@ -537,20 +656,14 @@ sub eidxq_process ($$) { # for reindexing warn "E: #$docid does not exist in over\n"; } $del->execute($docid); + ++${$sync->{nr}}; - my $cur = ++${$sync->{nr}}; - - # shards flush on their own, just don't queue up too many - # deletes - if ($self->{transact_bytes} >= $self->{batch_bytes}) { - $self->git->async_wait_all; - $self->{oidx}->commit_lazy; - $self->{oidx}->begin_lazy; - $pr->("reindexed $cur/$tot\n") if $pr; - $self->{transact_bytes} = 0; + if (checkpoint_due($sync)) { + $dbh = $del = $iter = undef; + reindex_checkpoint($self, $sync); # release lock + $dbh = $self->{oidx}->dbh; + goto restart; } - # this is only for SIGUSR1, shards do their own accounting: - reindex_checkpoint($self, $sync) if ${$sync->{need_checkpoint}}; } $self->git->async_wait_all; $pr->("reindexed ${$sync->{nr}}/$tot\n") if $pr; @@ -589,16 +702,29 @@ sub reindex_unseen ($$$$) { sub _reindex_check_unseen ($$$) { my ($self, $sync, $ibx) = @_; my $ibx_id = $ibx->{-ibx_id}; - my ($beg, $end) = (1, 1000); + my $slice = 1000; + my ($beg, $end) = (1, $slice); # first, check if we missed any messages in target $ibx - my $inx3 = $self->{oidx}->dbh->prepare(<<''); -SELECT DISTINCT(docid) FROM xref3 WHERE ibx_id = ? AND xnum = ? AND oidbin = ? - 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 + 1000; + $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); @@ -623,57 +749,88 @@ SELECT DISTINCT(docid) FROM xref3 WHERE ibx_id = ? AND xnum = ? AND oidbin = ? sub _reindex_check_stale ($$$) { my ($self, $sync, $ibx) = @_; - - # now, check if there's stale xrefs - my $get_xnum = $self->{oidx}->dbh->prepare(<<''); -SELECT docid,xnum,oidbin FROM xref3 WHERE ibx_id = ? ORDER BY docid ASC - - $get_xnum->execute($ibx->{-ibx_id}); - my $del_xref3 = $self->{oidx}->dbh->prepare(<<''); + 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 = ? - while (my ($docid, $xnum, $oidbin) = $get_xnum->fetchrow_array) { - last if $sync->{quit}; - 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 - } - warn $ibx->eidx_key . ":$xnum:$oidhex (#$docid): $err\n"; - $del_xref3->bind_param(1, $ibx->{-ibx_id}); - $del_xref3->bind_param(2, $xnum); - $del_xref3->bind_param(3, $oidbin, SQL_BLOB); - $del_xref3->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 + $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, $ibx) = @_; - _reindex_check_unseen($self, $sync, $ibx); - _reindex_check_stale($self, $sync, $ibx) unless $sync->{quit}; + 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; + } for my $ibx (@{$self->{ibx_list}}) { _reindex_inbox($self, $sync, $ibx); last if $sync->{quit}; @@ -682,10 +839,17 @@ sub eidx_reindex { eidxq_process($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) = @_; - 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}, ': ', @_); @@ -694,6 +858,8 @@ sub eidx_sync { # main entry point $self->{oidx}->rethread_prepare($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 @@ -709,17 +875,24 @@ sub eidx_sync { # main entry point for my $ibx (@{$self->{ibx_list}}) { $ibx->{-ibx_id} //= $self->{oidx}->ibx_id($ibx->eidx_key); } - eidx_reindex($self, $sync) if delete($opt->{reindex}); + if (delete($opt->{reindex})) { + local $sync->{checkpoint_unlocks} = 1; + eidx_reindex($self, $sync); + } # don't use $_ here, it'll get clobbered by reindex_checkpoint - for my $ibx (@{$self->{ibx_list}}) { - last if $sync->{quit}; - _sync_inbox($self, $sync, $ibx); + if ($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 @@ -762,18 +935,31 @@ sub idx_init { # similar to V2Writable 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, @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 { @@ -783,6 +969,22 @@ sub idx_init { # similar to V2Writable 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"; chomp(my $d = $line); @@ -800,16 +1002,125 @@ sub idx_init { # similar to V2Writable PublicInbox::V2Writable::write_alternates($info_dir, $mode, $o); } $self->parallel_init($self->{indexlevel}); - $self->umask_prepare; $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;