X-Git-Url: http://www.git.stargrave.org/?a=blobdiff_plain;f=lib%2FPublicInbox%2FSearchIdxShard.pm;h=000abd9413b5ba5159f2d540f8344805564c6ad9;hb=99a78c2ac51f63de41ccfc27e504d9b0e54d0592;hp=dcfeb0beec76c1c9e452cfad24a118e62c1cc007;hpb=44de182766037948d62bc2a8ba924de2264dd5fc;p=public-inbox.git diff --git a/lib/PublicInbox/SearchIdxShard.pm b/lib/PublicInbox/SearchIdxShard.pm index dcfeb0be..000abd94 100644 --- a/lib/PublicInbox/SearchIdxShard.pm +++ b/lib/PublicInbox/SearchIdxShard.pm @@ -1,4 +1,4 @@ -# Copyright (C) 2018-2020 all contributors +# Copyright (C) 2018-2021 all contributors # License: AGPL-3.0+ # Internal interface for a single Xapian shard in V2 inboxes. @@ -6,11 +6,8 @@ package PublicInbox::SearchIdxShard; use strict; use v5.10.1; -use parent qw(PublicInbox::SearchIdx); -use bytes qw(length); -use IO::Handle (); # autoflush -use PublicInbox::Eml; -use PublicInbox::Sigfd; +use parent qw(PublicInbox::SearchIdx PublicInbox::IPC); +use PublicInbox::OnDestroy; sub new { my ($class, $v2w, $shard) = @_; # v2w may be ExtSearchIdx @@ -21,195 +18,61 @@ sub new { $self->idx_acquire; $self->set_metadata_once; $self->idx_release; - $self->spawn_worker($v2w, $shard) if $v2w->{parallel}; - $self; -} - -sub spawn_worker { - my ($self, $v2w, $shard) = @_; - my ($r, $w); - pipe($r, $w) or die "pipe failed: $!\n"; - $w->autoflush(1); - my $oldset = PublicInbox::Sigfd::block_signals(); - my $pid = fork; - defined $pid or die "fork failed: $!\n"; - if ($pid == 0) { - # these signals are localized in parent - $SIG{$_} = 'IGNORE' for (qw(TERM INT QUIT)); - PublicInbox::Sigfd::sig_setmask($oldset); - my $bnote = $v2w->atfork_child; - close $w or die "failed to close: $!"; - - # F_SETPIPE_SZ = 1031 on Linux; increasing the pipe size here - # speeds V2Writable batch imports across 8 cores by nearly 20% - fcntl($r, 1031, 1048576) if $^O eq 'linux'; - - eval { shard_worker_loop($self, $v2w, $r, $shard, $bnote) }; - die "worker $shard died: $@\n" if $@; - die "unexpected MM $self->{mm}" if $self->{mm}; - exit; - } - PublicInbox::Sigfd::sig_setmask($oldset); - $self->{pid} = $pid; - $self->{w} = $w; - close $r or die "failed to close: $!"; -} - -sub eml ($$) { - my ($r, $len) = @_; - my $n = read($r, my $bref, $len) or die "read: $!\n"; - $n == $len or die "short read: $n != $len\n"; - PublicInbox::Eml->new(\$bref); -} - -# this reads all the writes to $self->{w} from the parent process -sub shard_worker_loop ($$$$$) { - my ($self, $v2w, $r, $shard, $bnote) = @_; - $0 = "shard[$shard]"; - $self->begin_txn_lazy; - while (my $line = readline($r)) { - $v2w->{current_info} = "[$shard] $line"; - if ($line eq "commit\n") { - $self->commit_txn_lazy; - } elsif ($line eq "close\n") { - $self->idx_release; - } elsif ($line eq "barrier\n") { - $self->commit_txn_lazy; - # no need to lock < 512 bytes is atomic under POSIX - print $bnote "barrier $shard\n" or - die "write failed for barrier $!\n"; - } elsif ($line =~ /\AD ([a-f0-9]{40,}) ([0-9]+)\n\z/s) { - $self->remove_by_oid($1, $2 + 0); - } elsif ($line =~ s/\A\+X //) { - my ($len, $docid, $oid, $eidx_key) = - split(/ /, $line, 4); - chomp $eidx_key; - $self->add_eidx_info($docid, $oid, $eidx_key, - eml($r, $len)); - } elsif ($line =~ s/\A-X //) { - my ($len, $docid, $oid, $eidx_key) = - split(/ /, $line, 4); - chomp $eidx_key; - $self->remove_eidx_info($docid, $oid, $eidx_key, - eml($r, $len)); - } else { - chomp $line; - my $eidx_key; - if ($line =~ s/\AX=(.+)\0//) { - $eidx_key = $1; - $v2w->{current_info} =~ s/\0/\\0 /; - } - # n.b. $mid may contain spaces(!) - my ($len, $bytes, $num, $oid, $ds, $ts, $tid, $mid) - = split(/ /, $line, 8); - $self->begin_txn_lazy; - my $smsg = bless { - bytes => $bytes, - num => $num + 0, - blob => $oid, - mid => $mid, - tid => $tid, - ds => $ds, - ts => $ts, - }, 'PublicInbox::Smsg'; - $smsg->{eidx_key} = $eidx_key if defined($eidx_key); - $self->add_message(eml($r, $len), $smsg); + if ($v2w->{parallel}) { + local $self->{-v2w_afc} = $v2w; + $self->ipc_worker_spawn("shard[$shard]"); + # F_SETPIPE_SZ = 1031 on Linux; increasing the pipe size for + # inputs speeds V2Writable batch imports across 8 cores by + # nearly 20%. Since any of our responses are small, make + # the response pipe as small as possible + if ($^O eq 'linux') { + fcntl($self->{-ipc_req}, 1031, 1048576); + fcntl($self->{-ipc_res}, 1031, 4096); } } - $self->worker_done; -} - -sub index_raw { - my ($self, $msgref, $eml, $smsg, $ibx) = @_; - if (my $w = $self->{w}) { - my @ekey = $ibx ? ('X='.$ibx->eidx_key."\0") : (); - $msgref //= \($eml->as_string); - $smsg->{raw_bytes} //= length($$msgref); - # mid must be last, it can contain spaces (but not LF) - print $w @ekey, join(' ', @$smsg{qw(raw_bytes bytes - num blob ds ts tid mid)}), - "\n", $$msgref or die "failed to write shard $!\n"; - } else { - if ($eml) { - undef($$msgref) if $msgref; - } else { # --xapian-only + --sequential-shard: - $eml = PublicInbox::Eml->new($msgref); - } - $self->begin_txn_lazy; - $smsg->{eidx_key} = $ibx->eidx_key if $ibx; - $self->add_message($eml, $smsg); - } + $self; } -sub shard_add_eidx_info { - my ($self, $docid, $oid, $xibx, $eml) = @_; - my $eidx_key = $xibx->eidx_key; - if (my $w = $self->{w}) { - my $hdr = $eml->header_obj->as_string; - my $len = length($hdr); - print $w "+X $len $docid $oid $eidx_key\n", $hdr or - die "failed to write shard: $!"; - } else { - $self->add_eidx_info($docid, $oid, $eidx_key, $eml); +sub _worker_done { + my ($self) = @_; + if ($self->need_xapian) { + die "$$ $0 xdb not released\n" if $self->{xdb}; } + die "$$ $0 still in transaction\n" if $self->{txn}; } -sub shard_remove_eidx_info { - my ($self, $docid, $oid, $xibx, $eml) = @_; - my $eidx_key = $xibx->eidx_key; - if (my $w = $self->{w}) { - my $hdr = $eml->header_obj->as_string; - my $len = length($hdr); - print $w "-X $len $docid $oid $eidx_key\n", $hdr or - die "failed to write shard: $!"; - } else { - $self->remove_eidx_info($docid, $oid, $eidx_key, $eml); - } +sub ipc_atfork_child { # called automatically before ipc_worker_loop + my ($self) = @_; + my $v2w = delete $self->{-v2w_afc} or die 'BUG: {-v2w_afc} missing'; + $v2w->atfork_child; # calls ipc_sibling_atfork_child on our siblings + $v2w->{current_info} = "[$self->{shard}]"; # for $SIG{__WARN__} + $self->begin_txn_lazy; + # caller must capture this: + PublicInbox::OnDestroy->new($$, \&_worker_done, $self); } -sub atfork_child { - close $_[0]->{w} or die "failed to close write pipe: $!\n"; +sub index_eml { + my ($self, $eml, $smsg, $eidx_key) = @_; + $smsg->{eidx_key} = $eidx_key if defined $eidx_key; + $self->ipc_do('add_xapian', $eml, $smsg); } -sub shard_barrier { - my ($self) = @_; - if (my $w = $self->{w}) { - print $w "barrier\n" or die "failed to print: $!"; - } else { - $self->commit_txn_lazy; - } +# wait for return to determine when ipc_do('commit_txn_lazy') is done +sub echo { + shift; + "@_"; } -sub shard_commit { +sub idx_close { my ($self) = @_; - if (my $w = $self->{w}) { - print $w "commit\n" or die "failed to write commit: $!"; - } else { - $self->commit_txn_lazy; - } + die "transaction in progress $self\n" if $self->{txn}; + $self->idx_release if $self->{xdb}; } sub shard_close { my ($self) = @_; - if (my $w = delete $self->{w}) { - my $pid = delete $self->{pid} or die "no process to wait on\n"; - print $w "close\n" or die "failed to write to pid:$pid: $!\n"; - close $w or die "failed to close pipe for pid:$pid: $!\n"; - waitpid($pid, 0) == $pid or die "remote process did not finish"; - $? == 0 or die ref($self)." pid:$pid exited with: $?"; - } else { - die "transaction in progress $self\n" if $self->{txn}; - $self->idx_release if $self->{xdb}; - } -} - -sub shard_remove { - my ($self, $oid, $num) = @_; - if (my $w = $self->{w}) { # triggers remove_by_oid in a shard child - print $w "D $oid $num\n" or die "failed to write remove $!"; - } else { # same process - $self->remove_by_oid($oid, $num); - } + $self->ipc_do('idx_close'); + $self->ipc_worker_stop; } 1;