X-Git-Url: http://www.git.stargrave.org/?a=blobdiff_plain;f=lib%2FPublicInbox%2FSearchIdxShard.pm;h=000abd9413b5ba5159f2d540f8344805564c6ad9;hb=99a78c2ac51f63de41ccfc27e504d9b0e54d0592;hp=544268819070c675bfa9d3e2a3edc92c11ffb2a2;hpb=592f8239dceb6604e538fd51e62bcce50d6e1972;p=public-inbox.git diff --git a/lib/PublicInbox/SearchIdxShard.pm b/lib/PublicInbox/SearchIdxShard.pm index 54426881..000abd94 100644 --- a/lib/PublicInbox/SearchIdxShard.pm +++ b/lib/PublicInbox/SearchIdxShard.pm @@ -1,119 +1,78 @@ -# Copyright (C) 2018-2020 all contributors +# Copyright (C) 2018-2021 all contributors # License: AGPL-3.0+ -# used to interface with a single Xapian shard in V2 repos. +# Internal interface for a single Xapian shard in V2 inboxes. # See L for more info on how we shard Xapian package PublicInbox::SearchIdxShard; use strict; use v5.10.1; -use parent qw(PublicInbox::SearchIdx); -use IO::Handle (); # autoflush -use PublicInbox::Eml; +use parent qw(PublicInbox::SearchIdx PublicInbox::IPC); +use PublicInbox::OnDestroy; sub new { - my ($class, $v2writable, $shard) = @_; - my $ibx = $v2writable->{-inbox}; - my $self = $class->SUPER::new($ibx, 1, $shard); + my ($class, $v2w, $shard) = @_; # v2w may be ExtSearchIdx + my $ibx = $v2w->{ibx}; + my $self = $ibx ? $class->SUPER::new($ibx, 1, $shard) + : $class->eidx_shard_new($v2w, $shard); # create the DB before forking: - $self->_xdb_acquire; - $self->set_indexlevel; - $self->_xdb_release; - $self->spawn_worker($v2writable, $shard) if $v2writable->{parallel}; + $self->idx_acquire; + $self->set_metadata_once; + $self->idx_release; + if ($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); + } + } $self; } -sub spawn_worker { - my ($self, $v2w, $shard) = @_; - my ($r, $w); - pipe($r, $w) or die "pipe failed: $!\n"; - $w->autoflush(1); - my $pid = fork; - defined $pid or die "fork failed: $!\n"; - if ($pid == 0) { - 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 $self->{mm}; - exit; +sub _worker_done { + my ($self) = @_; + if ($self->need_xapian) { + die "$$ $0 xdb not released\n" if $self->{xdb}; } - $self->{pid} = $pid; - $self->{w} = $w; - close $r or die "failed to close: $!"; + die "$$ $0 still in transaction\n" if $self->{txn}; } -sub shard_worker_loop ($$$$$) { - my ($self, $v2w, $r, $shard, $bnote) = @_; - $0 = "pi-v2-shard[$shard]"; +sub ipc_atfork_child { # called automatically before ipc_worker_loop + my ($self) = @_; + my $v2w = delete $self->{-v2w_afc} or die 'BUG: {-v2w_afc} missing'; + $v2w->atfork_child; # calls ipc_sibling_atfork_child on our siblings + $v2w->{current_info} = "[$self->{shard}]"; # for $SIG{__WARN__} $self->begin_txn_lazy; - while (my $line = readline($r)) { - $v2w->{current_info} = "[$shard] $line"; - if ($line eq "commit\n") { - $self->commit_txn_lazy; - } elsif ($line eq "close\n") { - $self->_xdb_release; - } elsif ($line eq "barrier\n") { - $self->commit_txn_lazy; - # no need to lock < 512 bytes is atomic under POSIX - print $bnote "barrier $shard\n" or - die "write failed for barrier $!\n"; - } elsif ($line =~ /\AD ([a-f0-9]{40,}) ([0-9]+)\n\z/s) { - $self->remove_by_oid($1, $2 + 0); - } else { - chomp $line; - # n.b. $mid may contain spaces(!) - my ($to_read, $bytes, $num, $blob, $ds, $ts, $mid) = - split(/ /, $line, 7); - $self->begin_txn_lazy; - my $n = read($r, my $msg, $to_read) or die "read: $!\n"; - $n == $to_read or die "short read: $n != $to_read\n"; - my $mime = PublicInbox::Eml->new(\$msg); - my $smsg = bless { - bytes => $bytes, - num => $num + 0, - blob => $blob, - mid => $mid, - ds => $ds, - ts => $ts, - }, 'PublicInbox::Smsg'; - $self->add_message($mime, $smsg); - } - } - $self->worker_done; + # caller must capture this: + PublicInbox::OnDestroy->new($$, \&_worker_done, $self); } -# called by V2Writable -sub index_raw { - my ($self, $msgref, $mime, $smsg) = @_; - if (my $w = $self->{w}) { - # mid must be last, it can contain spaces (but not LF) - print $w join(' ', @$smsg{qw(raw_bytes bytes - num blob ds ts mid)}), - "\n", $$msgref or die "failed to write shard $!\n"; - } else { - $$msgref = undef; - $self->begin_txn_lazy; - $self->add_message($mime, $smsg); - } +sub index_eml { + my ($self, $eml, $smsg, $eidx_key) = @_; + $smsg->{eidx_key} = $eidx_key if defined $eidx_key; + $self->ipc_do('add_xapian', $eml, $smsg); } -sub atfork_child { - close $_[0]->{w} or die "failed to close write pipe: $!\n"; +# wait for return to determine when ipc_do('commit_txn_lazy') is done +sub echo { + shift; + "@_"; } -# called by V2Writable: -sub remote_barrier { +sub idx_close { my ($self) = @_; - if (my $w = $self->{w}) { - print $w "barrier\n" or die "failed to print: $!"; - } 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) = @_; + $self->ipc_do('idx_close'); + $self->ipc_worker_stop; } 1;