X-Git-Url: http://www.git.stargrave.org/?a=blobdiff_plain;f=lib%2FPublicInbox%2FSearchIdxShard.pm;h=000abd9413b5ba5159f2d540f8344805564c6ad9;hb=99a78c2ac51f63de41ccfc27e504d9b0e54d0592;hp=15ec6578e67ee9190a246d27e40677ca233f8b2e;hpb=c477bdd8a80eecc319b680764edfb24bd12cb7b2;p=public-inbox.git diff --git a/lib/PublicInbox/SearchIdxShard.pm b/lib/PublicInbox/SearchIdxShard.pm index 15ec6578..000abd94 100644 --- a/lib/PublicInbox/SearchIdxShard.pm +++ b/lib/PublicInbox/SearchIdxShard.pm @@ -1,116 +1,78 @@ -# Copyright (C) 2018 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 warnings; -use base qw(PublicInbox::SearchIdx); +use v5.10.1; +use parent qw(PublicInbox::SearchIdx PublicInbox::IPC); +use PublicInbox::OnDestroy; sub new { - my ($class, $v2writable, $shard) = @_; - my $self = $class->SUPER::new($v2writable->{-inbox}, 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->_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, $v2writable, $shard) = @_; - my ($r, $w); - pipe($r, $w) or die "pipe failed: $!\n"; - binmode $r, ':raw'; - binmode $w, ':raw'; - my $pid = fork; - defined $pid or die "fork failed: $!\n"; - if ($pid == 0) { - my $bnote = $v2writable->atfork_child; - $v2writable = undef; - 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, $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 ($$$$) { - my ($self, $r, $shard, $bnote) = @_; - $0 = "pi-v2-shard[$shard]"; - my $current_info = ''; - my $warn_cb = $SIG{__WARN__} || sub { print STDERR @_ }; - local $SIG{__WARN__} = sub { - chomp $current_info; - $warn_cb->("[$shard] $current_info: ", @_); - }; +sub ipc_atfork_child { # called automatically before ipc_worker_loop + my ($self) = @_; + my $v2w = delete $self->{-v2w_afc} or die 'BUG: {-v2w_afc} missing'; + $v2w->atfork_child; # calls ipc_sibling_atfork_child on our siblings + $v2w->{current_info} = "[$self->{shard}]"; # for $SIG{__WARN__} $self->begin_txn_lazy; - while (my $line = $r->getline) { - $current_info = $line; - if ($line eq "commit\n") { - $self->commit_txn_lazy; - } elsif ($line eq "close\n") { - $self->_xdb_release; - } elsif ($line eq "barrier\n") { - $self->commit_txn_lazy; - # no need to lock < 512 bytes is atomic under POSIX - print $bnote "barrier $shard\n" or - die "write failed for barrier $!\n"; - } elsif ($line =~ /\AD ([a-f0-9]{40,}) (.+)\n\z/s) { - my ($oid, $mid) = ($1, $2); - $self->begin_txn_lazy; - $self->remove_by_oid($oid, $mid); - } else { - chomp $line; - my ($len, $artnum, $oid, $mid0) = split(/ /, $line); - $self->begin_txn_lazy; - my $n = read($r, my $msg, $len) or die "read: $!\n"; - $n == $len or die "short read: $n != $len\n"; - my $mime = PublicInbox::MIME->new(\$msg); - $artnum = int($artnum); - $self->add_message($mime, $n, $artnum, $oid, $mid0); - } - } - $self->worker_done; + # caller must capture this: + PublicInbox::OnDestroy->new($$, \&_worker_done, $self); } -# called by V2Writable -sub index_raw { - my ($self, $bytes, $msgref, $artnum, $oid, $mid0, $mime) = @_; - if (my $w = $self->{w}) { - print $w "$bytes $artnum $oid $mid0\n", $$msgref or die - "failed to write shard $!\n"; - $w->flush or die "failed to flush: $!\n"; - } else { - $$msgref = undef; - $self->begin_txn_lazy; - $self->add_message($mime, $bytes, $artnum, $oid, $mid0); - } +sub index_eml { + my ($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: $!"; - $w->flush or die "failed to flush: $!"; - } 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;