X-Git-Url: http://www.git.stargrave.org/?a=blobdiff_plain;f=lib%2FPublicInbox%2FXapcmd.pm;h=22e56f42263bf2e85b4a857cfd7b8c4a8381228c;hb=3c39f9c942a6975245fda878e9b957d8d3367662;hp=7e3d47fb9b26240e4fc563afcbf01ccb61b61b60;hpb=de1a674545481c779bc49feef028eacc1e181d5d;p=public-inbox.git diff --git a/lib/PublicInbox/Xapcmd.pm b/lib/PublicInbox/Xapcmd.pm index 7e3d47fb..22e56f42 100644 --- a/lib/PublicInbox/Xapcmd.pm +++ b/lib/PublicInbox/Xapcmd.pm @@ -17,34 +17,66 @@ our @COMPACT_OPT = qw(jobs|j=i quiet|q blocksize|b=s no-full|n fuller|F); sub commit_changes ($$$) { my ($ibx, $tmp, $opt) = @_; - + my $reshard = $opt->{reshard}; my $reindex = $opt->{reindex}; my $im = $ibx->importer(0); $im->lock_acquire if !$opt->{-coarse_lock}; $SIG{INT} or die 'BUG: $SIG{INT} not handled'; + my @old_shard; while (my ($old, $new) = each %$tmp) { - my @st = stat($old) or die "failed to stat($old): $!\n"; + my @st = stat($old); + if (!@st && !defined($opt->{reshard})) { + die "failed to stat($old): $!"; + } my $over = "$old/over.sqlite3"; if (-f $over) { # only for v1, v2 over is untouched + defined $new or die "BUG: $over exists when culling v2"; $over = PublicInbox::Over->new($over); my $tmp_over = "$new/over.sqlite3"; $over->connect->sqlite_backup_to_file($tmp_over); $over = undef; } - chmod($st[2] & 07777, $new) or die "chmod $old: $!\n"; + if (!defined($new)) { # culled shard + push @old_shard, $old; + next; + } + + if (@st) { + chmod($st[2] & 07777, $new) or die "chmod $old: $!\n"; + rename($old, "$new/old") or + die "rename $old => $new/old: $!\n"; + } # Xtmpdir->DESTROY won't remove $new after this: - rename($old, "$new/old") or die "rename $old => $new/old: $!\n"; rename($new, $old) or die "rename $new => $old: $!\n"; - my $prev = "$old/old"; - remove_tree($prev) or die "failed to remove $prev: $!\n"; + if (@st) { + my $prev = "$old/old"; + remove_tree($prev) or + die "failed to remove $prev: $!\n"; + } } + remove_tree(@old_shard); $tmp->done; if (!$opt->{-coarse_lock}) { $opt->{-skip_lock} = 1; + + if ($im->can('count_shards')) { + my $pr = $opt->{-progress}; + my $n = $im->count_shards; + if (defined $reshard && $n != $reshard) { + die +"BUG: counted $n shards after resharding to $reshard"; + } + my $prev = $im->{shards}; + if ($pr && $prev != $n) { + $pr->("shard count changed: $prev => $n\n"); + $im->{shards} = $n; + } + } + PublicInbox::Admin::index_inbox($ibx, $opt); # implicit lock_release } else { @@ -121,7 +153,7 @@ sub run { my ($ibx, $task, $opt) = @_; # task = 'cpdb' or 'compact' my $cb = \&${\"PublicInbox::Xapcmd::$task"}; PublicInbox::Admin::progress_prepare($opt ||= {}); - my $dir = $ibx->{mainrepo} or die "no mainrepo in inbox\n"; + my $dir = $ibx->{inboxdir} or die "no inboxdir in inbox\n"; runnable_or_die($XAPIAN_COMPACT) if $opt->{compact}; my $reindex; # v1:{ from => $x40 }, v2:{ from => [ $x40, $x40, .. ] } } my $from; # per-epoch ranges @@ -139,31 +171,59 @@ sub run { my $tmp = PublicInbox::Xtmpdirs->new; my $v = $ibx->{version} ||= 1; my @q; + my $reshard = $opt->{reshard}; + if (defined $reshard && $reshard <= 0) { + die "--reshard must be a positive number\n"; + } # we want temporary directories to be as deep as possible, - # so v2 partitions can keep "xap$SCHEMA_VERSION" on a separate FS. + # so v2 shards can keep "xap$SCHEMA_VERSION" on a separate FS. if ($v == 1) { + if (defined $reshard) { + warn +"--reshard=$reshard ignored for v1 $ibx->{inboxdir}\n"; + } my $old_parent = dirname($old); same_fs_or_die($old_parent, $old); - $tmp->{$old} = tempdir('xapcmd-XXXXXXXX', DIR => $old_parent); - push @q, [ $old, $tmp->{$old} ]; + my $v = PublicInbox::Search::SCHEMA_VERSION(); + my $wip = tempdir("xapian$v-XXXXXXXX", DIR => $old_parent); + $tmp->{$old} = $wip; + push @q, [ $old, $wip ]; } else { opendir my $dh, $old or die "Failed to opendir $old: $!\n"; + my @old_shards; while (defined(my $dn = readdir($dh))) { - if ($dn =~ /\A\d+\z/) { - my $tmpl = "$dn-XXXXXXXX"; - my $dst = tempdir($tmpl, DIR => $old); - same_fs_or_die($old, $dst); - my $cur = "$old/$dn"; - push @q, [ $cur, $dst ]; - $tmp->{$cur} = $dst; + if ($dn =~ /\A[0-9]+\z/) { + push @old_shards, $dn; } elsif ($dn eq '.' || $dn eq '..') { } elsif ($dn =~ /\Aover\.sqlite3/) { } else { warn "W: skipping unknown dir: $old/$dn\n" } } - die "No Xapian parts found in $old\n" unless @q; + die "No Xapian shards found in $old\n" unless @old_shards; + + my ($src, $max_shard); + if (!defined($reshard) || $reshard == scalar(@old_shards)) { + # 1:1 copy + $max_shard = scalar(@old_shards) - 1; + } else { + # M:N copy + $max_shard = $reshard - 1; + $src = [ map { "$old/$_" } @old_shards ]; + } + foreach my $dn (0..$max_shard) { + my $tmpl = "$dn-XXXXXXXX"; + my $wip = tempdir($tmpl, DIR => $old); + same_fs_or_die($old, $wip); + my $cur = "$old/$dn"; + push @q, [ $src // $cur , $wip ]; + $tmp->{$cur} = $wip; + } + # mark old shards to be unlinked + if ($src) { + $tmp->{$_} ||= undef for @$src; + } } my $im = $ibx->importer(0); my $max = $opt->{jobs} || scalar(@q); @@ -197,10 +257,11 @@ sub cpdb_retryable ($$) { } sub progress_pfx ($) { - my @p = split('/', $_[0]); + my ($wip) = @_; # tempdir v2: ([0-9])+-XXXXXXXX + my @p = split('/', $wip); # return "xap15/0" for v2, or "xapian15" for v1: - ($p[-1] =~ /\A\d+\z/) ? "$p[-2]/$p[-1]" : $p[-1]; + ($p[-1] =~ /\A([0-9]+)/) ? "$p[-2]/$1" : $p[-1]; } # xapian-compact wrapper @@ -243,12 +304,74 @@ sub compact ($$) { } } +sub cpdb_loop ($$$;$$) { + my ($src, $dst, $pr_data, $cur_shard, $reshard) = @_; + my ($pr, $fmt, $nr, $pfx); + if ($pr_data) { + $pr = $pr_data->{pr}; + $fmt = $pr_data->{fmt}; + $nr = \($pr_data->{nr}); + $pfx = $pr_data->{pfx}; + } + + my ($it, $end); + do { + eval { + $it = $src->postlist_begin(''); + $end = $src->postlist_end(''); + }; + } while (cpdb_retryable($src, $pfx)); + + do { + eval { + for (; $it != $end; $it++) { + my $docid = $it->get_docid; + if (defined $reshard) { + my $dst_shard = $docid % $reshard; + next if $dst_shard != $cur_shard; + } + my $doc = $src->get_document($docid); + $dst->replace_document($docid, $doc); + if ($pr_data && !(++$$nr & 1023)) { + $pr->(sprintf($fmt, $$nr)); + } + } + + # unlike copydatabase(1), we don't copy spelling + # and synonym data (or other user metadata) since + # the Perl APIs don't expose iterators for them + # (and public-inbox does not use those features) + }; + } while (cpdb_retryable($src, $pfx)); +} + # Like copydatabase(1), this is horribly slow; and it doesn't seem due # to the overhead of Perl. sub cpdb ($$) { my ($args, $opt) = @_; my ($old, $new) = @$args; - my $src = Search::Xapian::Database->new($old); + my ($src, $cur_shard); + my $reshard; + if (ref($old) eq 'ARRAY') { + ($cur_shard) = ($new =~ m!xap[0-9]+/([0-9]+)\b!); + defined $cur_shard or + die "BUG: could not extract shard # from $new"; + $reshard = $opt->{reshard}; + defined $reshard or die 'BUG: got array src w/o --reshard'; + + # resharding, M:N copy means have full read access + foreach (@$old) { + if ($src) { + my $sub = Search::Xapian::Database->new($_); + $src->add_database($sub); + } else { + $src = Search::Xapian::Database->new($_); + } + } + } else { + $src = Search::Xapian::Database->new($old); + } + my ($xtmp, $tmp); if ($opt->{compact}) { my $newdir = dirname($new); @@ -264,10 +387,9 @@ sub cpdb ($$) { # of other bugs: my $creat = Search::Xapian::DB_CREATE(); my $dst = Search::Xapian::WritableDatabase->new($tmp, $creat); - my ($it, $end); - my ($nr, $tot, $fmt); # progress output my $pr = $opt->{-progress}; - my $pfx = $opt->{-progress_pfx} = progress_pfx($old); + my $pfx = $opt->{-progress_pfx} = progress_pfx($new); + my $pr_data = { pr => $pr, pfx => $pfx, nr => 0 } if $pr; do { eval { @@ -275,37 +397,46 @@ sub cpdb ($$) { my $lc = $src->get_metadata('last_commit'); $dst->set_metadata('last_commit', $lc) if $lc; - $it = $src->postlist_begin(''); - $end = $src->postlist_end(''); - if ($pr) { - $nr = 0; - $tot = $src->get_doccount; - $fmt = "$pfx % ".length($tot)."u/$tot\n"; - $pr->("$pfx copying $tot documents\n"); + # only the first xapian shard (0) gets 'indexlevel' + if ($new =~ m!(?:xapian[0-9]+|xap[0-9]+/0)\b!) { + my $l = $src->get_metadata('indexlevel'); + if ($l eq 'medium') { + $dst->set_metadata('indexlevel', $l); + } } - }; - } while (cpdb_retryable($src, $pfx)); - - do { - eval { - while ($it != $end) { - my $docid = $it->get_docid; - my $doc = $src->get_document($docid); - $dst->replace_document($docid, $doc); - $it->inc; - if ($pr && !(++$nr & 1023)) { - $pr->(sprintf($fmt, $nr)); + if ($pr_data) { + my $tot = $src->get_doccount; + + # we can only estimate when resharding, + # because removed spam causes slight imbalance + my $est = ''; + if (defined $cur_shard && $reshard > 1) { + $tot = int($tot/$reshard); + $est = 'around '; } + my $fmt = "$pfx % ".length($tot)."u/$tot\n"; + $pr->("$pfx copying $est$tot documents\n"); + $pr_data->{fmt} = $fmt; + $pr_data->{total} = $tot; } - - # unlike copydatabase(1), we don't copy spelling - # and synonym data (or other user metadata) since - # the Perl APIs don't expose iterators for them - # (and public-inbox does not use those features) }; } while (cpdb_retryable($src, $pfx)); - $pr->(sprintf($fmt, $nr)) if $pr; + if (defined $reshard) { + # we rely on document IDs matching NNTP article number, + # so we can't have the Xapian sharding DB support rewriting + # document IDs. Thus we iterate through each shard + # individually. + $src = undef; + foreach (@$old) { + my $old = Search::Xapian::Database->new($_); + cpdb_loop($old, $dst, $pr_data, $cur_shard, $reshard); + } + } else { + cpdb_loop($src, $dst, $pr_data); + } + + $pr->(sprintf($pr_data->{fmt}, $pr_data->{nr})) if $pr; return unless $xtmp; $src = $dst = undef; # flushes and closes @@ -328,7 +459,7 @@ sub new { # http://www.tldp.org/LDP/abs/html/exitcodes.html $SIG{INT} = sub { exit(130) }; $SIG{HUP} = $SIG{PIPE} = $SIG{TERM} = sub { exit(1) }; - my $self = bless {}, $_[0]; # old partition => new (tmp) partition + my $self = bless {}, $_[0]; # old shard => new (WIP) shard $owner{"$self"} = $$; $self; } @@ -350,6 +481,7 @@ sub DESTROY { my $owner_pid = delete $owner{"$self"} or return; return if $owner_pid != $$; foreach my $new (values %$self) { + defined $new or next; # may be undef if resharding remove_tree($new) unless -d "$new/old"; } done($self);