X-Git-Url: http://www.git.stargrave.org/?a=blobdiff_plain;f=lib%2FPublicInbox%2FV2Writable.pm;h=553dd839f39a9a56227e8a092d55e791ca54b2e0;hb=94ae705673cb03045a109041eec9a6704b8a735b;hp=0119ea7634b96166e7cda8eba96ba30612d9ac96;hpb=46e4fa33863f136e8f36ffcc65231c2bc7de11e1;p=public-inbox.git diff --git a/lib/PublicInbox/V2Writable.pm b/lib/PublicInbox/V2Writable.pm index 0119ea76..553dd839 100644 --- a/lib/PublicInbox/V2Writable.pm +++ b/lib/PublicInbox/V2Writable.pm @@ -17,11 +17,11 @@ use PublicInbox::InboxWritable; use PublicInbox::OverIdx; use PublicInbox::Msgmap; use PublicInbox::Spawn qw(spawn popen_rd); -use PublicInbox::SearchIdx; -use PublicInbox::MultiMidQueue; +use PublicInbox::SearchIdx qw(log2stack crlf_adjust is_ancestor check_size); use IO::Handle; # ->autoflush -use File::Temp qw(tempfile); +use File::Temp (); +my $OID = qr/[a-f0-9]{40,}/; # an estimate of the post-packed size to the raw uncompressed size my $PACKING_FACTOR = 0.4; @@ -34,14 +34,13 @@ my $PACKING_FACTOR = 0.4; our $NPROC_MAX_DEFAULT = 4; sub detect_nproc () { - for my $nproc (qw(nproc gnproc)) { # GNU coreutils nproc - `$nproc 2>/dev/null` =~ /^(\d+)$/ and return $1; - } - # getconf(1) is POSIX, but *NPROCESSORS* vars are not for (qw(_NPROCESSORS_ONLN NPROCESSORS_ONLN)) { `getconf $_ 2>/dev/null` =~ /^(\d+)$/ and return $1; } + for my $nproc (qw(nproc gnproc)) { # GNU coreutils nproc + `$nproc 2>/dev/null` =~ /^(\d+)$/ and return $1; + } # should we bother with `sysctl hw.ncpu`? Those only give # us total processor count, not online processor count. @@ -108,21 +107,21 @@ sub new { my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION; my $self = { - -inbox => $v2ibx, + ibx => $v2ibx, im => undef, # PublicInbox::Import parallel => 1, transact_bytes => 0, total_bytes => 0, current_info => '', xpfx => $xpfx, - over => PublicInbox::OverIdx->new("$xpfx/over.sqlite3", 1), + over => PublicInbox::OverIdx->new("$xpfx/over.sqlite3"), lock_path => "$dir/inbox.lock", # limit each git repo (epoch) to 1GB or so rotate_bytes => int((1024 * 1024 * 1024) / $PACKING_FACTOR), - last_commit => [], # git repo -> commit + last_commit => [], # git epoch -> commit }; + $self->{over}->{-no_fsync} = 1 if $v2ibx->{-no_fsync}; $self->{shards} = count_shards($self) || nproc_shards($creat); - $self->{index_max_size} = $v2ibx->{index_max_size}; bless $self, $class; } @@ -147,22 +146,19 @@ sub init_inbox { # returns undef on duplicate or spam # mimics Import::add and wraps it for v2 sub add { - my ($self, $mime, $check_cb) = @_; - $self->{-inbox}->with_umask(sub { - _add($self, $mime, $check_cb) - }); + my ($self, $eml, $check_cb) = @_; + $self->{ibx}->with_umask(\&_add, $self, $eml, $check_cb); } # indexes a message, returns true if checkpointing is needed sub do_idx ($$$$) { my ($self, $msgref, $mime, $smsg) = @_; - $smsg->{bytes} = $smsg->{raw_bytes} + - PublicInbox::SearchIdx::crlf_adjust($$msgref); + $smsg->{bytes} = $smsg->{raw_bytes} + crlf_adjust($$msgref); $self->{over}->add_overview($mime, $smsg); my $idx = idx_shard($self, $smsg->{num} % $self->{shards}); $idx->index_raw($msgref, $mime, $smsg); my $n = $self->{transact_bytes} += $smsg->{raw_bytes}; - $n >= ($PublicInbox::SearchIdx::BATCH_BYTES * $self->{shards}); + $n >= $self->{batch_bytes}; } sub _add { @@ -170,7 +166,7 @@ sub _add { # spam check: if ($check_cb) { - $mime = $check_cb->($mime, $self->{-inbox}) or return; + $mime = $check_cb->($mime, $self->{ibx}) or return; } # All pipes (> $^F) known to Perl 5.6+ have FD_CLOEXEC set, @@ -199,7 +195,7 @@ sub _add { sub v2_num_for { my ($self, $mime) = @_; - my $mids = mids($mime->header_obj); + my $mids = mids($mime); if (@$mids) { my $mid = $mids->[0]; my $num = $self->{mm}->mid_insert($mid); @@ -219,7 +215,7 @@ sub v2_num_for { # AltId may pre-populate article numbers (e.g. X-Mail-Count # or NNTP article number), use that article number if it's # not in Over. - my $altid = $self->{-inbox}->{altid}; + my $altid = $self->{ibx}->{altid}; if ($altid && grep(/:file=msgmap\.sqlite3\z/, @$altid)) { my $num = $self->{mm}->num_for($mid); @@ -246,28 +242,27 @@ sub v2_num_for { } sub v2_num_for_harder { - my ($self, $mime) = @_; + my ($self, $eml) = @_; - my $hdr = $mime->header_obj; - my $dig = content_digest($mime); - my $mid0 = PublicInbox::Import::digest2mid($dig, $hdr); + my $dig = content_digest($eml); + my $mid0 = PublicInbox::Import::digest2mid($dig, $eml); my $num = $self->{mm}->mid_insert($mid0); unless (defined $num) { # it's hard to spoof the last Received: header - my @recvd = $hdr->header_raw('Received'); + my @recvd = $eml->header_raw('Received'); $dig->add("Received: $_") foreach (@recvd); - $mid0 = PublicInbox::Import::digest2mid($dig, $hdr); + $mid0 = PublicInbox::Import::digest2mid($dig, $eml); $num = $self->{mm}->mid_insert($mid0); # fall back to a random Message-ID and give up determinism: until (defined($num)) { $dig->add(rand); - $mid0 = PublicInbox::Import::digest2mid($dig, $hdr); + $mid0 = PublicInbox::Import::digest2mid($dig, $eml); warn "using random Message-ID <$mid0> as fallback\n"; $num = $self->{mm}->mid_insert($mid0); } } - PublicInbox::Import::append_mid($hdr, $mid0); + PublicInbox::Import::append_mid($eml, $mid0); ($num, $mid0); } @@ -276,22 +271,45 @@ sub idx_shard { $self->{idx_shards}->[$shard_i]; } +sub _idx_init { # with_umask callback + my ($self, $opt) = @_; + $self->lock_acquire unless $opt && $opt->{-skip_lock}; + $self->{over}->create; + + # xcpdb can change shard count while -watch is idle + my $nshards = count_shards($self); + $self->{shards} = $nshards if $nshards && $nshards != $self->{shards}; + $self->{batch_bytes} = $opt->{batch_size} // + $PublicInbox::SearchIdx::BATCH_BYTES; + $self->{batch_bytes} *= $self->{shards} if $self->{parallel}; + + # need to create all shards before initializing msgmap FD + # idx_shards must be visible to all forked processes + my $max = $self->{shards} - 1; + my $idx = $self->{idx_shards} = []; + push @$idx, PublicInbox::SearchIdxShard->new($self, $_) for (0..$max); + + # Now that all subprocesses are up, we can open the FDs + # for SQLite: + my $mm = $self->{mm} = PublicInbox::Msgmap->new_file( + "$self->{ibx}->{inboxdir}/msgmap.sqlite3", + $self->{ibx}->{-no_fsync} ? 2 : 1); + $mm->{dbh}->begin_work; +} + # idempotent sub idx_init { my ($self, $opt) = @_; return if $self->{idx_shards}; - my $ibx = $self->{-inbox}; + my $ibx = $self->{ibx}; # do not leak read-only FDs to child processes, we only have these # FDs for duplicate detection so they should not be # frequently activated. - delete $ibx->{$_} foreach (qw(git mm search)); - - my $indexlevel = $ibx->{indexlevel}; - if ($indexlevel && $indexlevel eq 'basic') { - $self->{parallel} = 0; - } + delete @$ibx{qw(mm search)}; + $ibx->git->cleanup; + $self->{parallel} = 0 if ($ibx->{indexlevel}//'') eq 'basic'; if ($self->{parallel}) { pipe(my ($r, $w)) or die "pipe failed: $!"; # pipe for barrier notifications doesn't need to be big, @@ -301,33 +319,8 @@ sub idx_init { $w->autoflush(1); } - my $over = $self->{over}; $ibx->umask_prepare; - $ibx->with_umask(sub { - $self->lock_acquire unless ($opt && $opt->{-skip_lock}); - $over->create; - - # xcpdb can change shard count while -watch is idle - my $nshards = count_shards($self); - if ($nshards && $nshards != $self->{shards}) { - $self->{shards} = $nshards; - } - - # need to create all shards before initializing msgmap FD - my $max = $self->{shards} - 1; - - # idx_shards must be visible to all forked processes - my $idx = $self->{idx_shards} = []; - for my $i (0..$max) { - push @$idx, PublicInbox::SearchIdxShard->new($self, $i); - } - - # Now that all subprocesses are up, we can open the FDs - # for SQLite: - my $mm = $self->{mm} = PublicInbox::Msgmap->new_file( - "$self->{-inbox}->{inboxdir}/msgmap.sqlite3", 1); - $mm->{dbh}->begin_work; - }); + $ibx->with_umask(\&_idx_init, $self, $opt); } # returns an array mapping [ epoch => latest_commit ] @@ -336,7 +329,7 @@ sub idx_init { sub _replace_oids ($$$) { my ($self, $mime, $replace_map) = @_; $self->done; - my $pfx = "$self->{-inbox}->{inboxdir}/git"; + my $pfx = "$self->{ibx}->{inboxdir}/git"; my $rewrites = []; # epoch => commit my $max = $self->{epoch_max}; @@ -379,24 +372,24 @@ sub content_matches ($$) { # used for removing or replacing (purging) sub rewrite_internal ($$;$$$) { - my ($self, $old_mime, $cmt_msg, $new_mime, $sref) = @_; + my ($self, $old_eml, $cmt_msg, $new_eml, $sref) = @_; $self->idx_init; my ($im, $need_reindex, $replace_map); if ($sref) { $replace_map = {}; # oid => sref - $need_reindex = [] if $new_mime; + $need_reindex = [] if $new_eml; } else { $im = $self->importer; } my $over = $self->{over}; - my $chashes = content_hashes($old_mime); - my @removed; - my $mids = mids($old_mime->header_obj); + my $chashes = content_hashes($old_eml); + my $removed = []; + my $mids = mids($old_eml); # We avoid introducing new blobs into git since the raw content # can be slightly different, so we do not need the user-supplied # message now that we have the mids and content_hash - $old_mime = undef; + $old_eml = undef; my $mark; foreach my $mid (@$mids) { @@ -422,15 +415,15 @@ sub rewrite_internal ($$;$$$) { } foreach my $num (keys %gone) { my ($smsg, $mime, $orig) = @{$gone{$num}}; - # @removed should only be set once assuming + # $removed should only be set once assuming # no bugs in our deduplication code: - @removed = (undef, $mime, $smsg); + $removed = [ undef, $mime, $smsg ]; my $oid = $smsg->{blob}; if ($replace_map) { $replace_map->{$oid} = $sref; } else { ($mark, undef) = $im->remove($orig, $cmt_msg); - $removed[0] = $mark; + $removed->[0] = $mark; } $orig = undef; if ($need_reindex) { # ->replace @@ -447,28 +440,26 @@ sub rewrite_internal ($$;$$$) { $self->{last_commit}->[$self->{epoch_max}] = $cmt; } if ($replace_map && scalar keys %$replace_map) { - my $rewrites = _replace_oids($self, $new_mime, $replace_map); + my $rewrites = _replace_oids($self, $new_eml, $replace_map); return { rewrites => $rewrites, need_reindex => $need_reindex }; } - defined($mark) ? @removed : undef; + defined($mark) ? $removed : undef; } # public (see PublicInbox::Import->remove), but note the 3rd element # (retval[2]) is not part of the stable API shared with Import->remove sub remove { - my ($self, $mime, $cmt_msg) = @_; - my @ret; - $self->{-inbox}->with_umask(sub { - @ret = rewrite_internal($self, $mime, $cmt_msg); - }); - defined($ret[0]) ? @ret : undef; + my ($self, $eml, $cmt_msg) = @_; + my $r = $self->{ibx}->with_umask(\&rewrite_internal, + $self, $eml, $cmt_msg); + defined($r) && defined($r->[0]) ? @$r: undef; } sub _replace ($$;$$) { - my ($self, $old_mime, $new_mime, $sref) = @_; - my $rewritten = $self->{-inbox}->with_umask(sub { - rewrite_internal($self, $old_mime, undef, $new_mime, $sref); - }) or return; + my ($self, $old_eml, $new_eml, $sref) = @_; + my $arg = [ $self, $old_eml, undef, $new_eml, $sref ]; + my $rewritten = $self->{ibx}->with_umask(\&rewrite_internal, + $self, $old_eml, undef, $new_eml, $sref) or return; my $rewrites = $rewritten->{rewrites}; # ->done is called if there are rewrites since we gc+prune from git @@ -492,18 +483,16 @@ sub purge { sub git_hash_raw ($$) { my ($self, $raw) = @_; # grab the expected OID we have to reindex: - open my $tmp_fh, '+>', undef or die "failed to open tmp: $!"; - $tmp_fh->autoflush(1); - print $tmp_fh $$raw or die "print \$tmp_fh: $!"; - sysseek($tmp_fh, 0, 0) or die "seek failed: $!"; - - my $git_dir = $self->{-inbox}->git->{git_dir}; + pipe(my($in, $w)) or die "pipe: $!"; + my $git_dir = $self->{ibx}->git->{git_dir}; my $cmd = ['git', "--git-dir=$git_dir", qw(hash-object --stdin)]; - my $r = popen_rd($cmd, undef, { 0 => $tmp_fh }); + my $r = popen_rd($cmd, undef, { 0 => $in }); + print $w $$raw or die "print \$w: $!"; + close $w or die "close \$w: $!"; local $/ = "\n"; chomp(my $oid = <$r>); close $r or die "git hash-object failed: $?"; - $oid =~ /\A[a-f0-9]{40}\z/ or die "OID not expected: $oid"; + $oid =~ /\A$OID\z/ or die "OID not expected: $oid"; $oid; } @@ -525,9 +514,7 @@ sub _check_mids_match ($$$) { # Message-IDs are pretty complex and rethreading hasn't been fully # implemented, yet. sub check_mids_match ($$) { - my ($old_mime, $new_mime) = @_; - my $old = $old_mime->header_obj; - my $new = $new_mime->header_obj; + my ($old, $new) = @_; _check_mids_match(mids($old), mids($new), 'Message-ID(s)'); _check_mids_match(references($old), references($new), 'References/In-Reply-To'); @@ -561,11 +548,11 @@ W: $list } # make sure we really got the OID: - my ($blob, $type, $bytes) = $self->{-inbox}->git->check($expect_oid); + my ($blob, $type, $bytes) = $self->{ibx}->git->check($expect_oid); $blob eq $expect_oid or die "BUG: $expect_oid not found after replace"; # don't leak FDs to Xapian: - $self->{-inbox}->git->cleanup; + $self->{ibx}->git->cleanup; # reindex modified messages: for my $smsg (@$need_reindex) { @@ -575,8 +562,8 @@ W: $list num => $smsg->{num}, mid => $smsg->{mid}, }, 'PublicInbox::Smsg'; - my $v2w = { autime => $smsg->{ds}, cotime => $smsg->{ts} }; - $new_smsg->populate($new_mime, $v2w); + my $sync = { autime => $smsg->{ds}, cotime => $smsg->{ts} }; + $new_smsg->populate($new_mime, $sync); do_idx($self, \$raw, $new_mime, $new_smsg); } $rewritten->{rewrites}; @@ -643,12 +630,12 @@ sub checkpoint ($;$) { my $barrier = $self->barrier_init(scalar @$shards); # each shard needs to issue a barrier command - $_->remote_barrier for @$shards; + $_->shard_barrier for @$shards; # wait for each Xapian shard $self->barrier_wait($barrier); } else { - $_->remote_commit for @$shards; + $_->shard_commit for @$shards; } # last_commit is special, don't commit these until @@ -668,35 +655,49 @@ sub checkpoint ($;$) { # public sub barrier { checkpoint($_[0], 1) }; +# true if locked and active +sub active { !!$_[0]->{im} } + # public sub done { my ($self) = @_; - my $im = delete $self->{im}; - $im->done if $im; # PublicInbox::Import::done - checkpoint($self); - my $mm = delete $self->{mm}; - $mm->{dbh}->commit if $mm; + my $err = ''; + if (my $im = delete $self->{im}) { + eval { $im->done }; # PublicInbox::Import::done + $err .= "import done: $@\n" if $@; + } + if (!$err) { + eval { checkpoint($self) }; + $err .= "checkpoint: $@\n" if $@; + } + if (my $mm = delete $self->{mm}) { + my $m = $err ? 'rollback' : 'commit'; + eval { $mm->{dbh}->$m }; + $err .= "msgmap $m: $@\n" if $@; + } my $shards = delete $self->{idx_shards}; if ($shards) { - $_->remote_close for @$shards; + for (@$shards) { + eval { $_->shard_close }; + $err .= "shard close: $@\n" if $@; + } } - $self->{over}->disconnect; + eval { $self->{over}->dbh_close }; + $err .= "over close: $@\n" if $@; delete $self->{bnote}; my $nbytes = $self->{total_bytes}; $self->{total_bytes} = 0; $self->lock_release(!!$nbytes) if $shards; - $self->{-inbox}->git->cleanup; + $self->{ibx}->git->cleanup; + die $err if $err; } sub fill_alternates ($$) { my ($self, $epoch) = @_; - my $pfx = "$self->{-inbox}->{inboxdir}/git"; - my $all = "$self->{-inbox}->{inboxdir}/all.git"; - - unless (-d $all) { - PublicInbox::Import::init_bare($all); - } + my $pfx = "$self->{ibx}->{inboxdir}/git"; + my $all = "$self->{ibx}->{inboxdir}/all.git"; + PublicInbox::Import::init_bare($all) unless -d $all; my $info_dir = "$all/objects/info"; my $alt = "$info_dir/alternates"; my (%alt, $new); @@ -706,7 +707,9 @@ sub fill_alternates ($$) { $mode = (stat($fh))[2] & 07777; # we assign a sort score to every alternate and favor - # the newest (highest numbered) one when we + # the newest (highest numbered) one because loose objects + # require scanning epochs and only the latest epoch is + # expected to see loose objects my $score; my $other = 0; # in case admin adds non-epoch repos %alt = map {; @@ -728,17 +731,19 @@ sub fill_alternates ($$) { } return unless $new; - my ($fh, $tmp) = tempfile('alt-XXXXXXXX', DIR => $info_dir); + my $fh = File::Temp->new(TEMPLATE => 'alt-XXXXXXXX', DIR => $info_dir); + my $tmp = $fh->filename; print $fh join("\n", sort { $alt{$b} <=> $alt{$a} } keys %alt), "\n" or die "print $tmp: $!\n"; chmod($mode, $fh) or die "fchmod $tmp: $!\n"; close $fh or die "close $tmp $!\n"; rename($tmp, $alt) or die "rename $tmp => $alt: $!\n"; + $fh->unlink_on_destroy(0); } sub git_init { my ($self, $epoch) = @_; - my $git_dir = "$self->{-inbox}->{inboxdir}/git/$epoch.git"; + my $git_dir = "$self->{ibx}->{inboxdir}/git/$epoch.git"; PublicInbox::Import::init_bare($git_dir); my @cmd = (qw/git config/, "--file=$git_dir/config", 'include.path', '../../all.git/config'); @@ -750,7 +755,7 @@ sub git_init { sub git_dir_latest { my ($self, $max) = @_; $$max = -1; - my $pfx = "$self->{-inbox}->{inboxdir}/git"; + my $pfx = "$self->{ibx}->{inboxdir}/git"; return unless -d $pfx; my $latest; opendir my $dh, $pfx or die "opendir $pfx: $!\n"; @@ -802,7 +807,7 @@ sub importer { sub import_init { my ($self, $git, $packed_bytes, $tmp) = @_; - my $im = PublicInbox::Import->new($git, undef, undef, $self->{-inbox}); + my $im = PublicInbox::Import->new($git, undef, undef, $self->{ibx}); $im->{bytes_added} = int($packed_bytes / $PACKING_FACTOR); $im->{lock_path} = undef; $im->{path_type} = 'v2'; @@ -814,18 +819,17 @@ sub import_init { sub diff ($$$) { my ($mid, $cur, $new) = @_; - my ($ah, $an) = tempfile('email-cur-XXXXXXXX', TMPDIR => 1); + my $ah = File::Temp->new(TEMPLATE => 'email-cur-XXXXXXXX', TMPDIR => 1); print $ah $cur->as_string or die "print: $!"; - close $ah or die "close: $!"; - my ($bh, $bn) = tempfile('email-new-XXXXXXXX', TMPDIR => 1); + $ah->flush or die "flush: $!"; PublicInbox::Import::drop_unwanted_headers($new); + my $bh = File::Temp->new(TEMPLATE => 'email-new-XXXXXXXX', TMPDIR => 1); print $bh $new->as_string or die "print: $!"; - close $bh or die "close: $!"; - my $cmd = [ qw(diff -u), $an, $bn ]; + $bh->flush or die "flush: $!"; + my $cmd = [ qw(diff -u), $ah->filename, $bh->filename ]; print STDERR "# MID conflict <$mid>\n"; my $pid = spawn($cmd, undef, { 1 => 2 }); waitpid($pid, 0) == $pid or die "diff did not finish"; - unlink($an, $bn); } sub get_blob ($$) { @@ -835,8 +839,7 @@ sub get_blob ($$) { return $msg if $msg; } # older message, should be in alternates - my $ibx = $self->{-inbox}; - $ibx->msg_by_smsg($smsg); + $self->{ibx}->msg_by_smsg($smsg); } sub content_exists ($$$) { @@ -861,8 +864,6 @@ sub content_exists ($$$) { sub atfork_child { my ($self) = @_; - my $fh = delete $self->{reindex_pipe}; - close $fh if $fh; if (my $shards = $self->{idx_shards}) { $_->atfork_child foreach @$shards; } @@ -874,199 +875,97 @@ sub atfork_child { $self->{bnote}->[1]; } -sub mark_deleted ($$$$) { - my ($self, $sync, $git, $oid) = @_; - return if PublicInbox::SearchIdx::too_big($self, $git, $oid); - my $msgref = $git->cat_file($oid); - my $mime = PublicInbox::Eml->new($$msgref); - my $mids = mids($mime->header_obj); - my $chash = content_hash($mime); - foreach my $mid (@$mids) { - $sync->{D}->{"$mid\0$chash"} = $oid; - } -} - -sub reindex_checkpoint ($$$) { - my ($self, $sync, $git) = @_; +sub reindex_checkpoint ($$) { + my ($self, $sync) = @_; - $git->cleanup; - $sync->{mm_tmp}->atfork_prepare; + $self->{ibx}->git->cleanup; # *async_wait + ${$sync->{need_checkpoint}} = 0; + my $mm_tmp = $sync->{mm_tmp}; + $mm_tmp->atfork_prepare if $mm_tmp; $self->done; # release lock if (my $pr = $sync->{-opt}->{-progress}) { - my ($bn) = (split('/', $git->{git_dir}))[-1]; - $pr->("$bn ".sprintf($sync->{-regen_fmt}, $sync->{nr})); + $pr->(sprintf($sync->{-regen_fmt}, ${$sync->{nr}})); } # allow -watch or -mda to write... - $self->idx_init; # reacquire lock - $sync->{mm_tmp}->atfork_parent; + $self->idx_init($sync->{-opt}); # reacquire lock + $mm_tmp->atfork_parent if $mm_tmp; } -# only for a few odd messages with multiple Message-IDs -sub reindex_oid_m ($$$$;$) { - my ($self, $sync, $git, $oid, $regen_num) = @_; - $self->{current_info} = "multi_mid $oid"; - my ($num, $mid0, $len); - my $msgref = $git->cat_file($oid, \$len); - my $mime = PublicInbox::Eml->new($$msgref); - my $mids = mids($mime->header_obj); - my $chash = content_hash($mime); - die "BUG: reindex_oid_m called for <=1 mids" if scalar(@$mids) <= 1; - - for my $mid (reverse @$mids) { - delete($sync->{D}->{"$mid\0$chash"}) and - die "BUG: reindex_oid should handle <$mid> delete"; +sub index_oid { # cat_async callback + my ($bref, $oid, $type, $size, $arg) = @_; + return if $size == 0; # purged + my ($num, $mid0); + my $eml = PublicInbox::Eml->new($$bref); + my $mids = mids($eml); + my $chash = content_hash($eml); + my $self = $arg->{v2w}; + + if (scalar(@$mids) == 0) { + warn "E: $oid has no Message-ID, skipping\n"; + return; } - my $over = $self->{over}; - for my $mid (reverse @$mids) { - ($num, $mid0) = $over->num_mid0_for_oid($oid, $mid); - next unless defined $num; - if (defined($regen_num) && $regen_num != $num) { - die "BUG: regen(#$regen_num) != over(#$num)"; + + # {unindexed} is unlikely + if ((my $unindexed = $arg->{unindexed}) && scalar(@$mids) == 1) { + $num = delete($unindexed->{$mids->[0]}); + if (defined $num) { + $mid0 = $mids->[0]; + $self->{mm}->mid_set($num, $mid0); + delete($arg->{unindexed}) if !keys(%$unindexed); } } - unless (defined($num)) { - for my $mid (reverse @$mids) { - # is this a number we got before? - my $n = $sync->{mm_tmp}->num_for($mid); - next unless defined $n; - next if defined($regen_num) && $regen_num != $n; - ($num, $mid0) = ($n, $mid); - last; + if (!defined($num)) { # reuse if reindexing (or duplicates) + my $over = $self->{over}; + for my $mid (@$mids) { + ($num, $mid0) = $over->num_mid0_for_oid($oid, $mid); + last if defined $num; } } - if (defined($num)) { - $sync->{mm_tmp}->num_delete($num); - } elsif (defined $regen_num) { - $num = $regen_num; - for my $mid (reverse @$mids) { - $self->{mm}->mid_set($num, $mid) == 1 or next; - $mid0 = $mid; - last; - } - unless (defined $mid0) { - warn "E: cannot regen #$num\n"; - return; - } - } else { # fixup bugs in old mirrors on reindex - for my $mid (reverse @$mids) { - $num = $self->{mm}->mid_insert($mid); - next unless defined $num; - $mid0 = $mid; - last; - } - if (defined $mid0) { - if ($sync->{reindex}) { - warn "reindex added #$num <$mid0>\n"; + $mid0 //= do { # is this a number we got before? + $num = $arg->{mm_tmp}->num_for($mids->[0]); + defined($num) ? $mids->[0] : undef; + }; + if (!defined($num)) { + for (my $i = $#$mids; $i >= 1; $i--) { + $num = $arg->{mm_tmp}->num_for($mids->[$i]); + if (defined($num)) { + $mid0 = $mids->[$i]; + last; } - } else { - warn "E: cannot find article #\n"; - return; } } - $sync->{nr}++; - my $smsg = bless { - raw_bytes => $len, - num => $num, - blob => $oid, - mid => $mid0, - }, 'PublicInbox::Smsg'; - $smsg->populate($mime, $self); - if (do_idx($self, $msgref, $mime, $smsg)) { - reindex_checkpoint($self, $sync, $git); - } -} - -sub check_unindexed ($$$) { - my ($self, $num, $mid0) = @_; - my $unindexed = $self->{unindexed} // {}; - my $n = delete($unindexed->{$mid0}); - defined $n or return; - if ($n != $num) { - die "BUG: unindexed $n != $num <$mid0>\n"; - } else { - $self->{mm}->mid_set($num, $mid0); - } -} - -sub multi_mid_q_push ($$$) { - my ($self, $sync, $oid) = @_; - my $multi_mid = $sync->{multi_mid} //= PublicInbox::MultiMidQueue->new; - if ($sync->{reindex}) { # no regen on reindex - $multi_mid->push_oid($oid, $self); - } else { - my $num = $sync->{regen}--; - die "BUG: ran out of article numbers" if $num <= 0; - $multi_mid->set_oid($num, $oid, $self); - } -} - -sub reindex_oid ($$$$) { - my ($self, $sync, $git, $oid) = @_; - return if PublicInbox::SearchIdx::too_big($self, $git, $oid); - my ($num, $mid0, $len); - my $msgref = $git->cat_file($oid, \$len); - return if $len == 0; # purged - my $mime = PublicInbox::Eml->new($$msgref); - my $mids = mids($mime->header_obj); - my $chash = content_hash($mime); - - if (scalar(@$mids) == 0) { - warn "E: $oid has no Message-ID, skipping\n"; - return; - } elsif (scalar(@$mids) == 1) { - my $mid = $mids->[0]; - - # was the file previously marked as deleted?, skip if so - if (delete($sync->{D}->{"$mid\0$chash"})) { - if (!$sync->{reindex}) { - $num = $sync->{regen}--; - $self->{mm}->num_highwater($num); - } - return; - } - - # is this a number we got before? - $num = $sync->{mm_tmp}->num_for($mid); - if (defined $num) { - $mid0 = $mid; - check_unindexed($self, $num, $mid0); - } else { - $num = $sync->{regen}--; - die "BUG: ran out of article numbers" if $num <= 0; - if ($self->{mm}->mid_set($num, $mid) != 1) { - warn "E: unable to assign $num => <$mid>\n"; - return; + if (defined($num)) { + $arg->{mm_tmp}->num_delete($num); + } else { # never seen + $num = $self->{mm}->mid_insert($mids->[0]); + if (defined($num)) { + $mid0 = $mids->[0]; + } else { # rare, try the rest of them, backwards + for (my $i = $#$mids; $i >= 1; $i--) { + $num = $self->{mm}->mid_insert($mids->[$i]); + if (defined($num)) { + $mid0 = $mids->[$i]; + last; + } } - $mid0 = $mid; - } - } else { # multiple MIDs are a weird case: - my $del = 0; - for (@$mids) { - $del += delete($sync->{D}->{"$_\0$chash"}) // 0; - } - if ($del) { - unindex_oid_remote($self, $oid, $_) for @$mids; - # do not delete from {mm_tmp}, since another - # single-MID message may use it. - } else { # handle them at the end: - multi_mid_q_push($self, $sync, $oid); } + } + if (!defined($num)) { + warn "E: $oid <", join('> <', @$mids), "> is a duplicate\n"; return; } - $sync->{mm_tmp}->mid_delete($mid0) or - die "failed to delete <$mid0> for article #$num\n"; - $sync->{nr}++; + ++${$arg->{nr}}; my $smsg = bless { - raw_bytes => $len, + raw_bytes => $size, num => $num, blob => $oid, mid => $mid0, }, 'PublicInbox::Smsg'; - $smsg->populate($mime, $self); - if (do_idx($self, $msgref, $mime, $smsg)) { - reindex_checkpoint($self, $sync, $git); + $smsg->populate($eml, $arg); + if (do_idx($self, $bref, $eml, $smsg)) { + ${$arg->{need_checkpoint}} = 1; } } @@ -1082,7 +981,7 @@ sub update_last_commit ($$$$) { last_epoch_commit($self, $i, $cmt); } -sub git_dir_n ($$) { "$_[0]->{-inbox}->{inboxdir}/git/$_[1].git" } +sub git_dir_n ($$) { "$_[0]->{ibx}->{inboxdir}/git/$_[1].git" } sub last_commits ($$) { my ($self, $epoch_max) = @_; @@ -1093,15 +992,13 @@ sub last_commits ($$) { $heads; } -*is_ancestor = *PublicInbox::SearchIdx::is_ancestor; - # returns a revision range for git-log(1) sub log_range ($$$$$) { my ($self, $sync, $git, $i, $tip) = @_; my $opt = $sync->{-opt}; my $pr = $opt->{-progress} if (($opt->{verbose} || 0) > 1); my $cur = $sync->{ranges}->[$i] or do { - $pr->("$i.git indexing all of $tip") if $pr; + $pr->("$i.git indexing all of $tip\n") if $pr; return $tip; # all of it }; @@ -1150,14 +1047,13 @@ sub sync_prepare ($$$) { my ($self, $sync, $epoch_max) = @_; my $pr = $sync->{-opt}->{-progress}; my $regen_max = 0; - my $head = $self->{-inbox}->{ref_head} || 'refs/heads/master'; + my $head = $self->{ibx}->{ref_head} || 'refs/heads/master'; # reindex stops at the current heads and we later rerun index_sync # without {reindex} my $reindex_heads = last_commits($self, $epoch_max) if $sync->{reindex}; for (my $i = $epoch_max; $i >= 0; $i--) { - die 'BUG: already indexing!' if $self->{reindex_pipe}; my $git_dir = git_dir_n($self, $i); -d $git_dir or next; # missing epochs are fine my $git = PublicInbox::Git->new($git_dir); @@ -1168,54 +1064,71 @@ sub sync_prepare ($$$) { next if $?; # new repo my $range = log_range($self, $sync, $git, $i, $tip) or next; - $sync->{ranges}->[$i] = $range; - # can't use 'rev-list --count' if we use --diff-filter $pr->("$i.git counting $range ... ") if $pr; - my $n = 0; - my $fh = $git->popen(qw(log --pretty=tformat:%H - --no-notes --no-color --no-renames - --diff-filter=AM), $range, '--', 'm'); - ++$n while <$fh>; - close $fh or die "git log failed: \$?=$?"; - $pr->("$n\n") if $pr; - $regen_max += $n; + # Don't bump num_highwater on --reindex by using {D}. + # We intentionally do NOT use {D} in the non-reindex case + # because we want NNTP article number gaps from unindexed + # messages to show up in mirrors, too. + $sync->{D} //= $sync->{reindex} ? {} : undef; # OID_BIN => NR + my $stk = log2stack($sync, $git, $range, $self->{ibx}); + my $nr = $stk ? $stk->num_records : 0; + $pr->("$nr\n") if $pr; + $sync->{stacks}->[$i] = $stk if $stk; + $regen_max += $nr; + } + + # XXX this should not happen unless somebody bypasses checks in + # our code and blindly injects "d" file history into git repos + if (my @leftovers = keys %{delete($sync->{D}) // {}}) { + warn('W: unindexing '.scalar(@leftovers)." leftovers\n"); + my $arg = { v2w => $self }; + my $all = $self->{ibx}->git; + for my $oid (@leftovers) { + $oid = unpack('H*', $oid); + $self->{current_info} = "leftover $oid"; + $all->cat_async($oid, \&unindex_oid, $arg); + } + $all->cat_async_wait; + } + if (!$regen_max && !keys(%{$self->{unindex_range}})) { + $sync->{-regen_fmt} = "%u/?\n"; + return 0; } - - return 0 if (!$regen_max && !keys(%{$self->{unindex_range}})); # reindex should NOT see new commits anymore, if we do, # it's a problem and we need to notice it via die() my $pad = length($regen_max) + 1; $sync->{-regen_fmt} = "% ${pad}u/$regen_max\n"; - $sync->{nr} = 0; + $sync->{nr} = \(my $nr = 0); return -1 if $sync->{reindex}; $regen_max + $self->{mm}->num_highwater() || 0; } sub unindex_oid_remote ($$$) { my ($self, $oid, $mid) = @_; - $_->remote_remove($oid, $mid) foreach @{$self->{idx_shards}}; - $self->{over}->remove_oid($oid, $mid); + my @removed = $self->{over}->remove_oid($oid, $mid); + for my $num (@removed) { + my $idx = idx_shard($self, $num % $self->{shards}); + $idx->shard_remove($oid, $num); + } } -sub unindex_oid ($$$;$) { - my ($self, $git, $oid, $unindexed) = @_; +sub unindex_oid ($$;$) { # git->cat_async callback + my ($bref, $oid, $type, $size, $sync) = @_; + my $self = $sync->{v2w}; + my $unindexed = $sync->{in_unindex} ? $sync->{unindexed} : undef; my $mm = $self->{mm}; - my $msgref = $git->cat_file($oid); - my $mime = PublicInbox::Eml->new($msgref); - my $mids = mids($mime->header_obj); - $mime = $msgref = undef; + my $mids = mids(PublicInbox::Eml->new($bref)); + undef $$bref; my $over = $self->{over}; foreach my $mid (@$mids) { my %gone; my ($id, $prev); while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) { $gone{$smsg->{num}} = 1 if $oid eq $smsg->{blob}; - 1; # continue } - my $n = scalar keys %gone; - next unless $n; + my $n = scalar(keys(%gone)) or next; if ($n > 1) { warn "BUG: multiple articles linked to $oid\n", join(',',sort keys %gone), "\n"; @@ -1231,21 +1144,24 @@ sub unindex_oid ($$$;$) { } } -my $x40 = qr/[a-f0-9]{40}/; +# this is rare, it only happens when we get discontiguous history in +# a mirror because the source used -purge or -edit sub unindex ($$$$) { my ($self, $sync, $git, $unindex_range) = @_; - my $unindexed = $self->{unindexed} ||= {}; # $mid0 => $num + my $unindexed = $sync->{unindexed} //= {}; # $mid0 => $num my $before = scalar keys %$unindexed; # order does not matter, here: my @cmd = qw(log --raw -r --no-notes --no-color --no-abbrev --no-renames); - my $fh = $self->{reindex_pipe} = $git->popen(@cmd, $unindex_range); + my $fh = $git->popen(@cmd, $unindex_range); + my $all = $self->{ibx}->git; + local $sync->{in_unindex} = 1; while (<$fh>) { - /\A:\d{6} 100644 $x40 ($x40) [AM]\tm$/o or next; - unindex_oid($self, $git, $1, $unindexed); + /\A:\d{6} 100644 $OID ($OID) [AM]\tm$/o or next; + $all->cat_async($1, \&unindex_oid, $sync); } - delete $self->{reindex_pipe}; close $fh or die "git log failed: \$?=$?"; + $all->cat_async_wait; return unless $sync->{-opt}->{prune}; my $after = scalar keys %$unindexed; @@ -1270,121 +1186,180 @@ sub sync_ranges ($$$) { $ranges; } +sub index_xap_only { # git->cat_async callback + my ($bref, $oid, $type, $size, $smsg) = @_; + my $self = $smsg->{v2w}; + my $idx = idx_shard($self, $smsg->{num} % $self->{shards}); + $smsg->{raw_bytes} = $size; + $idx->index_raw($bref, undef, $smsg); + $self->{transact_bytes} += $size; +} + +sub index_xap_step ($$$;$) { + my ($self, $sync, $beg, $step) = @_; + my $end = $sync->{art_end}; + return if $beg > $end; # nothing to do + + $step //= $self->{shards}; + my $ibx = $self->{ibx}; + if (my $pr = $sync->{-opt}->{-progress}) { + $pr->("Xapian indexlevel=$ibx->{indexlevel} ". + "$beg..$end (% $step)\n"); + } + for (my $num = $beg; $num <= $end; $num += $step) { + my $smsg = $ibx->over->get_art($num) or next; + $smsg->{v2w} = $self; + $ibx->git->cat_async($smsg->{blob}, \&index_xap_only, $smsg); + if ($self->{transact_bytes} >= $self->{batch_bytes}) { + ${$sync->{nr}} = $num; + reindex_checkpoint($self, $sync); + } + } +} + sub index_epoch ($$$) { my ($self, $sync, $i) = @_; my $git_dir = git_dir_n($self, $i); - die 'BUG: already reindexing!' if $self->{reindex_pipe}; -d $git_dir or return; # missing epochs are fine - fill_alternates($self, $i); my $git = PublicInbox::Git->new($git_dir); - if (my $unindex_range = delete $sync->{unindex_range}->{$i}) { + if (my $unindex_range = delete $sync->{unindex_range}->{$i}) { # rare unindex($self, $sync, $git, $unindex_range); } - defined(my $range = $sync->{ranges}->[$i]) or return; - if (my $pr = $sync->{-opt}->{-progress}) { - $pr->("$i.git indexing $range\n"); + defined(my $stk = $sync->{stacks}->[$i]) or return; + $sync->{stacks}->[$i] = undef; + my $all = $self->{ibx}->git; + while (my ($f, $at, $ct, $oid) = $stk->pop_rec) { + $self->{current_info} = "$i.git $oid"; + if ($f eq 'm') { + my $arg = { %$sync, autime => $at, cotime => $ct }; + if ($sync->{max_size}) { + $all->check_async($oid, \&check_size, $arg); + } else { + $all->cat_async($oid, \&index_oid, $arg); + } + } elsif ($f eq 'd') { + $all->cat_async($oid, \&unindex_oid, $sync); + } + if (${$sync->{need_checkpoint}}) { + reindex_checkpoint($self, $sync); + } } + $all->check_async_wait; + $all->cat_async_wait; + update_last_commit($self, $git, $i, $stk->{latest_cmt}); +} - my @cmd = qw(log --raw -r --pretty=tformat:%H.%at.%ct - --no-notes --no-color --no-abbrev --no-renames); - my $fh = $self->{reindex_pipe} = $git->popen(@cmd, $range); - my $cmt; - while (<$fh>) { - chomp; - $self->{current_info} = "$i.git $_"; - if (/\A($x40)\.([0-9]+)\.([0-9]+)$/o) { - $cmt //= $1; - $self->{autime} = $2; - $self->{cotime} = $3; - } elsif (/\A:\d{6} 100644 $x40 ($x40) [AM]\tm$/o) { - reindex_oid($self, $sync, $git, $1); - } elsif (/\A:\d{6} 100644 $x40 ($x40) [AM]\td$/o) { - mark_deleted($self, $sync, $git, $1); +sub xapian_only { + my ($self, $opt, $sync, $art_beg) = @_; + my $seq = $opt->{sequential_shard}; + $art_beg //= 0; + local $self->{parallel} = 0 if $seq; + $self->idx_init($opt); # acquire lock + if (my $art_end = $self->{ibx}->mm->max) { + $sync //= { + need_checkpoint => \(my $bool = 0), + -opt => $opt, + v2w => $self, + nr => \(my $nr = 0), + -regen_fmt => "%u/?\n", + }; + $sync->{art_end} = $art_end; + if ($seq || !$self->{parallel}) { + my $shard_end = $self->{shards} - 1; + for my $i (0..$shard_end) { + index_xap_step($self, $sync, $art_beg + $i); + if ($i != $shard_end) { + reindex_checkpoint($self, $sync); + } + } + } else { # parallel (maybe) + index_xap_step($self, $sync, $art_beg, 1); } } - close $fh or die "git log failed: \$?=$?"; - delete @$self{qw(reindex_pipe autime cotime)}; - update_last_commit($self, $git, $i, $cmt) if defined $cmt; + $self->{ibx}->git->cat_async_wait; + $self->done; } # public, called by public-inbox-index sub index_sync { my ($self, $opt) = @_; - $opt ||= {}; + $opt //= $_[1] //= {}; + goto \&xapian_only if $opt->{xapian_only}; + my $pr = $opt->{-progress}; my $epoch_max; my $latest = git_dir_latest($self, \$epoch_max); return unless defined $latest; + + my $seq = $opt->{sequential_shard}; + my $art_beg; # the NNTP article number we start xapian_only at + my $idxlevel = $self->{ibx}->{indexlevel}; + local $self->{ibx}->{indexlevel} = 'basic' if $seq; + $self->idx_init($opt); # acquire lock + fill_alternates($self, $epoch_max); + $self->{over}->rethread_prepare($opt); my $sync = { - D => {}, # "$mid\0$chash" => $oid + need_checkpoint => \(my $bool = 0), unindex_range => {}, # EPOCH => oid_old..oid_new reindex => $opt->{reindex}, - -opt => $opt + -opt => $opt, + v2w => $self, }; $sync->{ranges} = sync_ranges($self, $sync, $epoch_max); - $sync->{regen} = sync_prepare($self, $sync, $epoch_max); - - if ($sync->{regen}) { + if (sync_prepare($self, $sync, $epoch_max)) { # tmp_clone seems to fail if inside a transaction, so # we rollback here (because we opened {mm} for reading) # Note: we do NOT rely on DBI transactions for atomicity; # only for batch performance. $self->{mm}->{dbh}->rollback; $self->{mm}->{dbh}->begin_work; - $sync->{mm_tmp} = $self->{mm}->tmp_clone; - } - - # work backwards through history - for (my $i = $epoch_max; $i >= 0; $i--) { - index_epoch($self, $sync, $i); - } + $sync->{mm_tmp} = + $self->{mm}->tmp_clone($self->{ibx}->{inboxdir}); - # unindex is required for leftovers if "deletes" affect messages - # in a previous fetch+index window: - my $git; - if (my @leftovers = values %{delete $sync->{D}}) { - $git = $self->{-inbox}->git; - for my $oid (@leftovers) { - $self->{current_info} = "leftover $oid"; - unindex_oid($self, $git, $oid); + # xapian_only works incrementally w/o --reindex + if ($seq && !$opt->{reindex}) { + $art_beg = $sync->{mm_tmp}->max; + $art_beg++ if defined($art_beg); } } - if (my $multi_mid = delete $sync->{multi_mid}) { - $git //= $self->{-inbox}->git; - my $min = $multi_mid->{min}; - my $max = $multi_mid->{max}; - if ($sync->{reindex}) { - # we may need to create new Message-IDs if mirrors - # were initially indexed with old versions - for (my $i = $max; $i >= $min; $i--) { - my $oid; - $oid = $multi_mid->get_oid($i, $self) or next; - next unless defined $oid; - reindex_oid_m($self, $sync, $git, $oid); - } - } else { # regen on initial index - for my $num ($min..$max) { - my $oid; - $oid = $multi_mid->get_oid($num, $self) or next; - reindex_oid_m($self, $sync, $git, $oid, $num); - } - } + if ($sync->{max_size} = $opt->{max_size}) { + $sync->{index_oid} = \&index_oid; } - $git->cleanup if $git; + # work forwards through history + index_epoch($self, $sync, $_) for (0..$epoch_max); + $self->{over}->rethread_done($opt); $self->done; if (my $nr = $sync->{nr}) { my $pr = $sync->{-opt}->{-progress}; - $pr->('all.git '.sprintf($sync->{-regen_fmt}, $nr)) if $pr; + $pr->('all.git '.sprintf($sync->{-regen_fmt}, $$nr)) if $pr; + } + + # deal with Xapian shards sequentially + if ($seq && delete($sync->{mm_tmp})) { + $self->{ibx}->{indexlevel} = $idxlevel; + xapian_only($self, $opt, $sync, $art_beg); + } + + # --reindex on the command-line + if ($opt->{reindex} && !ref($opt->{reindex}) && $idxlevel ne 'basic') { + $self->lock_acquire; + my $s0 = PublicInbox::SearchIdx->new($self->{ibx}, 0, 0); + if (my $xdb = $s0->idx_acquire) { + my $n = $xdb->get_metadata('has_threadid'); + $xdb->set_metadata('has_threadid', '1') if $n ne '1'; + } + $s0->idx_release; + $self->lock_release; } # reindex does not pick up new changes, so we rerun w/o it: if ($opt->{reindex}) { my %again = %$opt; $sync = undef; - delete @again{qw(reindex -skip_lock)}; + delete @again{qw(rethread reindex -skip_lock)}; index_sync($self, \%again); } }