use PublicInbox::MID qw(mids);
use PublicInbox::ContentId qw(content_id content_digest);
use PublicInbox::Inbox;
-use PublicInbox::OverIdxFork;
+use PublicInbox::OverIdx;
use PublicInbox::Msgmap;
use PublicInbox::Spawn;
use IO::Handle;
parallel => 1,
transact_bytes => 0,
xpfx => $xpfx,
- over => PublicInbox::OverIdxFork->new("$xpfx/over.sqlite3"),
+ over => PublicInbox::OverIdx->new("$xpfx/over.sqlite3", 1),
lock_path => "$dir/inbox.lock",
# limit each git repo (epoch) to 1GB or so
rotate_bytes => int((1024 * 1024 * 1024) / $PACKING_FACTOR),
my $im = $self->importer;
my $cmt = $im->add($mime);
$cmt = $im->get_mark($cmt);
- my ($oid, $len, $msgref) = @{$im->{last_object}};
+ $self->{last_commit}->[$self->{epoch_max}] = $cmt;
+ my ($oid, $len, $msgref) = @{$im->{last_object}};
+ $self->{over}->add_overview($mime, $len, $num, $oid, $mid0);
my $nparts = $self->{partitions};
my $part = $num % $nparts;
- $self->{last_commit}->[$self->{epoch_max}] = $cmt;
my $idx = $self->idx_part($part);
$idx->index_raw($len, $msgref, $num, $oid, $mid0, $mime);
my $n = $self->{transact_bytes} += $len;
};
# crap, Message-ID is already known, hope somebody just resent:
- $self->barrier;
foreach my $m (@$mids) {
# read-only lookup now safe to do after above barrier
my $existing = $self->lookup_content($mime, $m);
# frequently activated.
delete $ibx->{$_} foreach (qw(git mm search));
+ if ($self->{parallel}) {
+ pipe(my ($r, $w)) or die "pipe failed: $!";
+ $self->{bnote} = [ $r, $w ];
+ $w->autoflush(1);
+ }
+
my $over = $self->{over};
$ibx->umask_prepare;
$ibx->with_umask(sub {
$self->lock_acquire;
- $over->create($self);
+ $over->create;
# -compact can change partition count while -watch is idle
my $nparts = count_partitions($self);
sub remove_internal {
my ($self, $mime, $cmt_msg, $purge) = @_;
- $self->barrier;
$self->idx_init;
my $im = $self->importer unless $purge;
- my $ibx = $self->{-inbox};
- my $srch = $ibx->search;
+ my $over = $self->{over};
my $cid = content_id($mime);
my $parts = $self->{idx_parts};
my $mm = $self->{mm};
foreach my $mid (@$mids) {
my %gone;
- $srch->reopen->each_smsg_by_mid($mid, sub {
- my ($smsg) = @_;
- $smsg->load_expand;
- my $msg = $ibx->msg_by_smsg($smsg);
+ my ($id, $prev);
+ while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
+ my $msg = get_blob($self, $smsg);
if (!defined($msg)) {
warn "broken smsg for $mid\n";
- return 1; # continue
+ next; # continue
}
my $orig = $$msg;
my $cur = PublicInbox::MIME->new($msg);
if (content_id($cur) eq $cid) {
$smsg->{mime} = $cur;
- $gone{$smsg->num} = [ $smsg, \$orig ];
+ $gone{$smsg->{num}} = [ $smsg, \$orig ];
}
- 1; # continue
- });
+ }
my $n = scalar keys %gone;
next unless $n;
if ($n > 1) {
($mark, undef) = $im->remove($orig, $cmt_msg);
}
$orig = undef;
- foreach my $idx (@$parts) {
- $idx->remote_remove($oid, $mid);
- }
+ $self->unindex_oid_remote($oid, $mid);
}
- $self->barrier;
}
if (defined $mark) {
}
}
-sub done {
- my ($self) = @_;
- my $im = delete $self->{im};
- $im->done if $im; # PublicInbox::Import::done
-
- my $mm = $self->{mm};
- $mm->{dbh}->commit if $mm;
-
- # order matters, we can only close {over} after all partitions
- # are done because the partitions also write to {over}
- my $parts = delete $self->{idx_parts};
- if ($parts) {
- $_->remote_commit for @$parts;
- $_->remote_close for @$parts;
- }
-
- my $over = $self->{over};
- $over->remote_commit;
- $over->remote_close;
-
- if ($mm) {
- $mm->{dbh}->begin_work;
- set_last_commits($self);
- $mm->{dbh}->commit;
- delete $self->{mm};
- }
-
- $self->{transact_bytes} = 0;
- $self->lock_release if $parts;
+sub barrier_init {
+ my ($self, $n) = @_;
+ $self->{bnote} or return;
+ --$n;
+ my $barrier = { map { $_ => 1 } (0..$n) };
}
-sub checkpoint {
- my ($self) = @_;
- my $im = $self->{im};
- $im->checkpoint if $im; # PublicInbox::Import::checkpoint
- $self->barrier(1);
+sub barrier_wait {
+ my ($self, $barrier) = @_;
+ my $bnote = $self->{bnote} or return;
+ my $r = $bnote->[0];
+ while (scalar keys %$barrier) {
+ defined(my $l = $r->getline) or die "EOF on barrier_wait: $!";
+ $l =~ /\Abarrier (\d+)/ or die "bad line on barrier_wait: $l";
+ delete $barrier->{$1} or die "bad part[$1] on barrier wait";
+ }
}
-# issue a write barrier to ensure all data is visible to other processes
-# and read-only ops. Order of data importance is: git > SQLite > Xapian
-sub barrier {
- my ($self, $fsync) = @_;
+sub checkpoint ($;$) {
+ my ($self, $wait) = @_;
if (my $im = $self->{im}) {
- $im->barrier;
+ if ($wait) {
+ $im->barrier;
+ } else {
+ $im->checkpoint;
+ }
}
my $parts = $self->{idx_parts};
if ($parts) {
my $dbh = $self->{mm}->{dbh};
- $dbh->commit; # SQLite msgmap data is second in importance
- my $over = $self->{over};
+ # SQLite msgmap data is second in importance
+ $dbh->commit;
- # Now deal with Xapian and overview DB
- $over->barrier_init(scalar(@$parts));
+ # SQLite overview is third
+ $self->{over}->commit_lazy;
- # each partition needs to issue a barrier command to over
- $_->remote_barrier foreach @$parts;
+ # Now deal with Xapian
+ if ($wait) {
+ my $barrier = $self->barrier_init(scalar @$parts);
- $over->barrier_wait; # wait for each Xapian partition
- $over->commit_fsync if $fsync;
+ # each partition needs to issue a barrier command
+ $_->remote_barrier for @$parts;
+
+ # wait for each Xapian partition
+ $self->barrier_wait($barrier);
+ } else {
+ $_->remote_commit for @$parts;
+ }
# last_commit is special, don't commit these until
# remote partitions are done:
$self->{transact_bytes} = 0;
}
+# issue a write barrier to ensure all data is visible to other processes
+# and read-only ops. Order of data importance is: git > SQLite > Xapian
+sub barrier { checkpoint($_[0], 1) };
+
+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 $parts = delete $self->{idx_parts};
+ if ($parts) {
+ $_->remote_close for @$parts;
+ }
+ $self->{over}->disconnect;
+ delete $self->{bnote};
+ $self->{transact_bytes} = 0;
+ $self->lock_release if $parts;
+}
+
sub git_init {
my ($self, $epoch) = @_;
my $pfx = "$self->{-inbox}->{mainrepo}/git";
} else {
$self->{im} = undef;
$im->done;
- $self->barrier(1);
$im = undef;
+ $self->checkpoint;
my $git_dir = $self->git_init(++$self->{epoch_max});
my $git = PublicInbox::Git->new($git_dir);
return $self->import_init($git, 0);
unlink($an, $bn);
}
-sub lookup_content {
- my ($self, $mime, $mid) = @_;
+sub get_blob ($$) {
+ my ($self, $smsg) = @_;
+ if (my $im = $self->{im}) {
+ my $msg = $im->cat_blob($smsg->{blob});
+ return $msg if $msg;
+ }
+ # older message, should be in alternates
my $ibx = $self->{-inbox};
+ $ibx->msg_by_smsg($smsg);
+}
- my $srch = $ibx->search->reopen;
+sub lookup_content {
+ my ($self, $mime, $mid) = @_;
+ my $over = $self->{over};
my $cid = content_id($mime);
my $found;
- $srch->each_smsg_by_mid($mid, sub {
- my ($smsg) = @_;
- $smsg->load_expand;
- my $msg = $ibx->msg_by_smsg($smsg);
+ my ($id, $prev);
+ while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
+ my $msg = get_blob($self, $smsg);
if (!defined($msg)) {
warn "broken smsg for $mid\n";
- return 1; # continue
+ next;
}
my $cur = PublicInbox::MIME->new($msg);
if (content_id($cur) eq $cid) {
$smsg->{mime} = $cur;
$found = $smsg;
- return 0; # break out of loop
+ last;
}
# XXX DEBUG_DIFF is experimental and may be removed
diff($mid, $cur, $mime) if $ENV{DEBUG_DIFF};
-
- 1; # continue
- });
+ }
$found;
}
$im->atfork_child;
}
die "unexpected mm" if $self->{mm};
+ close $self->{bnote}->[0] or die "close bnote[0]: $!\n";
+ $self->{bnote}->[1];
}
sub mark_deleted {
$mm_tmp->mid_delete($mid0) or
die "failed to delete <$mid0> for article #$num\n";
+ $self->{over}->add_overview($mime, $len, $num, $oid, $mid0);
my $nparts = $self->{partitions};
my $part = $num % $nparts;
my $idx = $self->idx_part($part);
\$regen_max;
}
+sub unindex_oid_remote {
+ my ($self, $oid, $mid) = @_;
+ $_->remote_remove($oid, $mid) foreach @{$self->{idx_parts}};
+ $self->{over}->remove_oid($oid, $mid);
+}
+
sub unindex_oid {
my ($self, $git, $oid) = @_;
my $msgref = $git->cat_file($oid);
my $mime = PublicInbox::MIME->new($msgref);
my $mids = mids($mime->header_obj);
$mime = $msgref = undef;
-
+ my $over = $self->{over};
foreach my $mid (@$mids) {
my %gone;
- $self->{-inbox}->search->reopen->each_smsg_by_mid($mid, sub {
- my ($smsg) = @_;
- $smsg->load_expand;
- $gone{$smsg->num} = 1 if $oid eq $smsg->{blob};
+ my ($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 $n;
if ($n > 1) {
join(',',sort keys %gone), "\n";
}
$self->{unindexed}->{$_}++ foreach keys %gone;
- $_->remote_remove($oid, $mid) foreach @{$self->{idx_parts}};
- $self->{over}->remove_oid($oid, $mid);
- $self->barrier;
+ $self->unindex_oid_remote($oid, $mid);
}
}
sub unindex {
my ($self, $opts, $git, $unindex_range) = @_;
my $un = $self->{unindexed} ||= {}; # num => removal count
- $self->barrier;
my $before = scalar keys %$un;
my @cmd = qw(log --raw -r
--no-notes --no-color --no-abbrev --no-renames);
sub index_sync {
my ($self, $opts) = @_;
$opts ||= {};
- my $ibx = $self->{-inbox};
my $epoch_max;
my $latest = git_dir_latest($self, \$epoch_max);
return unless defined $latest;