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 PublicInbox::Spawn qw(spawn);
+use PublicInbox::SearchIdx;
use IO::Handle;
# an estimate of the post-packed size to the raw uncompressed size
my $PACKING_FACTOR = 0.4;
# assume 2 cores if GNU nproc(1) is not available
-sub nproc () {
- int($ENV{NPROC} || `nproc 2>/dev/null` || 2);
+sub nproc_parts () {
+ my $n = int($ENV{NPROC} || `nproc 2>/dev/null` || 2);
+ # subtract for the main process and git-fast-import
+ $n -= 1;
+ $n < 1 ? 1 : $n;
}
-sub new {
- my ($class, $v2ibx, $creat) = @_;
- my $dir = $v2ibx->{mainrepo} or die "no mainrepo in inbox\n";
- unless (-d $dir) {
- if ($creat) {
- require File::Path;
- File::Path::mkpath($dir);
- } else {
- die "$dir does not exist\n";
- }
- }
-
+sub count_partitions ($) {
+ my ($self) = @_;
my $nparts = 0;
- my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION;
+ my $xpfx = $self->{xpfx};
# always load existing partitions in case core count changes:
+ # Also, partition count may change while -watch is running
+ # due to -compact
if (-d $xpfx) {
foreach my $part (<$xpfx/*>) {
-d $part && $part =~ m!/\d+\z! or next;
};
}
}
- $nparts = nproc() if ($nparts == 0);
+ $nparts;
+}
+
+sub new {
+ my ($class, $v2ibx, $creat) = @_;
+ my $dir = $v2ibx->{mainrepo} or die "no mainrepo in inbox\n";
+ unless (-d $dir) {
+ if ($creat) {
+ require File::Path;
+ File::Path::mkpath($dir);
+ } else {
+ die "$dir does not exist\n";
+ }
+ }
$v2ibx = PublicInbox::InboxWritable->new($v2ibx);
+
+ my $xpfx = "$dir/xap" . PublicInbox::Search::SCHEMA_VERSION;
my $self = {
-inbox => $v2ibx,
im => undef, # PublicInbox::Import
- partitions => $nparts,
parallel => 1,
transact_bytes => 0,
- over => PublicInbox::OverIdxFork->new("$xpfx/over.sqlite3"),
+ xpfx => $xpfx,
+ over => PublicInbox::OverIdx->new("$xpfx/over.sqlite3", 1),
lock_path => "$dir/inbox.lock",
- # limit each repo to 1GB or so
+ # limit each git repo (epoch) to 1GB or so
rotate_bytes => int((1024 * 1024 * 1024) / $PACKING_FACTOR),
- last_commit => [],
+ last_commit => [], # git repo -> commit
};
+ $self->{partitions} = count_partitions($self) || nproc_parts();
bless $self, $class;
}
my ($self, $parallel) = @_;
$self->{parallel} = $parallel;
$self->idx_init;
- my $max_git = -1;
- git_dir_latest($self, \$max_git);
- $self->git_init($max_git >= 0 ? $max_git : 0);
+ my $epoch_max = -1;
+ git_dir_latest($self, \$epoch_max);
+ $self->git_init($epoch_max >= 0 ? $epoch_max : 0);
$self->done;
}
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->{max_git}] = $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);
my $hdr = $mime->header_obj;
my $dig = content_digest($mime);
- $$mid0 = PublicInbox::Import::digest2mid($dig);
+ $$mid0 = PublicInbox::Import::digest2mid($dig, $hdr);
my $num = $self->{mm}->mid_insert($$mid0);
unless (defined $num) {
# it's hard to spoof the last Received: header
my @recvd = $hdr->header_raw('Received');
$dig->add("Received: $_") foreach (@recvd);
- $$mid0 = PublicInbox::Import::digest2mid($dig);
+ $$mid0 = PublicInbox::Import::digest2mid($dig, $hdr);
$num = $self->{mm}->mid_insert($$mid0);
# fall back to a random Message-ID and give up determinism:
until (defined($num)) {
$dig->add(rand);
- $$mid0 = PublicInbox::Import::digest2mid($dig);
+ $$mid0 = PublicInbox::Import::digest2mid($dig, $hdr);
warn "using random Message-ID <$$mid0> as fallback\n";
$num = $self->{mm}->mid_insert($$mid0);
}
# 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);
+ if ($nparts && $nparts != $self->{partitions}) {
+ $self->{partitions} = $nparts;
+ }
# need to create all parts before initializing msgmap FD
my $max = $self->{partitions} - 1;
$self->done;
my $pfx = "$self->{-inbox}->{mainrepo}/git";
my $purges = [];
- foreach my $i (0..$self->{max_git}) {
+ foreach my $i (0..$self->{epoch_max}) {
my $git = PublicInbox::Git->new("$pfx/$i.git");
my $im = $self->import_init($git, 0, 1);
$purges->[$i] = $im->purge_oids($purge);
$purges;
}
+sub content_ids ($) {
+ my ($mime) = @_;
+ my @cids = ( content_id($mime) );
+
+ # Email::MIME->as_string doesn't always round-trip, so we may
+ # use a second content_id
+ my $rt = content_id(PublicInbox::MIME->new(\($mime->as_string)));
+ push @cids, $rt if $cids[0] ne $rt;
+ \@cids;
+}
+
+sub content_matches ($$) {
+ my ($cids, $existing) = @_;
+ my $cid = content_id($existing);
+ foreach (@$cids) {
+ return 1 if $_ eq $cid
+ }
+ 0
+}
+
sub remove_internal {
my ($self, $mime, $cmt_msg, $purge) = @_;
- $self->barrier;
$self->idx_init;
my $im = $self->importer unless $purge;
- my $ibx = $self->{-inbox};
- my $srch = $ibx->search;
- my $cid = content_id($mime);
+ my $over = $self->{over};
+ my $cids = content_ids($mime);
my $parts = $self->{idx_parts};
my $mm = $self->{mm};
my $removed;
my $mark;
foreach my $mid (@$mids) {
- $srch->reopen->each_smsg_by_mid($mid, sub {
- my ($smsg) = @_;
- $smsg->load_expand;
- my $msg = $ibx->msg_by_smsg($smsg);
+ my %gone;
+ my ($id, $prev);
+ while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
+ my $msg = get_blob($self, $smsg);
if (!defined($msg)) {
warn "broken smsg for $mid\n";
- return 1; # continue
+ next; # continue
}
my $orig = $$msg;
my $cur = PublicInbox::MIME->new($msg);
- if (content_id($cur) eq $cid) {
- $mm->num_delete($smsg->num);
- # $removed should only be set once assuming
- # no bugs in our deduplication code:
- $removed = $smsg;
- $removed->{mime} = $cur;
- my $oid = $smsg->{blob};
- if ($purge) {
- $purge->{$oid} = 1;
- } else {
- ($mark, undef) =
- $im->remove(\$orig, $cmt_msg);
- }
- $orig = undef;
- $removed->num; # memoize this for callers
-
- foreach my $idx (@$parts) {
- $idx->remote_remove($oid, $mid);
- }
- $self->{over}->remove_oid($oid, $mid);
+ if (content_matches($cids, $cur)) {
+ $smsg->{mime} = $cur;
+ $gone{$smsg->{num}} = [ $smsg, \$orig ];
}
- 1; # continue
- });
- $self->barrier;
+ }
+ my $n = scalar keys %gone;
+ next unless $n;
+ if ($n > 1) {
+ warn "BUG: multiple articles linked to <$mid>\n",
+ join(',', sort keys %gone), "\n";
+ }
+ foreach my $num (keys %gone) {
+ my ($smsg, $orig) = @{$gone{$num}};
+ $mm->num_delete($num);
+ # $removed should only be set once assuming
+ # no bugs in our deduplication code:
+ $removed = $smsg;
+ my $oid = $smsg->{blob};
+ if ($purge) {
+ $purge->{$oid} = 1;
+ } else {
+ ($mark, undef) = $im->remove($orig, $cmt_msg);
+ }
+ $orig = undef;
+ $self->unindex_oid_remote($oid, $mid);
+ }
}
if (defined $mark) {
my $cmt = $im->get_mark($mark);
- $self->{last_commit}->[$self->{max_git}] = $cmt;
+ $self->{last_commit}->[$self->{epoch_max}] = $cmt;
}
if ($purge && scalar keys %$purge) {
return purge_oids($self, $purge);
$purges;
}
+sub last_commit_part ($$;$) {
+ my ($self, $i, $cmt) = @_;
+ my $v = PublicInbox::Search::SCHEMA_VERSION();
+ $self->{mm}->last_commit_xap($v, $i, $cmt);
+}
+
sub set_last_commits ($) {
my ($self) = @_;
- defined(my $max_git = $self->{max_git}) or return;
- my $mm = $self->{mm};
+ defined(my $epoch_max = $self->{epoch_max}) or return;
my $last_commit = $self->{last_commit};
- foreach my $i (0..$max_git) {
+ foreach my $i (0..$epoch_max) {
defined(my $cmt = $last_commit->[$i]) or next;
$last_commit->[$i] = undef;
- $mm->last_commit_n($i, $cmt);
+ last_commit_part($self, $i, $cmt);
}
}
-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;
+
+ # SQLite overview is third
+ $self->{over}->commit_lazy;
- # Now deal with Xapian and overview DB
- $over->barrier_init(scalar(@$parts));
+ # Now deal with Xapian
+ if ($wait) {
+ my $barrier = $self->barrier_init(scalar @$parts);
- # each partition needs to issue a barrier command to over
- $_->remote_barrier foreach @$parts;
+ # each partition needs to issue a barrier command
+ $_->remote_barrier for @$parts;
- $over->barrier_wait; # wait for each Xapian partition
- $over->commit_fsync if $fsync;
+ # 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, $new) = @_;
+ my ($self, $epoch) = @_;
my $pfx = "$self->{-inbox}->{mainrepo}/git";
- my $git_dir = "$pfx/$new.git";
+ my $git_dir = "$pfx/$epoch.git";
my @cmd = (qw(git init --bare -q), $git_dir);
PublicInbox::Import::run_die(\@cmd);
PublicInbox::Import::run_die(\@cmd);
my $alt = "$all/objects/info/alternates";
- my $new_obj_dir = "../../git/$new.git/objects";
+ my $new_obj_dir = "../../git/$epoch.git/objects";
my %alts;
if (-e $alt) {
open(my $fh, '<', $alt) or die "open < $alt: $!\n";
} else {
$self->{im} = undef;
$im->done;
- $self->barrier(1);
$im = undef;
- my $git_dir = $self->git_init(++$self->{max_git});
+ $self->checkpoint;
+ my $git_dir = $self->git_init(++$self->{epoch_max});
my $git = PublicInbox::Git->new($git_dir);
return $self->import_init($git, 0);
}
}
- my $new = 0;
+ my $epoch = 0;
my $max;
my $latest = git_dir_latest($self, \$max);
if (defined $latest) {
my $git = PublicInbox::Git->new($latest);
my $packed_bytes = $git->packed_bytes;
if ($packed_bytes >= $self->{rotate_bytes}) {
- $new = $max + 1;
+ $epoch = $max + 1;
} else {
- $self->{max_git} = $max;
+ $self->{epoch_max} = $max;
return $self->import_init($git, $packed_bytes);
}
}
- $self->{max_git} = $new;
- $latest = $self->git_init($new);
+ $self->{epoch_max} = $epoch;
+ $latest = $self->git_init($epoch);
$self->import_init(PublicInbox::Git->new($latest), 0);
}
sub diff ($$$) {
my ($mid, $cur, $new) = @_;
use File::Temp qw(tempfile);
- use PublicInbox::Spawn qw(spawn);
my ($ah, $an) = tempfile('email-cur-XXXXXXXX', TMPDIR => 1);
print $ah $cur->as_string or die "print: $!";
unlink($an, $bn);
}
-sub lookup_content {
- my ($self, $mime, $mid) = @_;
+sub get_blob ($$) {
+ my ($self, $smsg) = @_;
+ if (my $im = $self->{im}) {
+ my $msg = $im->cat_blob($smsg->{blob});
+ return $msg if $msg;
+ }
+ # older message, should be in alternates
my $ibx = $self->{-inbox};
+ $ibx->msg_by_smsg($smsg);
+}
- my $srch = $ibx->search->reopen;
- my $cid = content_id($mime);
- my $found;
- $srch->each_smsg_by_mid($mid, sub {
- my ($smsg) = @_;
- $smsg->load_expand;
- my $msg = $ibx->msg_by_smsg($smsg);
+sub lookup_content {
+ my ($self, $mime, $mid) = @_;
+ my $over = $self->{over};
+ my $cids = content_ids($mime);
+ my ($id, $prev);
+ while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
+ my $msg = get_blob($self, $smsg);
if (!defined($msg)) {
warn "broken smsg for $mid\n";
- return 1; # continue
+ next;
}
my $cur = PublicInbox::MIME->new($msg);
- if (content_id($cur) eq $cid) {
+ if (content_matches($cids, $cur)) {
$smsg->{mime} = $cur;
- $found = $smsg;
- return 0; # break out of loop
+ return $smsg;
}
+
# XXX DEBUG_DIFF is experimental and may be removed
diff($mid, $cur, $mime) if $ENV{DEBUG_DIFF};
-
- 1; # continue
- });
- $found;
+ }
+ undef;
}
sub atfork_child {
$im->atfork_child;
}
die "unexpected mm" if $self->{mm};
+ close $self->{bnote}->[0] or die "close bnote[0]: $!\n";
+ $self->{bnote}->[1];
}
sub mark_deleted {
$mm_tmp->mid_delete($mid0) or
die "failed to delete <$mid0> for article #$num\n";
+ $self->{over}->add_overview($mime, $len, $num, $oid, $mid0);
my $nparts = $self->{partitions};
my $part = $num % $nparts;
my $idx = $self->idx_part($part);
# only update last_commit for $i on reindex iff newer than current
sub update_last_commit {
my ($self, $git, $i, $cmt) = @_;
- my $last = $self->{mm}->last_commit_n($i);
+ my $last = last_commit_part($self, $i);
if (defined $last && is_ancestor($git, $last, $cmt)) {
my @cmd = (qw(rev-list --count), "$last..$cmt");
chomp(my $n = $git->qx(@cmd));
return if $n ne '' && $n == 0;
}
- $self->{mm}->last_commit_n($i, $cmt);
+ last_commit_part($self, $i, $cmt);
}
sub git_dir_n ($$) { "$_[0]->{-inbox}->{mainrepo}/git/$_[1].git" }
sub last_commits {
- my ($self, $max_git) = @_;
+ my ($self, $epoch_max) = @_;
my $heads = [];
- for (my $i = $max_git; $i >= 0; $i--) {
- $heads->[$i] = $self->{mm}->last_commit_n($i);
+ for (my $i = $epoch_max; $i >= 0; $i--) {
+ $heads->[$i] = last_commit_part($self, $i);
}
$heads;
}
-sub is_ancestor ($$$) {
- my ($git, $cur, $tip) = @_;
- return 0 unless $git->check($cur);
- my $cmd = [ 'git', "--git-dir=$git->{git_dir}",
- qw(merge-base --is-ancestor), $cur, $tip ];
- my $pid = spawn($cmd);
- defined $pid or die "spawning ".join(' ', @$cmd)." failed: $!";
- waitpid($pid, 0) == $pid or die join(' ', @$cmd) .' did not finish';
- $? == 0;
-}
+*is_ancestor = *PublicInbox::SearchIdx::is_ancestor;
sub index_prepare {
- my ($self, $opts, $max_git, $ranges) = @_;
+ my ($self, $opts, $epoch_max, $ranges) = @_;
my $regen_max = 0;
my $head = $self->{-inbox}->{ref_head} || 'refs/heads/master';
- for (my $i = $max_git; $i >= 0; $i--) {
+ for (my $i = $epoch_max; $i >= 0; $i--) {
die "already indexing!\n" if $self->{index_pipe};
my $git_dir = git_dir_n($self, $i);
-d $git_dir or next; # missing parts are fine
\$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 $max_git;
- my $latest = git_dir_latest($self, \$max_git);
+ my $epoch_max;
+ my $latest = git_dir_latest($self, \$epoch_max);
return unless defined $latest;
$self->idx_init; # acquire lock
my $mm_tmp = $self->{mm}->tmp_clone;
- my $ranges = $opts->{reindex} ? [] : $self->last_commits($max_git);
+ my $ranges = $opts->{reindex} ? [] : $self->last_commits($epoch_max);
my ($min, $max) = $mm_tmp->minmax;
- my $regen = $self->index_prepare($opts, $max_git, $ranges);
+ my $regen = $self->index_prepare($opts, $epoch_max, $ranges);
$$regen += $max if $max;
my $D = {};
my @cmd = qw(log --raw -r --pretty=tformat:%h
# work backwards through history
my $last_commit = [];
- for (my $i = $max_git; $i >= 0; $i--) {
+ for (my $i = $epoch_max; $i >= 0; $i--) {
my $git_dir = git_dir_n($self, $i);
die "already reindexing!\n" if delete $self->{reindex_pipe};
-d $git_dir or next; # missing parts are fine