use PublicInbox::Msgmap;
use PublicInbox::Spawn qw(spawn popen_rd);
use PublicInbox::SearchIdx;
-use PublicInbox::MultiMidQueue;
use IO::Handle; # ->autoflush
use File::Temp qw(tempfile);
+my $x40 = qr/[a-f0-9]{40}/;
# an estimate of the post-packed size to the raw uncompressed size
my $PACKING_FACTOR = 0.4;
# returns undef on duplicate or spam
# mimics Import::add and wraps it for v2
sub add {
- my ($self, $mime, $check_cb) = @_;
- $self->{-inbox}->with_umask(sub {
- _add($self, $mime, $check_cb)
- });
+ my ($self, $eml, $check_cb) = @_;
+ $self->{-inbox}->with_umask(\&_add, $self, $eml, $check_cb);
}
# indexes a message, returns true if checkpointing is needed
$self->{idx_shards}->[$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};
+
+ # 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->{-inbox}->{inboxdir}/msgmap.sqlite3", 1);
+ $mm->{dbh}->begin_work;
+}
+
# idempotent
sub idx_init {
my ($self, $opt) = @_;
# do not leak read-only FDs to child processes, we only have these
# FDs for duplicate detection so they should not be
# frequently activated.
+ # delete @$ibx{qw(git mm search)};
delete $ibx->{$_} foreach (qw(git mm search));
- my $indexlevel = $ibx->{indexlevel};
- if ($indexlevel && $indexlevel eq 'basic') {
- $self->{parallel} = 0;
- }
-
+ $self->{parallel} = 0 if ($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,
$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 ]
# 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->header_obj);
# 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) {
}
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
$self->{last_commit}->[$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->{-inbox}->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->{-inbox}->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
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: $!";
-
+ pipe(my($in, $w)) or die "pipe: $!";
my $git_dir = $self->{-inbox}->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 $$raw or die "print \$w: $!";
+ close $w or die "close \$w: $!";
local $/ = "\n";
chomp(my $oid = <$r>);
close $r or die "git hash-object failed: $?";
$self->{bnote}->[1];
}
-sub mark_deleted ($$$$) {
- my ($self, $sync, $git, $oid) = @_;
- return if PublicInbox::SearchIdx::too_big($self, $git, $oid);
- my $msgref = $git->cat_file($oid);
- my $mime = PublicInbox::Eml->new($$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 ($$$) {
my ($self, $sync, $git) = @_;
$sync->{mm_tmp}->atfork_parent;
}
-# only for a few odd messages with multiple Message-IDs
-sub reindex_oid_m ($$$$;$) {
- my ($self, $sync, $git, $oid, $regen_num) = @_;
- $self->{current_info} = "multi_mid $oid";
- my ($num, $mid0, $len);
- my $msgref = $git->cat_file($oid, \$len);
- my $mime = PublicInbox::Eml->new($$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";
- }
- 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(#$num)";
- }
- }
- unless (defined($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)) {
- $sync->{mm_tmp}->num_delete($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";
- }
- } 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) = @_;
+ if (my $D = $sync->{D}) { # don't waste I/O on deletes
+ return if $D->{pack('H*', $oid)};
+ }
return if PublicInbox::SearchIdx::too_big($self, $git, $oid);
my ($num, $mid0, $len);
my $msgref = $git->cat_file($oid, \$len);
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);
+ # {unindexed} is unlikely
+ if ((my $unindexed = $self->{unindexed}) && scalar(@$mids) == 1) {
+ $num = delete($unindexed->{$mids->[0]});
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;
- }
- $mid0 = $mid;
+ $mid0 = $mids->[0];
+ $self->{mm}->mid_set($num, $mid0);
+ delete($self->{unindexed}) if !keys(%$unindexed);
+ }
+ }
+ 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;
}
- } else { # multiple MIDs are a weird case:
- my $del = 0;
- for (@$mids) {
- $del += delete($sync->{D}->{"$_\0$chash"}) // 0;
+ }
+ $mid0 //= do { # is this a number we got before?
+ $num = $sync->{mm_tmp}->num_for($mids->[0]);
+ defined($num) ? $mids->[0] : undef;
+ };
+ if (!defined($num)) {
+ for (my $i = $#$mids; $i >= 1; $i--) {
+ $num = $sync->{mm_tmp}->num_for($mids->[$i]);
+ if (defined($num)) {
+ $mid0 = $mids->[$i];
+ last;
+ }
}
- 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)) {
+ $sync->{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;
+ }
+ }
}
+ }
+ 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}++;
my $smsg = bless {
raw_bytes => $len,
$range;
}
+# don't bump num_highwater on --reindex
+sub mark_deleted ($$$) {
+ my ($git, $sync, $range) = @_;
+ my $D = $sync->{D} //= {}; # pack("H*", $oid) => NR
+ my $fh = $git->popen(qw(log --raw --no-abbrev
+ --pretty=tformat:%H
+ --no-notes --no-color --no-renames
+ --diff-filter=AM), $range, '--', 'd');
+ while (<$fh>) {
+ if (/\A:\d{6} 100644 $x40 ($x40) [AM]\td$/o) {
+ $D->{pack('H*', $1)}++;
+ }
+ }
+ close $fh or die "git log failed: \$?=$?";
+}
+
sub sync_prepare ($$$) {
my ($self, $sync, $epoch_max) = @_;
my $pr = $sync->{-opt}->{-progress};
# without {reindex}
my $reindex_heads = last_commits($self, $epoch_max) if $sync->{reindex};
- for (my $i = $epoch_max; $i >= 0; $i--) {
+ for my $i (0..$epoch_max) {
die 'BUG: already indexing!' if $self->{reindex_pipe};
my $git_dir = git_dir_n($self, $i);
-d $git_dir or next; # missing epochs are fine
close $fh or die "git log failed: \$?=$?";
$pr->("$n\n") if $pr;
$regen_max += $n;
+ mark_deleted($git, $sync, $range) if $sync->{reindex};
}
-
return 0 if (!$regen_max && !keys(%{$self->{unindex_range}}));
# reindex should NOT see new commits anymore, if we do,
sub unindex_oid_remote ($$$) {
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});
+ $idx->remote_remove($oid, $num);
+ }
}
sub unindex_oid ($$$;$) {
my ($id, $prev);
while (my $smsg = $over->next_by_mid($mid, \$id, \$prev)) {
$gone{$smsg->{num}} = 1 if $oid eq $smsg->{blob};
- 1; # continue
}
- my $n = scalar keys %gone;
- next unless $n;
+ my $n = scalar(keys(%gone)) or next;
if ($n > 1) {
warn "BUG: multiple articles linked to $oid\n",
join(',',sort keys %gone), "\n";
}
}
-my $x40 = qr/[a-f0-9]{40}/;
sub unindex ($$$$) {
my ($self, $sync, $git, $unindex_range) = @_;
my $unindexed = $self->{unindexed} ||= {}; # $mid0 => $num
if (my $pr = $sync->{-opt}->{-progress}) {
$pr->("$i.git indexing $range\n");
}
-
- my @cmd = qw(log --raw -r --pretty=tformat:%H.%at.%ct
+ my @cmd = qw(log --reverse --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;
+ my $D = $sync->{D};
while (<$fh>) {
chomp;
$self->{current_info} = "$i.git $_";
if (/\A($x40)\.([0-9]+)\.([0-9]+)$/o) {
- $cmt //= $1;
+ $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);
+ # allow re-add if there was user error
+ my $oid = $1;
+ if ($D) {
+ my $oid_bin = pack('H*', $oid);
+ my $nr = --$D->{$oid_bin};
+ delete($D->{$oid_bin}) if $nr <= 0;
+ }
+ unindex_oid($self, $git, $oid);
}
}
close $fh or die "git log failed: \$?=$?";
my $latest = git_dir_latest($self, \$epoch_max);
return unless defined $latest;
$self->idx_init($opt); # acquire lock
+ $self->{over}->rethread_prepare($opt);
my $sync = {
- D => {}, # "$mid\0$chash" => $oid
unindex_range => {}, # EPOCH => oid_old..oid_new
reindex => $opt->{reindex},
-opt => $opt
};
$sync->{ranges} = sync_ranges($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;
$sync->{mm_tmp} = $self->{mm}->tmp_clone;
}
- # work backwards through history
- for (my $i = $epoch_max; $i >= 0; $i--) {
- index_epoch($self, $sync, $i);
- }
-
- # 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);
- }
- }
- 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);
- }
- }
- }
- $git->cleanup if $git;
+ # work forwards through history
+ index_epoch($self, $sync, $_) for (0..$epoch_max);
$self->done;
if (my $nr = $sync->{nr}) {
my $pr = $sync->{-opt}->{-progress};
$pr->('all.git '.sprintf($sync->{-regen_fmt}, $nr)) if $pr;
}
+ $self->{over}->rethread_done($opt);
# 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);
}
}