user/dev discussion of public-inbox itself
 help / color / mirror / code / Atom feed
From: Eric Wong <e@80x24.org>
To: meta@public-inbox.org
Subject: [RFC] v2writable: reindex handles 3-headed monsters [was: BUG: --reindex broken on multiple Message-IDs reuse]
Date: Thu, 17 Oct 2019 11:22:15 +0000	[thread overview]
Message-ID: <20191017112215.GA13175@dcvr> (raw)
In-Reply-To: <20191016211415.GA6084@dcvr>

Eric Wong <e@80x24.org> wrote:
> Initial indexing went fine,

Actually, no.  Initial indexing was broken when dealing with
this horror show.  But I think I can fix that...

> but --reindex fails because it's
> trying to use $sync->{regen} when it should not...

Yes, --reindex can be fixed if initial indexing isn't totally
broken to begin with.  Or, if over.sqlite3 is deleted..

But --reindex won't fix things if initial indexing was
broken....

> So that's a bug in the reindexing logic somewhere that
> needs to be fixed w/o disrupting normal indexing...
> 
> One of the messages has 3 Message-IDs:
> https://lore.kernel.org/linux-renesas-soc/1923946.Jvi0TDUXFC@wasted.cogentembedded.com

So we had broken code trying to workaround broken data :<

Anyways, the following patch is probably OK, but I'm way too
tired and need to do some more testing on existing repos.
Unfortunately, --reindex alone isn't yet strong enough to fix
cases that were already broken...

----------8<------------
Subject: [RFC] v2writable: reindex handles 3-headed monsters

And maybe 8-headed ones, too; but not very gracefully...

I noticed --reindex failing on the linux-renesas-soc mirror due
one 3-headed monster of a message having 3 sets of headers;
while another normal message had a Message-ID that matched one
of the 3 IDs of the 3-headed monster.

Since we do reindexing backwards, we can try to do our best to
preserve NNTP article numbers by matching blob OIDs if the
overview DB exists by re-reindexing.

Some of these may benefit from using List::Util::shuffle
in case we hit infinite loops with reindex_q...

Link: https://public-inbox.org/meta/20191016211415.GA6084@dcvr/
---
 TODO                          |   3 +
 lib/PublicInbox/OverIdx.pm    |  14 +++++
 lib/PublicInbox/V2Writable.pm | 107 ++++++++++++++++++++++++++++------
 t/v2reindex.t                 |  65 +++++++++++++++++++++
 4 files changed, 171 insertions(+), 18 deletions(-)

diff --git a/TODO b/TODO
index a327ca06..61c44a84 100644
--- a/TODO
+++ b/TODO
@@ -133,3 +133,6 @@ all need to be considered for everything we introduce)
   for coderepos
 
 * configurable diff output for solver-generated blobs
+
+* fix search for messages with multiple Subject:/To:/From:/Date:
+  headers (some wacky examples out there...)
diff --git a/lib/PublicInbox/OverIdx.pm b/lib/PublicInbox/OverIdx.pm
index 7fd1905d..1ff435d7 100644
--- a/lib/PublicInbox/OverIdx.pm
+++ b/lib/PublicInbox/OverIdx.pm
@@ -343,6 +343,20 @@ sub remove_oid {
 	$nr;
 }
 
+sub num_for_oid {
+	my ($self, $oid, $mid) = @_;
+	my $num;
+	$self->begin_lazy;
+	each_by_mid($self, $mid, ['ddd'], sub {
+		my ($smsg) = @_;
+		my $blob = $smsg->{blob};
+		return 1 if (!defined($blob) || $blob ne $oid); # continue;
+		$num = $smsg->{num};
+		0; # done
+	});
+	$num;
+}
+
 sub create_tables {
 	my ($dbh) = @_;
 
diff --git a/lib/PublicInbox/V2Writable.pm b/lib/PublicInbox/V2Writable.pm
index 6a88f62a..6e1ac5e3 100644
--- a/lib/PublicInbox/V2Writable.pm
+++ b/lib/PublicInbox/V2Writable.pm
@@ -845,6 +845,51 @@ sub mark_deleted ($$$$) {
 	}
 }
 
+sub regen_art_num ($$$$$) {
+	my ($self, $sync, $git, $oid, $mids) = @_;
+	my $num = $sync->{regen}--;
+	my $mm = $self->{mm};
+	if ($num > 0) {
+		foreach my $mid (reverse @$mids) {
+			if ($mm->mid_set($num, $mid) == 1) {
+				return ($num, $mid);
+			}
+		}
+
+		# nope, not using the regen number because either we
+		# have or overlap Message-IDs with another message.
+		$sync->{regen}++;
+		$num = undef;
+	}
+
+	# --reindex was broken when Message-IDs were reused with a
+	# message that already had multiple Message-IDs.  This causes
+	# failures when using --reindex again.  The least bad option is
+	# to show messages twice to NNTP clients, rather than lose/drop
+	# messages entirely.
+	#
+	# create a new article number?
+	for my $mid (@$mids) {
+		$num = $self->{mm}->mid_insert($mid);
+		return ($num, $mid) if defined $num;
+	}
+
+	# swap article numbers with a previously regenerated (newer message)
+	# which has an overlapping Message-ID.  The current $oid is older
+	# than $smsg->{blob} in git history since we regen walks backwards.
+	my $reindex_q = $sync->{reindex_q} //= [];
+	for my $mid (@$mids) {
+		my $n = $self->{mm}->num_for($mid);
+		next unless defined $n;
+		if (my $smsg = $self->{over}->get_art($n)) {
+			push @$reindex_q, $smsg->{blob};
+			return ($n, $mid);
+		}
+	}
+	warn "W: ran out of article numbers on $oid\n";
+	(undef, undef);
+}
+
 sub reindex_oid ($$$$) {
 	my ($self, $sync, $git, $oid) = @_;
 	my $len;
@@ -853,31 +898,43 @@ sub reindex_oid ($$$$) {
 	my $mids = mids($mime->header_obj);
 	my $cid = content_id($mime);
 
-	# get the NNTP article number we used before, highest number wins
-	# and gets deleted from sync->{mm_tmp};
+	# get the NNTP article number we used before, which gets deleted
+	# from sync->{mm_tmp};
 	my $mid0;
 	my $num = -1;
 	my $del = 0;
-	foreach my $mid (@$mids) {
-		$del += delete($sync->{D}->{"$mid\0$cid"}) ? 1 : 0;
-		my $n = $sync->{mm_tmp}->num_for($mid);
-		if (defined $n && $n > $num) {
-			$mid0 = $mid;
-			$num = $n;
-			$self->{mm}->mid_set($num, $mid0);
+	my $reindex = $sync->{reindex};
+	if ($reindex) {
+		# see if it's already in the overview DB, but keep in mind
+		# --reindex may be used blindly w/o overview DB.
+		foreach my $mid (@$mids) {
+			$del += delete($sync->{D}->{"$mid\0$cid"}) ? 1 : 0;
+			my $n = $self->{over}->num_for_oid($oid, $mid);
+			if (defined $n) {
+				($num, $mid0) = ($n, $mid);
+				$self->{mm}->mid_set($num, $mid0);
+				last; # yay! reused
+			}
 		}
 	}
-	if (!defined($mid0) && !$del) {
-		$num = $sync->{regen}--;
-		die "BUG: ran out of article numbers\n" if $num <= 0;
-		my $mm = $self->{mm};
-		foreach my $mid (reverse @$mids) {
-			if ($mm->mid_set($num, $mid) == 1) {
-				$mid0 = $mid;
-				last;
+
+	# not reindexing, or reindexing with broken/incomplete overview DB:
+	if (!defined($mid0)) {
+		# highest number wins for unseen messages
+		foreach my $mid (@$mids) {
+			$del += delete($sync->{D}->{"$mid\0$cid"}) ? 1 : 0;
+			my $n = $sync->{mm_tmp}->num_for($mid);
+			if (defined $n && $n > $num) {
+				($num, $mid0) = ($n, $mid);
+				$self->{mm}->mid_set($num, $mid0);
 			}
 		}
+	}
+
+	if (!defined($mid0) && !$del) {
+		($num, $mid0) = regen_art_num($self, $sync, $git, $oid, $mids);
 		if (!defined($mid0)) {
+			my $mm = $self->{mm};
 			my $id = '<' . join('> <', @$mids) . '>';
 			warn "Message-ID $id unusable for $num\n";
 			foreach my $mid (@$mids) {
@@ -890,7 +947,11 @@ sub reindex_oid ($$$$) {
 	if (!defined($mid0) || $del) {
 		if (!defined($mid0) && $del) { # expected for deletes
 			$num = $sync->{regen}--;
-			$self->{mm}->num_highwater($num) if !$sync->{reindex};
+			if (!$reindex) {
+				($num <= 0) and
+					die "BUG: ran out of article numbers\n";
+				$self->{mm}->num_highwater($num);
+			}
 			return
 		}
 
@@ -1147,6 +1208,16 @@ sub index_epoch ($$$) {
 		} elsif (/\A:\d{6} 100644 $x40 ($x40) [AM]\td$/o) {
 			mark_deleted($self, $sync, $git, $1);
 		}
+
+		while (my $reindex_q = delete $sync->{reindex_q}) {
+			my $all = $self->{-inbox}->git;
+			for my $oid (@$reindex_q) {
+				$self->{current_info} = "$i.git $oid";
+				warn "reindexing to shuffle article numbers\n";
+				reindex_oid($self, $sync, $all, $oid);
+			}
+			$all->cleanup;
+		}
 	}
 	$fh = undef;
 	delete $self->{reindex_pipe};
diff --git a/t/v2reindex.t b/t/v2reindex.t
index 7c5a6b07..5de067c6 100644
--- a/t/v2reindex.t
+++ b/t/v2reindex.t
@@ -431,4 +431,69 @@ ok(!-d $xap, 'Xapian directories removed again');
 		  ], 'msgmap as expected' );
 }
 
+# A real example from linux-renesas-soc on lore where a 3-headed monster
+# of a message has 3 sets of common headers.  Another normal message
+# previously existed with a single Message-ID that conflicts with one
+# of the Message-IDs in the 3-headed monster.
+{
+	my @warn;
+	local $SIG{__WARN__} = sub { push @warn, @_ };
+	my %config = %$ibx_config;
+	$config{indexlevel} = 'basic';
+	my $ibx = PublicInbox::Inbox->new(\%config);
+	my $im = PublicInbox::V2Writable->new($ibx);
+	my $m3 = PublicInbox::MIME->new(<<'EOF');
+Date: Tue, 24 May 2016 14:34:22 -0700 (PDT)
+Message-Id: <20160524.143422.552507610109476444.d@example.com>
+To: t@example.com
+Cc: c@example.com
+Subject: Re: [PATCH v2 2/2]
+From: <f@example.com>
+In-Reply-To: <1463825855-7363-2-git-send-email-y@example.com>
+References: <1463825855-7363-1-git-send-email-y@example.com>
+	<1463825855-7363-2-git-send-email-y@example.com>
+Date: Wed, 25 May 2016 10:01:51 +0900
+From: h@example.com
+To: g@example.com
+Cc: m@example.com
+Subject: Re: [PATCH]
+Message-ID: <20160525010150.GD7292@example.com>
+References: <1463498133-23918-1-git-send-email-g+r@example.com>
+In-Reply-To: <1463498133-23918-1-git-send-email-g+r@example.com>
+From: s@example.com
+To: h@example.com
+Cc: m@example.com
+Subject: [PATCH 12/13]
+Date: Wed, 01 Jun 2016 01:32:35 +0300
+Message-ID: <1923946.Jvi0TDUXFC@wasted.example.com>
+In-Reply-To: <13205049.n7pM8utpHF@wasted.example.com>
+References: <13205049.n7pM8utpHF@wasted.example.com>
+
+Somehow we got a message with 3 sets of headers into one
+message, could've been something broken on the archiver side.
+EOF
+
+	my $m1 = PublicInbox::MIME->new(<<'EOF');
+From: a@example.com
+To: t@example.com
+Subject: [PATCH 12/13]
+Date: Wed, 01 Jun 2016 01:32:35 +0300
+Message-ID: <1923946.Jvi0TDUXFC@wasted.example.com>
+In-Reply-To: <13205049.n7pM8utpHF@wasted.example.com>
+References: <13205049.n7pM8utpHF@wasted.example.com>
+
+This is probably one of the original messages
+
+EOF
+	$im->add($m1);
+	$im->add($m3);
+	$im->done;
+	remove_tree($xap);
+	eval { $im->index_sync() };
+	is($@, '', 'no error from initial indexing');
+	eval { $im->index_sync({reindex=>1}) };
+	is($@, '', 'no error from reindexing after reused Message-ID (x3)');
+	is_deeply(\@warn, [], 'no warnings');
+}
+
 done_testing();

  reply	other threads:[~2019-10-17 11:22 UTC|newest]

Thread overview: 11+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-10-16 21:14 BUG: --reindex broken on multiple Message-IDs reuse Eric Wong
2019-10-17 11:22 ` Eric Wong [this message]
2019-10-22  8:09   ` [RFC/HELP] search: multiple From/To/Cc/Subject (what about Date?) Eric Wong
2019-10-21 11:02 ` [PATCH 0/3] fix reindex on multiple + overlapping Message-IDs Eric Wong
2019-10-21 11:02   ` [PATCH 1/3] v2writable: set unindexed article number Eric Wong
2019-10-21 11:02   ` [PATCH 2/3] v2writable: improve "num_for" API and disambiguate Eric Wong
2019-10-21 11:02   ` [PATCH 3/3] v2writable: reindex handles 3-headered monsters Eric Wong
2019-10-21 11:34     ` Eric Wong
2019-10-22  1:28     ` [PATCH 4/3] v2writable: move git->cleanup to the correct place Eric Wong
2019-10-22  1:29       ` [PATCH 5/3] v2writable: use msgmap as multi_mid queue Eric Wong
2019-10-23 18:19   ` [PUSHED] fix reindex on multiple + overlapping Message-IDs Eric Wong

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

  List information: https://public-inbox.org/README

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20191017112215.GA13175@dcvr \
    --to=e@80x24.org \
    --cc=meta@public-inbox.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
Code repositories for project(s) associated with this public inbox

	https://80x24.org/public-inbox.git

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).