* [PATCH] extindex: guard against buggy unrefs @ 2021-10-14 5:49 Eric Wong 2021-10-14 6:06 ` [PATCH v2] " Eric Wong 0 siblings, 1 reply; 4+ messages in thread From: Eric Wong @ 2021-10-14 5:49 UTC (permalink / raw) To: meta I noticed some unref messages which shouldn't have been happening, but they were. Which is troubling. So add a guard around an unref path until we can get to the bottom of this. --- lib/PublicInbox/ExtSearchIdx.pm | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/lib/PublicInbox/ExtSearchIdx.pm b/lib/PublicInbox/ExtSearchIdx.pm index 750ced5c..363e5665 100644 --- a/lib/PublicInbox/ExtSearchIdx.pm +++ b/lib/PublicInbox/ExtSearchIdx.pm @@ -18,6 +18,7 @@ use strict; use v5.10.1; use parent qw(PublicInbox::ExtSearch PublicInbox::Lock); use Carp qw(croak carp); +use Scalar::Util qw(blessed); use Sys::Hostname qw(hostname); use POSIX qw(strftime); use File::Glob qw(bsd_glob GLOB_NOSORT); @@ -143,6 +144,14 @@ sub _unref_doc ($$$$$;$) { $smsg = $docid; $docid = $smsg->{num}; } + if (defined($oidbin) && defined($xnum) && blessed($ibx) $ibx->over) { + my $smsg = $ibx->over->get_art($xnum); + if ($smsg && pack('H*', $smsg->{blob}) eq $oidbin) { + carp("BUG: (non-fatal) ".$ibx->eidx_key. + " #$xnum $smsg->{blob} still valid"); + return; + } + } my $s = 'DELETE FROM xref3 WHERE oidbin = ?'; $s .= ' AND ibx_id = ?' if defined($ibx); $s .= ' AND xnum = ?' if defined($xnum); ^ permalink raw reply related [flat|nested] 4+ messages in thread
* [PATCH v2] extindex: guard against buggy unrefs 2021-10-14 5:49 [PATCH] extindex: guard against buggy unrefs Eric Wong @ 2021-10-14 6:06 ` Eric Wong 2021-10-14 21:09 ` Eric Wong 0 siblings, 1 reply; 4+ messages in thread From: Eric Wong @ 2021-10-14 6:06 UTC (permalink / raw) To: meta I noticed some unref messages which shouldn't have been happening, but they were. Which is troubling. So add a guard around an unref path until we can get to the bottom of this. --- v2: added missing '&&', v1 didn't even compile :x lib/PublicInbox/ExtSearchIdx.pm | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/lib/PublicInbox/ExtSearchIdx.pm b/lib/PublicInbox/ExtSearchIdx.pm index 750ced5c..3d7a6e7d 100644 --- a/lib/PublicInbox/ExtSearchIdx.pm +++ b/lib/PublicInbox/ExtSearchIdx.pm @@ -18,6 +18,7 @@ use strict; use v5.10.1; use parent qw(PublicInbox::ExtSearch PublicInbox::Lock); use Carp qw(croak carp); +use Scalar::Util qw(blessed); use Sys::Hostname qw(hostname); use POSIX qw(strftime); use File::Glob qw(bsd_glob GLOB_NOSORT); @@ -143,6 +144,14 @@ sub _unref_doc ($$$$$;$) { $smsg = $docid; $docid = $smsg->{num}; } + if (defined($oidbin) && defined($xnum) && blessed($ibx) && $ibx->over) { + my $smsg = $ibx->over->get_art($xnum); + if ($smsg && pack('H*', $smsg->{blob}) eq $oidbin) { + carp("BUG: (non-fatal) ".$ibx->eidx_key. + " #$xnum $smsg->{blob} still valid"); + return; + } + } my $s = 'DELETE FROM xref3 WHERE oidbin = ?'; $s .= ' AND ibx_id = ?' if defined($ibx); $s .= ' AND xnum = ?' if defined($xnum); ^ permalink raw reply related [flat|nested] 4+ messages in thread
* Re: [PATCH v2] extindex: guard against buggy unrefs 2021-10-14 6:06 ` [PATCH v2] " Eric Wong @ 2021-10-14 21:09 ` Eric Wong 2021-10-16 1:48 ` [PATCH] extindex: avoid triggering a buggy unref Eric Wong 0 siblings, 1 reply; 4+ messages in thread From: Eric Wong @ 2021-10-14 21:09 UTC (permalink / raw) To: meta Eric Wong <e@80x24.org> wrote: > I noticed some unref messages which shouldn't have been > happening, but they were. Which is troubling. So add > a guard around an unref path until we can get to the bottom > of this. And now I'm not seeing it... Heisenbugs :< ^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH] extindex: avoid triggering a buggy unref 2021-10-14 21:09 ` Eric Wong @ 2021-10-16 1:48 ` Eric Wong 0 siblings, 0 replies; 4+ messages in thread From: Eric Wong @ 2021-10-16 1:48 UTC (permalink / raw) To: meta Eric Wong <e@80x24.org> wrote: > Eric Wong <e@80x24.org> wrote: > > I noticed some unref messages which shouldn't have been > > happening, but they were. Which is troubling. So add > > a guard around an unref path until we can get to the bottom > > of this. > > And now I'm not seeing it... Heisenbugs :< OK, it worked as intended and help me find a bug \o/ -----------8<---------- Subject: [PATCH] extindex: avoid triggering a buggy unref We can't attempt to unref messages beyond the highwater mark of an inbox. This bugfix was found by commit c485036d0b1ce7ed (extindex: guard against buggy unrefs, 2021-10-14), which actually did its intended job and guarded against a buggy unref. The window for the bug is when one process is doing -index against an inbox, -extdindex --reindex is working on the same inbox. --- lib/PublicInbox/ExtSearchIdx.pm | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-) diff --git a/lib/PublicInbox/ExtSearchIdx.pm b/lib/PublicInbox/ExtSearchIdx.pm index a08a9451..69d048fb 100644 --- a/lib/PublicInbox/ExtSearchIdx.pm +++ b/lib/PublicInbox/ExtSearchIdx.pm @@ -839,7 +839,7 @@ sub _unref_stale_range ($$$) { $r = $sync->{self}->{oidx}->dbh->selectall_arrayref( <<EOS, undef, $ibx->{-ibx_id}); SELECT docid,xnum,oidbin FROM xref3 -WHERE ibx_id = ? AND xnum $lt_or_gt LIMIT $lim +WHERE ibx_id = ? AND $lt_or_gt LIMIT $lim EOS return if $sync->{quit}; for (@$r) { # hopefully rare, not worth optimizing: @@ -859,7 +859,7 @@ sub _reindex_check_ibx ($$$) { my $opt = { limit => $slice }; my ($beg, $end) = (1, $slice); my $err = sync_inbox($self, $sync, $ibx) and return; - my $max = $ibx->over->max; + my $max = $ibx->mm->num_highwater; $end = $max if $end > $max; # first, check if we missed any messages in target $ibx @@ -869,7 +869,7 @@ sub _reindex_check_ibx ($$$) { local $sync->{-regen_fmt} = "$ekey checking %u/$max\n"; ${$sync->{nr}} = 0; my $fast = $sync->{-opt}->{fast}; - my $dsu; # _unref_stale_range (< $lo) called + my $usr; # _unref_stale_range (< $lo) called my ($lo, $hi); while (scalar(@{$msgs = $ibx->over->query_xover($beg, $end, $opt)})) { ${$sync->{nr}} = $beg; @@ -880,7 +880,7 @@ sub _reindex_check_ibx ($$$) { reindex_checkpoint($self, $sync); # release lock } ($lo, $hi) = ($msgs->[0]->{num}, $msgs->[-1]->{num}); - $dsu //= _unref_stale_range($sync, $ibx, "< $lo"); + $usr //= _unref_stale_range($sync, $ibx, "xnum < $lo"); my $x3a = $self->{oidx}->dbh->selectall_arrayref( <<"", undef, $ibx_id, $lo, $hi); SELECT xnum,oidbin,docid FROM xref3 WHERE @@ -921,7 +921,8 @@ ibx_id = ? AND xnum >= ? AND xnum <= ? } } } - _unref_stale_range($sync, $ibx, "> $hi") if defined($hi); + defined($hi) and ($hi < $max) and + _unref_stale_range($sync, $ibx, "xnum > $hi AND xnum <= $max"); } sub _reindex_inbox ($$$) { ^ permalink raw reply related [flat|nested] 4+ messages in thread
end of thread, other threads:[~2021-10-16 1:48 UTC | newest] Thread overview: 4+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2021-10-14 5:49 [PATCH] extindex: guard against buggy unrefs Eric Wong 2021-10-14 6:06 ` [PATCH v2] " Eric Wong 2021-10-14 21:09 ` Eric Wong 2021-10-16 1:48 ` [PATCH] extindex: avoid triggering a buggy unref Eric Wong
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).