[PATCH] nodatacow: quiet chattr errors [was: Test failures with 1.7.0]

[Eric Wong <e@80x24.org>]

Dominique Martinet <asmadeus@codewreck.org> wrote:
> Eric Wong wrote on Wed, Dec 08, 2021 at 04:08:36AM +0000:
> > > And another reviewer fails 3 times with:
> > > > t/extsearch.t              (Wstat: 2048 Tests: 145 Failed: 8)
> > > >   Failed tests:  68-69, 75-76, 86, 98, 102, 139
> > > >   Non-zero exit status: 8
> > > > t/imapd.t                  (Wstat: 256 Tests: 186 Failed: 1)
> > > >   Failed test:  183
> > > >   Non-zero exit status: 1
> > > > t/nntpd.t                  (Wstat: 256 Tests: 110 Failed: 1)
> > > >   Failed test:  104
> > > >   Non-zero exit status: 1
> 
> (I'm that one)
> 
> So instead of giving the output of prove -bvw t/nntpd.t (I couldn't
> reproduce manually..), I'll go straight to the solution: it looks like
> there's a problem with the chattr fallback when btrfs is present on the
> machine (to disable cow, lib/PublicInbox/NDC_PP.pm)

Hi, thanks for the extra info.

Note: You might want to rm -rf ~/.cache/public-inbox/inline-c or
try a different HOME directory to get a clean repro (more on
this later).

I also forget to note, setting TAIL='tail -F' can dump daemon
output for tests (works better for GNU tail w/ inotify support).

> I had initial messages about chattr not being found; adding chattr to
> the build inputs gave another error message that chattr +C doesn't work
> on tmpfs.
> The problem here seems to be that the fallback here just checks
> /proc/self/mounts and calls chattr "just in case", saying it's ok if it
> fails, but some tests probably rely on stderr output being empty?

Yes on tests requiring stderr to be empty.  Below is a patch
which should fix it; however it should only be calling chattr on
btrfs mounts.

So I wonder if our /proc/self/mounts parsing is broken...
Can you share your contents of /proc/self/mounts?
(maybe spaces/odd chars aren't handled properly...)

You can also try:

  BTRFS_TESTDIR=/path/to/your/btrfs-mount prove -bvw t/nodatacow.t

?

> If so it looks more like a test problem than anything else to me, but
> perhaps the fallback path could be a bit more prudent in its calling of
> chattr (or simply try silencing its output?)

Agreed, the patch below silences stderr.

> For testing, being a bit forceful and removing that chattr call made all
> the tests pass for me.

> Now the $100 question is I don't know why the Inline::C version wasn't
> used; this one properly calls statfs() and does the ioctl directly only
> if required so would have worked.
> As far as I can see we're installing Inline::C in the build chroot, I
> see it in the PERL5LIB paths; and make/gcc are also available so it
> probably should work. Is there a way to check that?

For public-inbox-* stuff, Inline::C isn't enabled by default
even if installed.   Either ~/.cache/public-inbox/inline-c needs
to exist or PERL_INLINE_DIRECTORY must be set in env.

However, lei will auto-create ~/.cache/public-inbox/inline-c
if Socket::MsgHdr is missing (since lei is new w/ no existing users).
Thus running lei will enable Inline::C for public-inbox-* commands.

<snip>

Thanks again for all the extra info.  Hopefully this below is
one step towards making things work more smoothly, but I also
suspect /proc/self/mount parsing is off...

------8<-----
Subject: [PATCH] nodatacow: quiet chattr errors

Maybe this helps when chattr is missing or our /proc/self/mount
parsing is broken, somehow.

Link: https://public-inbox.org/meta/YbCPWGaJEkV6eWfo@codewreck.org/
---
 lib/PublicInbox/NDC_PP.pm | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/lib/PublicInbox/NDC_PP.pm b/lib/PublicInbox/NDC_PP.pm
index 57abccbe0625..a84682d371c8 100644
--- a/lib/PublicInbox/NDC_PP.pm
+++ b/lib/PublicInbox/NDC_PP.pm
@@ -16,11 +16,13 @@ sub nodatacow_dir ($) {
 		# weird chars are escaped as octal
 		$mnt_path =~ s/\\(0[0-9]{2})/chr(oct($1))/egs;
 		$mnt_path .= '/' unless $mnt_path =~ m!/\z!;
-		if (index($path, $mnt_path) == 0) {
-			# error goes to stderr, but non-fatal for us
-			system('chattr', '+C', $path);
-			last;
-		}
+		next if index($path, $mnt_path) < 0;
+
+		open my $olderr, '>&', \*STDERR or die "dup(STDERR): $!";
+		open STDERR, '>', '/dev/null' or die "open(/dev/null): $!";
+		system(qw(chattr +C), $path); # ignore error
+		open STDERR, '>&', $olderr or die "dup2(..,STDERR): $!";
+		last;
 	}
 }