Nixpkgs: e2fsprogs 1.45.2 transient failures

[FRidh]

Describe the bug
Since we have e2fsprogs version 1.45.2 I've noticed transient failures of certain tests when building e2fsprogs.

https://hydra.nixos.org/build/97065990

Maintainer information:

# a list of nixpkgs attributes affected by the problem
attribute: e2fsprogs
# a list of nixos modules affected by the problem
module:

[FRidh]

we now have 1.45.3 in staging-next.

[FRidh]

failures with 1.45.3: m_image_mmp m_mmp m_mmp_bad_csum m_mmp_bad_magic t_mmp_1on t_mmp_2off

[FRidh]

@vcunat what do you think, disable the tests?

[vcunat]

Probably, disable these individual tests. It would be nice to know a bit more about these failures and/or work with upstream, but I currently won't have capacity for that.

Apparently it was not an _immediate_ blocker so far (it seemed like one restart typically worked around that), but I do believe we should resolve it somehow within a couple staging-next iterations.

[lheckemann]

It seems to fail consistently on zfs for me, but work if I mount a tmpfs to /tmp.

[FRidh]

Tests are now disabled in master.

[AndersonTorres]

Sorry, but the line 23 here says tests are enabled:

https://github.com/NixOS/nixpkgs/blob/a69a6c11176d6aa96ab7c8f7308be4b95206598e/pkgs/tools/filesystems/e2fsprogs/default.nix#L53

EDIT: Sorry again, I have looked around the wrong place...

[FRidh]

I disabled only the failing tests, see the postPatch.

[FRidh]

I could have been clearer in my earlier comment ;-)

[AndersonTorres]

I tried to update the system today and it is still failing:

d_fallocate_bigalloc: fallocate sparse files and big files with bigalloc: ok
Creating test_script...
349 tests succeeded     1 tests failed
Tests failed: m_hugefile
make[1]: *** [Makefile:396: test_post] Error 1
make[1]: Leaving directory '/build/e2fsprogs-1.45.3/tests'
make: *** [Makefile:420: check-recursive] Error 1
builder for '/nix/store/xvjgrwscnjmv6kglchy0ph1ws68x0c2l-e2fsprogs-1.45.3.drv' failed with exit code 2
cannot build derivation '/nix/store/q0f7m00hc269w7vksri6b8anrcr3fd0x-libarchive-3.4.0.drv': 1 dependencies couldn't be built
cannot build derivation '/nix/store/fdxdfxspxdzr0lbms3wcgigapmrqawy4-cmake-3.15.1.drv': 1 dependencies couldn't be built
cannot build derivation '/nix/store/5lphgzwxch2a85xj1w9k7xbr2yfxxz7n-aws-sdk-cpp-1.7.90.drv': 1 dependencies couldn't be built
cannot build derivation '/nix/store/vapypxs4hb81frl2p1clzri19hwl7l9d-brotli-1.0.7.drv': 1 dependencies couldn't be built
cannot build derivation '/nix/store/acr7w1m1anm35dmi1awzgfzjcz1sqbab-nix-2.3.drv': 1 dependencies couldn't be built
error: build of '/nix/store/acr7w1m1anm35dmi1awzgfzjcz1sqbab-nix-2.3.drv' failed

[vcunat]

m_hugefile isn't in the removed set. Perhaps these depend on the FS used in the temporary dir?

[AndersonTorres]

Well, it is a possibility. My /tmp dir is an EXT2 filesystem.

[bqv]

So, I've just had it too. My /tmp dir was F2FS on one machine, TMPFS (over BTRFS) on another, then I loop mounted an EXT4 file onto the former and tried it again and now it works fine.

[tytso]

The MMP feature requires that the underlying storage device (e.g., the block device, or a file if you are using an image) that supports O_DIRECT. So for the MMP tests, it requires that /tmp is either tmpfs (for which we have a special case to skip the test) or it must be a file system which supports O_DIRECT. ext2 and btrfs doesn't support O_DIRECT.

[lheckemann]

Can we reenable the tests and only enable the ones with the particular requirements, or run them in a VM to ensure the right filesystem for them to work?

[vcunat]

@lheckemann: I assume you got confused by the post saying

Tests are now disabled in master.

as the commit 151f52bf96a only disabled the problematic (mmp) tests.

[vcunat]

Additionally running the excluded tests separately inside a VM – that sounds possible (we do NixOS VM tests already), but it would seem an overkill to me.