bug#71160: --fallback and (fallback #t) do not apply when receiving a corrupt nar

bug#71160: --fallback and (fallback #t) do not apply when receiving a corrupt nar

[Richard Sent]

Hi Guix!

As part of building linux-libre-arm64-generic, Guix tries substituting
linux-libre-arm64-generic.tar.xz. Unfortunately it looks like a corrupt
nar snuck into bordeaux.

--8<---------------cut here---------------start------------->8---
root@lifeline ~# guix build linux-libre-arm64-generic --system=aarch64-linux --fallback
The following derivations will be built:
  /gnu/store/9626zaczwl5x4ypxmmdklvkclqx2dlpi-linux-libre-arm64-generic-6.8.10.drv
  /gnu/store/1wi10rg7236ck8k5vdrdfap5l7a9s9z0-linux-libre-6.8.10-guix.tar.xz.drv
199.7 MB will be downloaded:
# Snip
  /gnu/store/y813phs2n9xnb7zbcr07g0j9509bzbsb-linux-libre-6.8.10-guix.tar.xz
substituting /gnu/store/0jfsx4hljddyand45z7i77ynpvr0mhb5-module-import-compiled...
downloading from https://bordeaux.guix.gnu.org/nar/lzip/0jfsx4hljddyand45z7i77ynpvr0mhb5-module-import-compiled ...
 module-import-compiled  171KiB                                                                                     676KiB/s 00:00 ▕██████████████████▏ 100.0%

substituting /gnu/store/y813phs2n9xnb7zbcr07g0j9509bzbsb-linux-libre-6.8.10-guix.tar.xz...
downloading from https://bordeaux.guix.gnu.org/nar/none/y813phs2n9xnb7zbcr07g0j9509bzbsb-linux-libre-6.8.10-guix.tar.xz ...
 linux-libre-6.8.10-guix.tar.xz  136.5MiB                                                                          20.5MiB/s 00:06 ▕█████████████████▌▏  97.6%guix substitute: error: corrupt input while restoring '/gnu/store/y813phs2n9xnb7zbcr07g0j9509bzbsb-linux-libre-6.8.10-guix.tar.xz' from #<input: string 7f1190966ee0>
substitution of /gnu/store/y813phs2n9xnb7zbcr07g0j9509bzbsb-linux-libre-6.8.10-guix.tar.xz failed
guix build: error: corrupt input while restoring archive from #<closed: file 7fea23dffaf0>
root@lifeline ~# 
--8<---------------cut here---------------end--------------->8---

Trying to pass --fallback on the command line has no effect, even though
both the documentation and [1] imply that should work.

Similarly with Cuirass, I have a spec that tries to build an
operating-system with a linux-libre-arm64-generic kernel. Even though
cuirass-configuration has (fallback #t), no attempt is made to recover
by building the derivation locally.

--8<---------------cut here---------------start------------->8---
;; In config
(define %rsent-cuirass-service
  (service cuirass-service-type
           (cuirass-configuration
            (specifications %rsent-cuirass-specs)
            (host "0.0.0.0")
            (fallback? #t))))

;; From /var/log/cuirass.log
2024-05-23 22:23:36 Uncaught exception in task:
2024-05-23 22:23:36 In fibers.scm:
2024-05-23 22:23:36     172:8  8 (_)
2024-05-23 22:23:36 In ice-9/boot-9.scm:
2024-05-23 22:23:36   1752:10  7 (with-exception-handler _ _ #:unwind? _ # _)
2024-05-23 22:23:36 In guix/store.scm:
2024-05-23 22:23:36    684:37  6 (thunk)
2024-05-23 22:23:36 In cuirass/base.scm:
2024-05-23 22:23:36    421:14  5 (_ _)
2024-05-23 22:23:36    267:10  4 (spawn-builds #<store-connection 256.100 7ff1213670f0> _ ?)
2024-05-23 22:23:36 In ice-9/boot-9.scm:
2024-05-23 22:23:36   1752:10  3 (with-exception-handler _ _ #:unwind? _ # _)
2024-05-23 22:23:36   1685:16  2 (raise-exception _ #:continuable? _)
2024-05-23 22:23:36   1683:16  1 (raise-exception _ #:continuable? _)
2024-05-23 22:23:36   1685:16  0 (raise-exception _ #:continuable? _)
2024-05-23 22:23:36 ice-9/boot-9.scm:1685:16: In procedure raise-exception:
2024-05-23 22:23:36 ERROR:
2024-05-23 22:23:36   1. &nar-error:
2024-05-23 22:23:36       file: #f
2024-05-23 22:23:36       port: #<closed: file 7ff1213615b0>
--8<---------------cut here---------------end--------------->8---

I'm no rocket scientist, but that error looks very similar to the error
found when building from the CLI. Given that cuirass-configuration has
fallback #t, it should be recoverable.

Possibly related: [2] and [3]

[1]: https://issues.guix.gnu.org/23103
[2]: https://issues.guix.gnu.org/55820
[3]: Guix 3f59fd6d114548480c719d4b8f8509bdf3e8dcca

-- 
Take it easy,
Richard Sent
Making my computer weirder one commit at a time.

bug#71160: --fallback and (fallback #t) do not apply when receiving a corrupt nar

[Richard Sent]

Richard Sent <richard@freakingpenguin.com> writes:

> Trying to pass --fallback on the command line has no effect, even though
> both the documentation and [1] imply that should work.

This issue might have some spiciness to it. I have two machines with
identical guix commits and --fallback works on one but not the other.

--8<---------------cut here---------------start------------->8---
# Failing machine:
root@lifeline ~# guix describe
Generation 2    May 25 2024 21:03:46    (current)
  guix 94c8cec
    repository URL: https://git.savannah.gnu.org/git/guix.git
    branch: master
    commit: 94c8cec99969fe9f65777637fde1f05e1c576a3f

# Good machine:
Generation 3	May 25 2024 21:58:15	(current)
  guix 94c8cec
    repository URL: https://git.savannah.gnu.org/git/guix.git
    branch: master
    commit: 94c8cec99969fe9f65777637fde1f05e1c576a3f
--8<---------------cut here---------------end--------------->8---

On the failing machine, I get an error like this:

--8<---------------cut here---------------start------------->8---
root@lifeline ~# guix build linux-libre-arm64-generic --system=aarch64-linux --fallback
# snip
downloading from https://bordeaux.guix.gnu.org/nar/none/y813phs2n9xnb7zbcr07g0j9509bzbsb-linux-libre-6.8.10-guix.tar.xz ...
 linux-libre-6.8.10-guix.tar.xz  136.5MiB                                                                          19.0MiB/s 00:07 ▕█████████████████▉▏  99.9%guix substitute: error: corrupt input while restoring '/gnu/store/y813phs2n9xnb7zbcr07g0j9509bzbsb-linux-libre-6.8.10-guix.tar.xz' from #<input: string 7f558e7233f0>
substitution of /gnu/store/y813phs2n9xnb7zbcr07g0j9509bzbsb-linux-libre-6.8.10-guix.tar.xz failed
guix build: error: corrupt input while restoring archive from #<closed: file 7f51c615b380>
root@lifeline ~# logout
--8<---------------cut here---------------end--------------->8---

whereas on the good machine:

--8<---------------cut here---------------start------------->8---
root@gibraltar ~# guix build linux-libre-arm64-generic --system=aarch64-linux --fallback
# snip
downloading from https://bordeaux.guix.gnu.org/nar/none/y813phs2n9xnb7zbcr07g0j9509bzbsb-linux-libre-6.8.10-guix.tar.xz ...
 linux-libre-6.8.10-guix.tar.xz  136.5MiB                                                                          19.6MiB/s 00:07 ▕█████████████████▊▏  98.8%guix substitute: error: corrupt input while restoring '/gnu/store/y813phs2n9xnb7zbcr07g0j9509bzbsb-linux-libre-6.8.10-guix.tar.xz' from #<input: string 7f7b2a223b60>
substitution of /gnu/store/y813phs2n9xnb7zbcr07g0j9509bzbsb-linux-libre-6.8.10-guix.tar.xz failed
building /gnu/store/ny56fdcig9cd9bd3pssmlraz2c1q10q8-linux-libre-6.8.10-guix.tar.xz.drv...
--8<---------------cut here---------------end--------------->8---

I thought perhaps there was some hyper-odd race condition going on here
(lifeline is consistently at a higher percent than gibraltar when the
error is detected), but I just had a outlier that seems to disprove
that, where lifeline had the same error with a 97.2% download progress
bar.

-- 
Take it easy,
Richard Sent
Making my computer weirder one commit at a time.