Re: Hydra status

[ludo@gnu.org (Ludovic Courtès)]

ludo@gnu.org (Ludovic Courtès) skribis:

> ludo@gnu.org (Ludovic Courtès) skribis:
>
>> Going to http://hydra.gnu.org/build/49569/ shows that the texlive builds
>> were actually aborted.  The “Nix error output” box shows this:
>>
>> offloading '/gnu/store/j0b550qfqaffffc1w550v0zq2nlb0sqc-texlive-2013.drv' to 'hydra.gnunet.org'...
>> @ build-remote /gnu/store/j0b550qfqaffffc1w550v0zq2nlb0sqc-texlive-2013.drv hydra.gnunet.org
>> retrieving 2 files from 'hydra.gnunet.org'...
>> Backtrace:
>> In ice-9/boot-9.scm:
>> 2401: 19 [save-module-excursion #<procedure 1acc980 at ice-9/boot-9.scm:4045:3 ()>]
>> 4050: 18 [#<procedure 1acc980 at ice-9/boot-9.scm:4045:3 ()>]
>> 1724: 17 [%start-stack load-stack ...]
>> 1729: 16 [#<procedure 1ad85d0 ()>]
>> In unknown file:
>>    ?: 15 [primitive-load "/usr/local/bin/guix"]
>> In guix/ui.scm:
>>  597: 14 [run-guix-command offload "x86_64-linux" "3600" "1" "72000"]
>> In guix/scripts/offload.scm:
>>  686: 13 [guix-offload "x86_64-linux" "3600" "1" "72000"]
>> In ice-9/boot-9.scm:
>>  157: 12 [catch srfi-34 ...]
>> In guix/scripts/offload.scm:
>>  413: 11 [transfer-and-offload # # # ...]
>>  488: 10 [retrieve-files # #]
>> In ice-9/boot-9.scm:
>>  157: 9 [catch srfi-34 ...]
>> In guix/scripts/offload.scm:
>>  497: 8 [#<procedure 262b660 at guix/scripts/offload.scm:488:11 ()>]
>> In guix/utils.scm:
>>  240: 7 [call-with-decompressed-port xz #<closed: file 0> ...]
>> In guix/nar.scm:
>>  426: 6 [loop 1 ()]
>>  256: 5 [restore "/gnu/store/guix-8uVCz3"]
>>  256: 4 [restore "/gnu/store/guix-8uVCz3/texmf-dist"]
>>  256: 3 [restore "/gnu/store/guix-8uVCz3/texmf-dist/doc"]
>>  256: 2 [restore "/gnu/store/guix-8uVCz3/texmf-dist/doc/generic"]
>>  248: 1 [restore "/gnu/store/guix-8uVCz3/texmf-dist/doc/generic/babel-swedish"]
>> In unknown file:
>>    ?: 0 [mkdir "/gnu/store/guix-8uVCz3/texmf-dist/doc/generic/babel-swedish" ...]
>>
>> ERROR: In procedure mkdir:
>> ERROR: In procedure mkdir: No such file or directory
>> lsh: io.c: write failed (errno = 32): Broken pipe
>> lsh: channel.c: stdout: I/O error, Broken pipe
>> lsh: Ignoring data on channel which is closing
>> builder for `/gnu/store/j0b550qfqaffffc1w550v0zq2nlb0sqc-texlive-2013.drv' failed with exit code 1
>> @ hook-failed /gnu/store/j0b550qfqaffffc1w550v0zq2nlb0sqc-texlive-2013.drv - 256 builder for `/gnu/store/j0b550qfqaffffc1w550v0zq2nlb0sqc-texlive-2013.drv' failed with exit code 1
>> error: build of `/gnu/store/j0b550qfqaffffc1w550v0zq2nlb0sqc-texlive-2013.drv' failed
>>
>> So the problem occurred while retrieving and restoring the build result
>> (the log shows just these two lines because apparently, the result was
>> already available on the build machine hydra.gnunet.org–which is no
>> surprise given that /gnu is 1.5T on that machine, vs. 400G on
>> hydra.gnu.org.)
>>
>> The ‘mkdir’ error here suggests that the temporary place were we are
>> restoring, /gnu/store/guix-8uVCz3, has been removed right in the middle
>> of the process by the GC.
>>
>> I believe this is fixed by 9132b9b.
>
> Actually no (I’ve seen other occurrences of that in the meantime.)
>
> 9132b9b had a thinko, which is (hopefully!) fixed by 6071b55.  I’ve just
> installed it on hydra.gnu.org.

There was another problem with that “fix”, namely that the daemon
ignores normal GC roots pointing to “abnormal” store files like
/gnu/store/guix-8uVCz3.

This led to the (now abandoned) proposal at
<https://lists.gnu.org/archive/html/guix-devel/2014-04/msg00198.html>.
I finally came up with a simpler solution in commit 50db7d8: a
connection to the daemon is opened for the sake of holding a per-process
GC root to that temporary store file.

Ludo’.