Arch Linux User Repository

Upgrading to postgresql-17 is awkward because of pgvecto.rs if you have been using 0.2.0 as provided by AUR package pgvecto.rs-immich. If you have been using 0.3.0 as provided by AUR package pgvecto.rs-immich-bin then these extra steps are likely not required and I suggest the wiki instructions for upgrades for major version increments of postgresql are followed.

If you would like to stick at postgresql-16 please add the following to pacman.conf:

IgnorePkgs = postgresql postgresql-libs

Following the manual dump and reload in the wiki does appear to work, with a few extra steps to consider the pgvecto.rs upgrade.

Starting point is postgresql-old-upgrade installed with pgvector.rs-0.2.0. Symlink the installation outputs of pgvecto.rs into the directories used by the parallel installation of postgresql-old-upgrade.

ln -s /usr/share/postgresql/extension/vectors.control /opt/pgsql-16/share/extension/vectors.control
ln -s /usr/share/postgresql/extension/vectors--0.2.0.sql /opt/pgsql-16/share/extension/vectors--0.2.0.sql
ln -s /usr/lib/postgresql/vectors.so /opt/pgsql-16/lib/vectors.so

Then perform the database start (/opt/psql-16/bin/pg_ctl .. start), "dump" (pg_dumpall...) and database stop (/opt/psql-16/bin/pg_ctl .. start) steps in the wiki.

Upgrade to postgresql-17 and pgvecto.rs-0.3.0 (provided by pgvector.rs-immich-bin). Clear out /var/lib/postgresql/data then initialise the database (initdb...) as per the wiki.

Start the service using systemctl start postgresql and install the pgvecto.rs using the following command:

psql -c 'ALTER SYSTEM SET shared_preload_libraries = "vectors.so"'

Finally restart the postgresql service and perform the reload (psql...) steps in the wiki.

EDITED: Various clarifications

From v1.111.0, this package does not build with nodejs < v20.

That's an option as well. Any clue how to do that properly without breaking everything?

I've installed at least python-insightface, python-opencv, python-tokenizers, onnxruntime, python-aiocache, python-python-multipart although I have a lot of python stuff installed already and haven't tried in a clean environment - check machine-learning/pyproject.toml - I think everything is packaged other than opencv-python-headless which I think python-opencv can be used instead.

I've modified the machine-learning.service file to run /usr/bin/gunicorn rather than /opt/immich-machine-learning/venv/bin/gunicorn and the PKGBUILD to not run poetry install. I then changed SUPPORTED_PROVIDERS to ["CPUExecutionProvider"] in /opt/immich-machine-learning/app/models/constants.py as the arch packaged one has cuda support and I only have CPU.

So to have 1.123 built correctly temporarily, you will need to : - install python312 from AUR (in parallel to whichever version of python you use)

• upgrade the pgvecto.rs-immich-bin package
• follow the instructions on https://docs.vectorchord.ai/admin/upgrading.html to finish the upgrade.

Note that SELECT pgvectors_upgrade(); returned an error to me (function not found) but the upgrade seemed to work anyway. If you don't do the upgrade, immich-server will fail with the pgvectors not up to date message in your syslog.

After this, change line 108 of the PKGBUILD to

    python3.12 -m venv "${srcdir}/venv"

instead of

    python -m venv "${srcdir}/venv"

It should then compile and be installable. Install the immich-server package that's created.

That's an option as well. Any clue how to do that properly without breaking everything?

The python-onnx upstream maintainers don't seem to be in a rush to get their 1.18 version out (which supports python 3.13) despite the fact that 3.13 is out since october, replacing the onnx with onnx-weekly through poetry is a massive pain, and we basically can't upgrade python if we have immich installed because the machine-learning hardlinks to libpython3.12.so.1.0.

I'm trying to get python312 from AUR installed as well, not sure it will help much.

EDIT: installing python312 from AUR in parallel does allow immich-machine-learning to run for version 1.121. Trying to upgrade to v1.123 to see what's what. EDIT2: yeah forgot that you can't build v1.123 because of the python-onnx dep. Trying to update the pkgbuild to temporarily link to python312.

there are existing packages for the machine learning dependencies like python-insightface, python-opencv, python-tokenizers, onnxruntime - these seems to work ok for me so perhaps can be considered instead of the poetry install

I've tried a compilation of immich after uprading all packages of immich-machine-learning to their latest revision.

With scipy 1.14.1, it compiles but there's still python-onnx that currently fails (v1.18 is unreleased but will support python 3.13).

This is a waiting game for now. However, and I know this is not the right place to say this... Don't upgrade your python. Immich-machine-learning will fail if you install python-3.13.

system-wide downgrade to 2024-11-15 works fine for me. FWIW, you can use https://aur.archlinux.org/packages/pacdate to do that.

Did a system wide downgrade to 2024-11-15, and everything compiled and working as it should now. Gonna leave it there for now.

I've confirmed python 3.13 is the culprit for the failing builds. Downgrading python,python-build,python-cachecontrol,python-cleo, python-dulwich,python-fastjsonschema,python-filelock,python-idna, python-installer,python-lark-parser,python-msgpack,python-packaging,python-pkginfo,python-platformdirs,python-poetry,python-poetry-core,python-poetry-plugin-export python-pyproject-hooks,python-requests,python-requests-toolbelt,python-tomlkit,python-urllib3,python-virtualenv

fixed the issue and immich-server builds again

Getting build errors with 1.123.0-1.

../meson.build:80:0: ERROR: Unknown compiler(s): [['gfortran'], ['flang-new'], ['flang'], ['nvfortran'], ['pgfortran'], ['ifort'], ['ifx'], ['g95']]       

 at /usr/lib/python3.13/site-packages/poetry/installation/chef.py:164 in _prepare
      160│ 
      161│                 error = ChefBuildError("\n\n".join(message_parts))
      162│ 
      163│             if error is not None:
    → 164│                 raise error from None
      165│ 
      166│             return path
      167│ 
      168│     def _prepare_sdist(self, archive: Path, destination: Path | None = None) -> Path:

Note: This error originates from the build backend, and is likely not a problem with poetry but with scipy (1.11.4) not supporting PEP 517 builds. You can verify this by running 'pip wheel --no-cache-dir --use-pep517 "scipy (==1.11.4)"'.

Anyone else running into this?

tried to do a build on a fresh VM as well, same result.

My current build wont run as its not compatible with python 3.13. A recompile will fix it, but can't compile.

EDIT: Was able to get the missing compiler error to go away by installing gcc-fortran openblas pkgconf. However the build still fails:

  FAILED: scipy/special/cython_special.cpython-313-x86_64-linux-gnu.so.p/cython_special.c 
  /tmp/tmppc2bluel/.venv/bin/cython -3 --fast-fail --output-file scipy/special/cython_special.cpython-313-x86_64-linux-gnu.so.p/cython_special.c --include-dir . scipy/special/cython_special.pyx

  Error compiling Cython file:
  ------------------------------------------------------------
  ...

  cdef inline number_t zpow(number_t x, double y) noexcept nogil:
      cdef np.npy_cdouble r, z
      # FIXME
      if number_t is double_complex:
          z.real = y
          ^
  ------------------------------------------------------------

  scipy/special/_complexstuff.pxd:123:9: Assignment of Python object not allowed without gil
  [545/1619] Generating 'scipy/special/_test_internal.cpython-313-x86_64-linux-gnu.so.p/_test_internal.c'
  [546/1619] Generating 'scipy/linalg/_solve_toeplitz.cpython-313-x86_64-linux-gnu.so.p/_solve_toeplitz.c'
  [547/1619] Generating 'scipy/linalg/_matfuncs_sqrtm_triu.cpython-313-x86_64-linux-gnu.so.p/_matfuncs_sqrtm_triu.c'
  [548/1619] Generating 'scipy/special/_ufuncs.cpython-313-x86_64-linux-gnu.so.p/_ufuncs.c'
  FAILED: scipy/special/_ufuncs.cpython-313-x86_64-linux-gnu.so.p/_ufuncs.c 
  /tmp/tmppc2bluel/.venv/bin/cython -3 --fast-fail --output-file scipy/special/_ufuncs.cpython-313-x86_64-linux-gnu.so.p/_ufuncs.c --include-dir . scipy/special/_ufuncs.pyx

  Error compiling Cython file:
  ------------------------------------------------------------
  ...

  cdef inline number_t zpow(number_t x, double y) noexcept nogil:
      cdef np.npy_cdouble r, z
      # FIXME
      if number_t is double_complex:
          z.real = y
          ^
  ------------------------------------------------------------

At this point I might be out of my depth to figure out a fix. Funny thing is, I get the same error when I try to compile previous versions of immich. I think some of these packages are not yet ready for 3.13

The reason /etc/nginx/sites-available/immich.conf was not in the backup is because it is a sample config, not intended to be used directly. Although I agree this doesn't really make sense since it can be used directly through a symlink in /etc/nginx/sites-enabled if the Debian/Ubuntu model of nginx configs is followed.

bjo is correct though, this model isn't commonly used on Arch. It's mentioned in the wiki but an installed package shouldn't be directing a user either way on such a decision. I thought the best solution was to move it to /usr/share/docs/immich/examples which is where it now resides at v1.121.0-2.