[PATCH 2/3] migrations: Don't fail if it's not possible to rehome a patch
dja at axtens.net
Thu Mar 5 11:44:16 AEDT 2020
Aaah this is terrifying. I now wish I had reviewed this code much more
closely, and I am now super nervous about releasing this on the world.
Stephen Finucane <stephen at that.guru> writes:
> Migration 0039 attempts to move patches that have ended up in an
> arbitrary series due to race conditions into the correct series.
> However, because we weren't previously considering versions when
> choosing a series, this might not be possible resulting in the issues
> seen in #340 . If this happens, we could try rehome _those_ patches
> but that's really complicated and requires bringing in a whole load of
> parsing functionality from 'patchwork.parser' so we can e.g. rebuild the
> list of references from the headers. Instead, print a big warning to the
> admin with information about how they could fix the issue and then just
> remove the offending references. This means we're still left with
> orphaned series but these could be fixed manually, if necessary.
>  https://github.com/getpatchwork/patchwork/issues/340
> Signed-off-by: Stephen Finucane <stephen at that.guru>
> Closes: #340
> .../0039_unique_series_references.py | 35 +++++++++++++++++--
> 1 file changed, 32 insertions(+), 3 deletions(-)
> diff --git a/patchwork/migrations/0039_unique_series_references.py b/patchwork/migrations/0039_unique_series_references.py
> index 99b10fcc..c8ac0dc6 100644
> --- a/patchwork/migrations/0039_unique_series_references.py
> +++ b/patchwork/migrations/0039_unique_series_references.py
> @@ -1,6 +1,8 @@
> from django.db import connection, migrations, models
> from django.db.models import Count
> +from django.db.utils import IntegrityError
> import django.db.models.deletion
> +from django.utils import six
> def merge_duplicate_series(apps, schema_editor):
> @@ -49,14 +51,41 @@ def merge_duplicate_series(apps, schema_editor):
> if series_ref == chosen_ref:
> + has_conflict = False
> # update the patches to point to our chosen series instead, on the
> # assumption that all other metadata is correct
> for patch in Patch.objects.filter(series=series_ref.series):
> patch.series = chosen_ref.series
> - patch.save()
> + try:
> + patch.save()
> + except IntegrityError as exc:
> + has_conflict = True
> + print(
> + "We attempted to merge patch '%d' into the correct "
> + "series, '%d', but it appears there is already a "
> + "patch with the same number in this series. We have "
> + "deleted the series references for the bad series, "
> + "'%d', but you may wish to do further manual work to "
> + "resolve this issue. For more information, refer to "
> + "https://git.io/JvVvV. Error: %s" % (
Should we use the full canonical url rather than a link shortener that
might die at some point?
Also, given that this is potentially destructive, would we be better to
just leave old data alone? Or at least give admins the option to leave
old data alone? The value of correct series information ages out pretty
> + patch.id,
> + chosen_ref.series.id,
> + series_ref.series.id,
> + six.text_type(exc),
> + )
> + )
> - # delete the other series (which will delete the series ref)
> - series_ref.series.delete()
> + if not has_conflict:
> + # assuming there has been no conflict and all patches have been
> + # moved to a new series, delete the other series (which will
> + # delete the series ref)
> + series_ref.series.delete()
> + else:
> + # otherwise just delete the series references and keep the old
> + # series for later cleanup
> + for series_ref in series_ref.series.references:
> + series_ref.delete()
> def copy_project_field(apps, schema_editor):
> Patchwork mailing list
> Patchwork at lists.ozlabs.org
More information about the Patchwork