[PATCH 3/4] migrations: 0043: split the data migration into chunks

Stephen Finucane stephen at that.guru
Fri Aug 13 02:55:26 AEST 2021


On Sat, 2021-07-17 at 03:19 +1000, Daniel Axtens wrote:
> Migrate 10000 rows at a time. This:
>  - provides a view on progress
>  - means replication happens in manageable chunks
>  - hopefully prevents db lockups
> 
> Signed-off-by: Daniel Axtens <dja at axtens.net>
> ---
>  .../migrations/0043_merge_patch_submission.py | 146 ++++++++++--------
>  1 file changed, 82 insertions(+), 64 deletions(-)
> 
> diff --git a/patchwork/migrations/0043_merge_patch_submission.py b/patchwork/migrations/0043_merge_patch_submission.py
> index 465e527812ba..1d072ca18230 100644
> --- a/patchwork/migrations/0043_merge_patch_submission.py
> +++ b/patchwork/migrations/0043_merge_patch_submission.py
> @@ -1,76 +1,94 @@
>  from django.conf import settings
>  from django.db import connection, migrations, models
>  import django.db.models.deletion
> +from django.db.models import Max
>  
>  import patchwork.fields
>  
> +CHUNK_SIZE = 10000
>  
>  def migrate_data(apps, schema_editor):
> -    if connection.vendor == 'postgresql':
> -        schema_editor.execute(
> -            """
> -            UPDATE patchwork_submission
> -              SET archived = patchwork_patch.archived,
> -                  commit_ref = patchwork_patch.commit_ref,
> -                  delegate_id = patchwork_patch.delegate_id,
> -                  diff = patchwork_patch.diff,
> -                  hash = patchwork_patch.hash,
> -                  number = patchwork_patch.number,
> -                  pull_url = patchwork_patch.pull_url,
> -                  related_id = patchwork_patch.related_id,
> -                  series_id = patchwork_patch.series_id,
> -                  state_id = patchwork_patch.state_id
> -            FROM patchwork_patch
> -              WHERE patchwork_submission.id = patchwork_patch.submission_ptr_id
> -            """
> -        )
> -    elif connection.vendor == 'mysql':
> -        schema_editor.execute(
> -            """
> -            UPDATE patchwork_submission, patchwork_patch
> -              SET patchwork_submission.archived = patchwork_patch.archived,
> -                  patchwork_submission.commit_ref = patchwork_patch.commit_ref,
> -                  patchwork_submission.delegate_id = patchwork_patch.delegate_id,
> -                  patchwork_submission.diff = patchwork_patch.diff,
> -                  patchwork_submission.hash = patchwork_patch.hash,
> -                  patchwork_submission.number = patchwork_patch.number,
> -                  patchwork_submission.pull_url = patchwork_patch.pull_url,
> -                  patchwork_submission.related_id = patchwork_patch.related_id,
> -                  patchwork_submission.series_id = patchwork_patch.series_id,
> -                  patchwork_submission.state_id = patchwork_patch.state_id
> -            WHERE patchwork_submission.id = patchwork_patch.submission_ptr_id
> -            """  # noqa
> -        )
> -    else:
> -        schema_editor.execute(
> -            """
> -            UPDATE patchwork_submission
> -              SET (
> -                archived, commit_ref, delegate_id, diff, hash, number,
> -                pull_url, related_id, series_id, state_id
> -              ) = (
> -                SELECT
> -                  patchwork_patch.archived,
> -                  patchwork_patch.commit_ref,
> -                  patchwork_patch.delegate_id,
> -                  patchwork_patch.diff,
> -                  patchwork_patch.hash,
> -                  patchwork_patch.number,
> -                  patchwork_patch.pull_url,
> -                  patchwork_patch.related_id,
> -                  patchwork_patch.series_id,
> -                  patchwork_patch.state_id
> +    Patch = apps.get_model('patchwork', 'patch')
> +    max_id = Patch.objects.all().aggregate(Max('submission_ptr_id'))['submission_ptr_id__max']

Any risk of a TOCTOU bug with this on busy instances? We probably want a final
check at the end unless we lock the database entirely for this operation
(perhaps that happens as-is? I actually don't know off the top of my head)

> +    old_max = 0
> +    # process CHUNK_SIZE at a time. +CHUNK_SIZE+1 is to ensure that if we get
> +    # max_id = 50000, we do a migration of ids >= 50000, < 60000 - otherwise
> +    # we would miss id=50000
> +    print("\nMigrating data - %d rows" % max_id)
> +    for new_max in range(CHUNK_SIZE, max_id+CHUNK_SIZE+1, CHUNK_SIZE):

nit: space around the '+'. I guess pep8 is disabled on this directory?

> +        print("Migrating rows >= %d, < %d (max %d)" % (old_max, new_max, max_id))

Should we use print statements or logging? I suspect the latter might be wiser.
The Django docs should have answers.


<snip rest of mostly unchanged but re-indented code>

Cheers,
Stephen



More information about the Patchwork mailing list