Page MenuHomePhabricator

Refactor fundraising database schema
Open, Needs TriagePublic

Description

The current database schema has seven problems:

  • German column names - This leads to confusion for non-German speaking developers, diminishing developer experience, making onboarding more involved and leading to subtle errors: adresstyp vs address_type, note the different amount of "d"
  • Large Object (LOB) Columns (data fields) - This makes it hard to export, query and update (purge) existing data.
  • Persistence layer breaks the boundaries of bounded contexts. Example: Both Donations and Memberships have a relation to Address Changes, but Address Change is its own bounded context and the Donation and Membership Entities (Business Objects) don't have references to Address Change. The reference is only there as a convenience, to trigger the creation of an AddressChange record when we save Donations/Memberships. Here, the framework/storage layer bleeds into the BC.
  • Overloaded Status - For donations, the payment status (waiting for payment, received payment notification from external provider), donation status (new, deleted in Fundraising Operation Center, Canceled on the Confirmation Page) and moderation status (Address contains suspicious words, donation comment contains suspicious words) and kept in one field, leading to data loss on status changes because the status has no "permutations" for the three different aspects. For memberships the situation is slightly better: Status is kept as flags in bit field on the database level, but that makes it hard to comprehend & has errors in the implementation.
  • Inefficient Export status - We use nullable date fields as indicators for a donation being exported. This makes the index bloated (we'll never query for specific dates) and slow.
  • Too many nullable values - using NULL instead of empty strings for optional or anonymizable values leads to unnecessary null checks in properly typed PHP (e.g. fullName)
  • Nullable IDs - Our repositories (mis)use the null value to determine if they need to create a new donation or update an existing one. If we wanted to be type safe, we'd need to litter our code with null checks, because the rest of the code only deals with loaded donations where we assume the ID to be always set (but don't check explicitly).

During the refactoring, these problems should be rectified, with the following constraints:

  • The FundraisingFrontend, Fundraising Operation Center and Export (SpendenDumper) must be operational at all times.
  • To enforce the separation of the bounded contexts, we want to get rid of FundraisingStore "library" dependency in the fundraising-donation, fundraising-memberships and fundraising-subscriptions code repositories. Each bounded context should manage its own database schema.
  • Entities in different bounded contexts are not allowed to reference each other (creating a dependency on entities outside the bounded context). If there is a reference on the storage level, for example to aggregate data for export, the reference should be inserted/changed at the storage level and should not show up in the entity definition.
  • We want to keep the code of our domain objects unaffected by database technology - no ORM annotations on them. Use a separate XML file instead

Acceptance Criteria:

  • All column names are in English
  • If using ENUM or ENUM-Like columns (e.g. address_type), the values are in English.
  • The data stored in the LOB is split out into tables
  • The data stored in the "status" field is split among the entities, with a separate database field for each purpose.
  • The export flags are booleans.
  • All database access to Domain entities is behind one or more interfaces, using the Repository Pattern.
  • We're still able to do database migrations across all bounded contexts.
  • Entity IDs are non-nullable. They *might* be their own classes that implement the Null Object pattern.
  • String and integer values are non-nullable and have empty strings or zero as defaults.
  • We have a clear model of the various states the data can be in, with optional relationships depending on the state:
    • Unconfirmed donations with external providers might have no payment data
    • Anonymized donations don't have donor data
    • Anonymized membership applications don't have applicant data
    • etc
  • The unit tests prove that the new repositories can transparently load and update data in the various states, without losing data or accidentally creating optional data

Implementation Notes:

  • To keep the Fundraising Operation Center and export working without huge refactorings, we could store the data redundantly, using the StranglerApplication pattern:
    1. Copy the DoctrineDonationRepository and DoctrineMembershipApplicationRepository to the FundraisingFrontend, change the Doctrine prefix to Legacy
    2. Change the DoctrineDonationRepository in fundraising-donation and fundraising-membership to map the Domain entities to database tables.
    3. In FundraisingFrontend, create DonationRepositoryWrapper and MembershipApplicationRepositoryWrapper classes that implement the repository interfaces and have the Legacy and newly created repositories as dependencies. Whenever an interface method is called, both repositories must receive the method call. There must be a "mapping table" that maps between the automatically generated IDs for the legacy and "new" entities.
  • For more information on where to put the code for ID generation, see https://matthiasnoback.nl/2018/05/when-and-where-to-determine-the-id-of-an-entity/
  • Not all data, especially the data for diaplying information to the user (e.g. confirmation page or comment list) needs the full entities. In fact passing entities is an anti-pattern. Ideally, we would pass simple value objects instead. These value objects can also be queried without the ORM, with a SQL query builder.

Possible tables/entities (to be discussed)

We don't need a 1:1 mapping between Domain object and tables. Objects (like DonorName) could be implemented as Doctrine Mapping Types.

For nullable relations, we should investigate if it makes sense to use null objects instead of null, to avoid excessive null checks.

Resources:

Related Objects

StatusSubtypeAssignedTask
OpenNone
OpenNone
OpenNone
OpenNone
ResolvedAbbanWMDE
Resolvedgabriel-wmde
OpenNone
Resolvedgabriel-wmde
ResolvedNone
ResolvedNone
ResolvedNone
ResolvedNone
Resolvedgabriel-wmde
ResolvedNone
Resolvedgabriel-wmde
ResolvedNone
DeclinedNone
OpenNone
OpenNone
OpenNone
OpenNone
OpenNone
OpenNone
OpenNone
OpenNone
ResolvedNone
OpenNone
ResolvedNone
ResolvedNone
ResolvedNone
OpenNone
OpenNone

Event Timeline

Restricted Application added a subscriber: Aklapper. · View Herald Transcript
gabriel-wmde renamed this task from Refactor database schema to Refactor fundraising database schema.Sep 6 2018, 3:30 PM