New Fun Schema

From GCD
Jump to navigation Jump to search

[This was the direction the schema was going towards at the time of the server crisis in August of 2009. That crisis required a more modest migration of the database, followed by numerous gradual releases which may or may not end up at a release that looks like this schema. The documentation here is largely untouched since the crisis, and therefore quite out of date. It remains a useful reference for concepts that have not yet made it into production.]

This page documents the proposed database schema for the "New Fun" release, i.e. the first release of the new web site that will go into production. It is not set in stone, but is nearing its final form. At some point, we will declare a soft schema freeze, after which only minor changes (such as adding a column that doesn't affect table relationships or major display options) will be allowed.

Projects and Applications

Django organizes projects into applications, which may be more or less independent of one another. The project corresponds to the top-level Python package, with each application receiving a sub-package. The package name appears as a prefix on database tables. The GCD project will use the following applications (with possibly more to be added as needed):

  • data: Holds the data for display. In theory, dumping the data tables should provide all of the information that we wish to distribute publicly.
  • oi: Holds the state of the online indexing system such as reservations, approval queues, indexing credits and space for data to be edited outside of the publicly visible tables.
  • resource: Holds files such as cover scans. Also holds information about linking those scans to the public data, since that linkage is of no interest without the scans. Also holds other files like logo scans or anything else we decide to include. Some of this may be publicly accessible, but unlike the data application that is not the default assumption.
  • account: User account information, holding user preferences and all data that doesn't fit in Django's standard authorization and prefs apps.
  • migration: Space for temporary data and status tracking while we migrate fully to the new schema. This keeps transient junk out of our "proper" data design.

Storage Architecture vs Application Architecture

This version of the schema is the first to dramatically separate the representation of the data in storage (the database) from the application layer (and the UI above it). In addition to the standard database vs UI separations, this is enabled by our decision to use a separate search server to implement search. The server will organize an index of our data tables in a way that optimizes for searches, and all end-user searches will run against that index. This means that we can use a table structure that would otherwise result in very complex search queries without the performance impact of those queries. So when looking at this schema, don't think to yourself "wow that will be hard/expensive to search against", because it won't.

There are several key concepts involved in this separation that you should understand before diving into the table layout. Most notably, these involve dealing with what we have always called series and issues, and how they relate to each other.

Series Classification

One of the most important fields introduced by the New Fun schema is the series classification. This selects exactly one of several very broad categories that supplies a context for interpreting the information associated with the series. For the most part, this field tells the UI (and to a lesser extent, the application) how display things from this series, and what fields to show the indexers when they index or create something in the series. This is our primary tool for cutting down the high number of potential fields and choices into something manageable and relevant for each task at hand.

An important thing to note is that while classification helps us optimize the UI for both display and indexing, all fields are available to all classifications. You will just have to go to some sort of "advanced" option to use fields that are rare or that few indexers prefer to see for the given classification. And some fields will probably always be "advanced" in any classification.

The tentative list of classifications is (names very much subject to change):

Periodical / pamphlet series These are series of things that are generally called "issues" and are more like magazines than books.

Periodical / pamphlet one-shot These are things that look like something from the periodical / pamphlet series classification but are only ever intended to have one physical item within the series. The only difference between this classification and the series is that the UI may do some display optimization that it would *not* do for a series that only has one issue simply because the second one hasn't come out yet. For instance, the indexing UI will never ask for a publishing frequency for a one-shot. If your series lists a publishing frequency other than "published as one-shot", it should be classified as a "periodical pamphlet / series" that was cancelled after the first issue (for instance, Timely's Red Raven Comics should be a "periodical / pamphlet series" despite having only one issue). If this proves too confusing, the distinction will be dropped and all such series will automatically become "periodical / pamphlet series".

Series of books Original books or collections of periodicals, but they look more like books than magazines. Items are probably referred to as "books" or "volumes" or some similar concept. They also quite likely have their own title in addition to some sort of numeric designation.

Single book Original book or collection, but not part of a real series. Again, this is mostly for UI optimization. Note that a book may appear in two series, one in this classification (to catalogue it stand-alone) and another in the "series of books" classification (as part of a larger series). Books in this classification are very likely to have their own titles, and are less likely to have a volume or book number (or need to have "[nn]" displayed). Again, if this proves too confusing, the distinction will be dropped and all such series will automatically become "series of books".

We could take this further and separate books that are collections of periodicals from other books, but that can easily be done later if the concept proves useful. Sticking with the two choice areas that have come up over and over again on the lists (single vs true series, book vs periodical) seems like a good start. There are clear demands for these- quite recently with the book vs periodical distinction, and during the last prototype work when Jochen implemented a view of single-issue series that combined the series and issue pages. I objected at the time but the idea stuck with me and now I think there is a proper framework in which to implement it.

Items vs Issues vs Books

During the debates over handling books, it became extremely confusing to refer to things as "issues". But using "books" is arguably more accurate (as in "comic books"), it is still confusing. In the database, we will use the neutral term item for the smallest physical thing that is tracked in the database. An item is anything you can't split further without undoing its binding (even if that binding is just folding, as with the current Wednesday Comics from DC.

At the application layer, the distinction will be made, based on the classification, whether a given item is an issue, a book, or some other term we add in the future. As an item can be in more than one series, it can potentially be classified in different ways (although making something both an issue and a book is probably an error, but it's hard to predict for certain).

What's a Descriptor?

We all know that the term "issue number" is problematic. They neither universally apply to "issues", nor are they all "numbers". As has been discussed ad nauseum. Sometimes they don't even serve as faux-numbers, but behave more like names. And then there are volume numbers. Which are sometimes Roman numerals since apparently the publishing industry didn't get the memo about these great new numbers Europe nicked from the Hindus by way of the Arabs and Persians. And some volume numbers are probably completely non-numeric. And then there are people who object to calling them volume numbers and say they should be book numbers. Whether that is true depends on the series in question and whether you find the term "volume" pretentious or not, apparently. Whatever, it's all a horrible mess. Add to it the fact that these things can contradict each other depending on where they appear on the item, and you have at least six fields, three for each of the concepts (indicia, cover, and inferred a.k.a. indexer-assigned). And someone came up with some example with three different sorts of numbers. Or at least two plus a name.

So... what to do about this? At one point, I proposed calling the issue-level ones labels, but that term is already needed for certain UI-related things, and would just add more confusion on the technical side. And it didn't solve the larger problems of how to organize the fields clearly for the various puproses either. Tossing in an issue name/title field brings the number of fields up to seven, eight or nine, depending on how you look at it.

So I decided to abandon the basic notion of a fixed set of fields, and allow us to link in whatever is needed to the item. As with "item", I needed a neutral term for this. Abandoning all pretense of human comprehensibility, I appropriated the techie term "descriptor" (as in "file descriptor", if you must know). It is accurate, as these things describe the item.

Descriptor Properities

Descriptors have a few properties.

  • They have a label, by which I mean a term that a normal human would use by sticking it in front of the actual descriptor. Like "number", or "volume" or "book". Or "title" which you wouldn't actually say, but would make sense as a label in the UI or the indexing form.
  • They may apply to either a single item, or to groups of items.
    • The group may be pathological in that it only has one item in it, but the intention is usually clear from the label. Or in some cases, label plus classification (i.e. "volume" applies to groups in periodicals, but single items in series of books).

Another aspect of descriptors is that unlike many other things that are factored into their own table, we're mostly not trying to ensure that there's only one row per value and all uses link to that row. For instance, the number "1" for the issue of one series is not meaningfully the same as the number "1" for another series. The only reason to store them both as the same row is to reduce the storage and table scan time. But as far as the application is concerned, descriptor's aren't really shared.

Naming and Design Conventions

Formatting

A few notes on database table and column names vs code class and variable names:

Table names are lower case with underscores. Model class names are capitalized with no spaces or underscores. It should be obvious which model fits which table, as the words in the names should be the same. Column and variable names are all lower case with underscores, and should match exactly except for foreign keys.

All foreign keys end in _id in the database, but do not have that suffix in the code. So wherever you see something like issue_id in this table, expect to see issue in the code. This is because in the code you get an object, not just an ID.

Tables are prefixed with an application name, which does not appear in the model class name. The main app name is currently usually core, which is arguably a lousy name. I'm contemplating changing it to data but I'm not sure about that. It can't be gcd because that is the name of the overall project. Currently, the only other application is migration, which is for tables that exist only to assist in migration of data from old to new forms.

Meaningful Terms

Many of the names in the database and/or code attempt to follow some sort of systematic convention. Here are some notable ones:

_inferred Many fields will have another field with the same name plus the _inferred suffix. This other field will be a boolean that tells you whether the value of the primary field was really present on the item at hand (false) or was inferred (or just made up) by the indexer (true). In the old schema, this was mostly handled by enclosing the value in brackets, which should no longer be done. However, the UI may continue to display inferred values by enclosing them in brackets. Note that "[nn]" or "nn" for the issue number should not appear at all, even with an inferred boolean- see the section about item descriptors to understand why this is and how the display will still be able to show "[nn]" when appropriate.

_uncertain Analagous to _inferred, but indicates that the value is a guess of some sort. Both flags may be set on the same value.

sort_code always refers to a field that is *not* shown to the user, but is instead used behind the scenes to order things. In the UI, rather than viewing and entering numbers, you will use a more intuitive system such as specifying that an item follows another item, or by dragging and dropping items in a list. As auto-generated numbers, they may not make much sense if you look at them raw, except that they *will* define a clear ordering.

_links is a suffix that shows up in the code when many-to-many relationships are involved, which is very common in the new schema. In general, the set of relationship objects (rows in the join table containing data about the relationship rather than about either related object) will end in _links. This will help distinguish these sets from either set of related objects on the ends of the relationship.

label is used for fields that describe other fields. The value of label fields is generally intended to be used by the UI to label a field that does not always have an obvious name. This is hard to explain without an example, so look through the tables to see how this is used. This did not end up being a term that replaces "issue number", although that is one area where a label field is used.

Proxy Models

Typically, each database table corresponds to one model class in the code. However, several of our tables, instead of holding homogenous data, hold rows that may fall into one or more subtypes that are treated somewhat differently in the applications. Publishers and Items (on which more is explained below) come to mind. These are dealt with by defining Proxy Models, which are classes that provide different views of existing tables. In particular, they are used to make models that act like they can only access a subset of the table that is relevant for a particular application-level concept.

Indexes and Keys

Don't read much into the presence or absence of indexes in the current table definitions. Aside from all foreign keys having indexes, the presence or absence of an index for a column is mostly just left over from the old schema, plus a few nearly arbitrary decisions made as I worked on the tables. The reason for this is that we'll be using an external search server, so many columns that would normally need an index may not- we'll have to see how the search server works. Once we know that, and have a better feel for what sort of queries will actually be run, we will revisit the indexes and also consider multi-column indexes.

Table Documentation

Note: This documentation does not cover every table. It focuses on the more complex new areas of the schema. Hopefully we'll get the whole thing documented as we go along, but for the definitive schema, see the files in subversion. Email the gcd-tech list if you have questions or would like to help write documentation.

Publishers

The Publisher table is unusual in that it stores several distinct types of entities, all of which fall under the general umbrella of "Publisher". Several of them may play multiple roles, which is the primary reason for keeping them in one table. Each type of publishing entity is represented by a proxy model in the Django code. Except as noted, publishers are connected to each other via the PublisherRelationship table described further down. Before going through the fields, here is the list of proxy models and their definitions:

MasterPublisher

This is the legacy notion of a top-level publisher, linked at the series. Defined by setting is_master = 1. The nature and future role of master publishers is very much a matter of debate.

LegacyImprint

This is the legacy notion of an imprint, effectively a hodge-podge of anything and everything that did not fit as a MasterPublisher. These are linked at the series level, and will be retained only until they can be migrated to the other fields, after which they will be dropped from the application. Something is a legacy imprint if it has a relationship of type LEGACY with a MasterPublisher.

PublishingCompany

An actual corporate entity of some sort involved in publication. These are linked at the item level and defined as having is_company = 1.

Brand

A marketing brand involving a particular logo or set of closely related logos. These are linked at the item level and defined as having is_brand = 1.

Distributor

A company that distributes issues to retailers. Technically not a publisher, but the same fields apply and several publishers have served as their own distributors. These are linked at the item level and are defined as having is_distributor = 1.

The Publisher Table

data_publisher
Column Type Description Usage Notes
Core publisher data fields:
id int(11) auto_increment DB-generated primary key.
name varchar(255) NOT NULL Name of the publishing entity. Different for each type. Master publishers use names chosen for easy recognition. Brand names are based on the logos that denote them. Publishing company names should be exactly as found in the indicia that reference them (minor variations, such as "Co." vs "Company" may be harmonized). Imprints have arbitrarily complicated garbage for names that mash together all of the above. Don't even try to figure it out.
country_id foreign key to data_country The country from which the publisher operates.
year_began year Year (by cover date) of first publication.
year_ended year Year (by cover date) of last publication, if any.
url varchar(255) The URL for the publisher's website. Should be a complete URL, including the protocol (i.e. "http://")
notes longtext Space for arbitrary notes.
Type flags. A row may have more than one set to true.
is_master boolean Defines which rows appear in the MasterPublisher proxy model.
is_company boolean Defines which rows appear in the PublishingCompany proxy model.
is_brand boolean Defines which rows appear in the Brand proxy model.
is_distributor boolean Defines which rows appear in the Distributor proxy model.
Change tracking fields:
series_count int(11) The number of series associated with the publishing entity. This may have to go due to possible confusion over using a row as more than one type of publishing entity (for instance, both a master publisher and a brand).
item_count int(11) The number of items associated with the publishing entity. This may have to go due to possible confusion over using a row as more than one type of publishing entity (for instance, both a master publisher and a brand).
Change tracking fields:
created datetime Row creation date and time. Previously NULL or zero entries are now set to January 1, 1901.
modified datetime Last modified date and time. Previously NULL or zero entries are now set equal to "created".

Series

Series is fairly straightforward, except that series names are held in a separate table to allow multiple names to be assigned to a given series. See the code for details.

data_series_name
Column Type Description Usage Notes
id int(11) auto_increment DB-generated primary key.
value varchar(255) NOT NULL A name used for a series. The name may come from the cover, the indicia, common usage or any other source.
source_id Foreign key to 'data_source Explains where (indicia, cover, gcd-assigned, etc.) the name comes from.
is_primary boolean Exactly one name per series should have this set, indicating that this is the most important name. This is the name that will be put at the top of the series' page, and will be used in display formats where only one name can be shown (and it's not a search result that matched one of the non-primary names).
data_series
Column Type Description Usage Notes
Core series data fields:
id int(11) auto_increment DB-generated primary key.
year_began int(11) Year (by cover date) of first publication.
year_ended int(11) Year (by cover date) of last publication, if any.
classification_id foreign key to data_classification Used to guide the application and UI layer as to how best to display or index items in this series.
language_id foreign key to data_country The language in which the series is written. Currently there is no facility to record multi-lingual comics.
format varchar(255) legacy format field This field will be retained only until its contents have been migrated to the issue and sequence-level format fields. It's contents are a mess of at least five different sorts of data (including size, binding, paper stock, coloration and frequency of publication).
notes mediumtext Space for arbitrary notes.
tracking_notes mediumtext Notes about how a given series is continued in or continued from another series. Largely replaced by the data_series_relationship table, after most data is migrated this field will only be used for tracking notes where one end of the relationship is not and cannot be in the database (tracking from a newspaper or web comic, for instance).
has_gallery tinyint(1) True if there's a cover gallery to display for this series.
Change tracking fields:
created datetime Row creation date.
modified datetime Last modified date.

Items (a.k.a. Issues)

As recent discussions delved more into the problems created by trying to treat books as issues, the terminology got quite confusing. For that reason, and to better reflect the requirements of the project, I've renamed the "issues" to "items" at the database level. Of course, they will still be "issues" (or in some cases "books") in the UI. In general, this section has a lot of techy terminology that will not be at all visible in the UI, or even in the middle layers of the web application. This is intentional, with the aim to break faulty associations between objects as we conceive of them visually and the database-level storage requirements. Within the web application there will be classes such as "Issue" that make sense out of this very complex sub-area of the schema.

For instance, it's difficult to see from looking at the next few tables how to figure out something as conceptually simple as "what is the number of this issue?" However, in the typical case an indexer will just enter the issue number into a very obvious issue number field, or edit it there if they need to change something later. The point being, unless you really care about database design, don't get hung up on how much this doesn't look anything like a comic book issue anymore. Unless you squint really hard.

item_item_descriptor
DB Column Type Description Usage Notes
id int(11) auto_increment DB-generated primary key.
series_item_id foreign key to data_series_item The item in a particular series to which this descriptor applies.
scope enum('item', 'group') Indicates whether this descriptor applies to one item or a group of items. A single item is typically an issue or a book, or more technically, one row in the data_item table. A group is a set of items within the series (possibly the entire series, and possibly a set with only one member, but at least theoretically multiple items). Volume numbers in periodicals are the most typical example of descriptors at group scope. Indexers will never directly select or see this field as it will be handled behind the scenes. But it will, for instance, determine whether the UI should display a descriptor of type "volume" as the main number for an item (as in a trade paperback collection) or as a prefix to the main item (as in a a periodical issue).
source_id Foreign key to data_source Where the descriptor originated. Typical values include indicia, cover, title page, spine, etc. as well as "inferred", meaning the indexer made it up or got it from somewhere other than the actual item (long standing fan convention, for instance).
label_id Foreign key to data_descriptor_label The display label for this descriptor, such as "issue" or "volume".
value varchar(255) The actual issue/volume/book/ISBN/ISSN/etc. number/name/title/whatever. Note that the special no-number value "[nn]" should never be used (nor should the older "nn"). An item with no number should simply not have an entry in this table, or possibly should have an entry with a NULL value (still working on that). Or should only have an entry that is marked as "inferred", meaning that it was assigned by the indexer and does not appear on the item. The UI will translate this condition to "[nn]" or whatever else is appropriate.


data_descriptor_label
DB Column Type Description Usage Notes
id int(11) auto_increment DB-generated primary key.
name varchar(255) The human-readable display name for this kind of descriptor. Note that unlike most literal data, these labels may not all be in English and therefore may have unusual translation rules. The exact strategy here is still TBD, but part of the need for this feature is dealing with different names in various countries that don't directly translate outside of that country/language.
for_item boolean If true, may be used at the item level. i.e. descriptors where scope='item'
for_group boolean If true may be used at the group level. i.e. descriptors where scope='group'


data_source
DB Column Type Description Usage Notes
id int(11) auto_increment DB-generated primary key.
name varchar(255) The human-readable display name for the source location.
is_inferred boolean If true, the descriptor marked with this source did not come from the item at hand, and should be displayed accordingly.


data_item
DB Column Type Description Usage Notes
id int(11) auto_increment DB-generated primary key.
Basic item information
price varchar(255) Legacy price field. A decimal price followed by the ISO code, i.e. 0.10 USD See Cover Price, but superseded by data_item_price
page_count varchar(10) Page count including covers but not inserts or dust jackets. Should be decimal, but needs migration work.
editing mediumtext Editor for the item as a whole.
notes mediumtext Notes for the item.
Date fields:
publication_date varchar(255) Legacy field. Publication date of the item as given in the indicia, but fully spelled out. Publication Date, but superseded by the other publication fields in this table.
publication_year int(11) Year of publication.
publication_year_inferred boolean True if the indexer guessed or got the year from somewhere other than the item. For use with an undated item, or for a book cover-dated something like "December-January 1950" that came out in early 1950 and would therefore have an inferred first year of 1949.
publication_second_year int(11) For use with dates like "December-January 1941-1942"
publication_year_inferred boolean True if the indexer inferred the 2nd year. For instance, a book cover-dated "Winter 1941" that came out in late 1941 would have an inferred 2nd year of 1942.
publication_month_id foreign key to data_publication_month The month or vaguely month-like part of the publication date. May be two months, may be a season, may be a value like "Holiday"
publication_month_modifier enum('early', 'mid', 'late') Can be applied to the selected month. This cuts down substantially on the size of the drop down menu of "months".
publication_month_inferred boolean True if the indexer guessed or got the month from somewhere other than the item.
publication_day int(11) Day of the month of the publication date, if any.
publication_day_inferred int(11) True if the indexer guessed or got the day for somewhere other than the items. This is not meant to be used for monthlies or other publications that should not have a publication day, only for weeklies or similar items that simply omitted the day from the cover or publication information.
Format fields:
size_id Foreign key to data_size References a standard size (e.g. U.S. Golden Age) from a pre-defined set.
height decimal(10,3) Height in either inches or centimeters
width decimal(10,3) Width in either inches or centimeters
size_in_metric boolean This does not affect the standard sizes in the other table, which have their own metric flag.
interior_paper_id Foreign key to data_paper Paper stock of the interior of the publication.
cover_paper_id Foreign key to data_paper Paper stock of the cover of the publication.
binding_id Foreign key to data_binding How the publication is bound, although what that means remains a matter of some debate.
Field relating to sequences:
sequence_count int(11) Number of stories linked to this issue.
Change tracking fields:
created datetime Row creation date.
modified datetime Last modified date.


data_series_item
DB Column Type Description Usage Notes
id int(11) auto_increment DB-generated primary key.
series_id Foreign key into data_series
item_id Foreign key into data_item
sort_code int(11) Non-user-visible number that orders items within the series.

Sequences

The core_sequence table is the new name of the old stories table, using a more generic term.

core_sequence
DB Column Type Description Usage Notes
id int(11) auto_increment DB-generated primary key.
title varchar(255) The story title. In quotes if taken from the first line of the story.
title_inferred boolean True if the title was chosen by the indexer (either made up or quoted from the script). This replaces the use of brackets, which should no longer appear in the title field.
feature varchar(255) The name of the feature, if any. Should match the logo on the splash page, or possibly the bolded text in the intro blurb (for stories that don't get fancy splash pages and logos).
type_id Foreign key to data_type Type of "story", or more accurately sequence. Actual comic story vs text story vs ad vs letters page, etc.
page_count varchar(10) Number of pages. Page Count. Should be decimal but needs migration work.
notes mediumtext General notes about the sequence.
reprint_notes mediumtext Reprint information that can't be stored as links because the other end can't be put in the database. Most of the original purpose of this field has been superseded by the various reprint join tables.
Fields for creator credits:
script mediumtext Script, writer, plotter, etc.
pencils mediumtext Pencils, painted art, layouts, photographs, etc.
inks mediumtext Inks, painted art, photographs
colors mediumtext Colors, painted art, color photographs
letters mediumtext Lettering credit
editing mediumtext Editor credit for the sequence. Rarely used at the sequence level, but particularly useful for anthologies or compilations where individual sequences are drawn from original publications with various editors.
job_number varchar(25) Job Number
Fields related to content:
characters mediumtext
synopsis mediumtext
genre varchar(255)
Change tracking fields:
created datetime Row creation date.
modified datetime Last modified date.