For the past few days I’ve been working on adding Frontier-like object database (ODB) support to my database framework.
It’s not finished yet — it doesn’t even build.
What it is (or, what it will be)
It’s hierarchical key-value storage. No schemas. Tables can contain tables, with no limit.
This implementation is the lowest level: the part that gets, sets, and deletes data from the database.
It’s application-agnostic, at this level — it doesn’t know about all of Frontier’s data types, for instance. A level on top of this will be needed for new-Frontier.
SQLite, my favorite hammer
I’m not actually writing a new database — I’m using SQLite. And that’s because I’ve been using SQLite for 15 years, and I love it and know it well, and I know how incredibly stable it is. I’m not willing to write my own thing, and I’m not willing to use a thing less mature and rock-solid than SQLite.
How it works:
The schema is pretty simple. There are tables and values.
Every table has an id. Every table (except the root table) has a parent_id that points to its parent table.
And every value has an odb_table_id that points to its parent table.
This way it’s easy to get a table’s children: it takes just two select statements.
(Both tables and values also have a name, since this is key-value storage.)
Tables and values will be cached in memory, so not every call will require a database read.
(Before you suggest I use something other than SQLite, know that I won’t change my mind on this.)
(Also, again: it’s not done yet. Doesn’t even build.)
Why I’m doing this now instead of something else
I’m using schema-less storage for feeds in Evergreen. (Articles and article status, on the other hand, are stored using a schema, in SQLite.)
Currently I’m writing a big binary plist with all the feed data, and it has to be rewritten every time a feed property changes. The writes are coalesced — but still, this isn’t great.
I’m using schema-less storage in part because of syncing systems: I don’t know, and can’t guess, what I’ll need to store. Different systems will have different requirements.
Also: I may add features later that require additional feed properties. I don’t know what those are.
I realized that what I really want for this is a feature from Frontier: hierarchical key-value storage.
Each system will gets its own database on the client. For each, I’ll create an odb table called feeds. Each feed will have its own subtable. The key will be its id (which may or may not be its URL, depending on the syncing system).
And inside each subtable I can put whatever I want, at any time, without having to change any schemas or implementations.
Example:
For the On My Mac account — not synced; reads feeds directly — we keep track of Etag headers in order to support conditional GET. So, for example, I’d want to get, set, and delete feeds.[feedID].etag.ifModifiedSince.
But with most syncing systems we get the feed content from the system itself — not by directly reading the feed. There might be some other data from the service to store: feeds.[feedID].syncToken, for instance.
