0
0
Fork 0
mirror of https://github.com/matrix-construct/construct synced 2024-11-09 19:41:22 +01:00
construct/include/ircd/m/dbs
2020-04-22 02:09:32 -07:00
..
dbs.h ircd:Ⓜ️:dbs::init: Copy string to save original dbpath. 2020-04-22 02:09:32 -07:00
event_column.h ircd:Ⓜ️:dbs: Split dbs unit per column; naming simplifications; major reorg. 2020-03-25 16:08:17 -07:00
event_horizon.h ircd:Ⓜ️:dbs: Split dbs unit per column; naming simplifications; major reorg. 2020-03-25 16:08:17 -07:00
event_idx.h ircd:Ⓜ️:dbs: Split dbs unit per column; naming simplifications; major reorg. 2020-03-25 16:08:17 -07:00
event_json.h ircd:Ⓜ️:dbs: Split dbs unit per column; naming simplifications; major reorg. 2020-03-25 16:08:17 -07:00
event_refs.h ircd:Ⓜ️:dbs: Split dbs unit per column; naming simplifications; major reorg. 2020-03-25 16:08:17 -07:00
event_sender.h ircd:Ⓜ️:dbs: Split dbs unit per column; naming simplifications; major reorg. 2020-03-25 16:08:17 -07:00
event_state.h ircd:Ⓜ️:dbs: Split dbs unit per column; naming simplifications; major reorg. 2020-03-25 16:08:17 -07:00
event_type.h ircd:Ⓜ️:dbs: Split dbs unit per column; naming simplifications; major reorg. 2020-03-25 16:08:17 -07:00
README.md ircd:Ⓜ️:dbs: Start a directory README; add some comments. 2019-08-20 21:23:18 -07:00
room_events.h ircd:Ⓜ️:dbs: Split dbs unit per column; naming simplifications; major reorg. 2020-03-25 16:08:17 -07:00
room_head.h ircd:Ⓜ️:dbs: Split dbs unit per column; naming simplifications; major reorg. 2020-03-25 16:08:17 -07:00
room_joined.h ircd:Ⓜ️:dbs: Split dbs unit per column; naming simplifications; major reorg. 2020-03-25 16:08:17 -07:00
room_state.h ircd:Ⓜ️:dbs: Split dbs unit per column; naming simplifications; major reorg. 2020-03-25 16:08:17 -07:00
room_state_space.h ircd:Ⓜ️:dbs: Split dbs unit per column; naming simplifications; major reorg. 2020-03-25 16:08:17 -07:00
room_type.h ircd:Ⓜ️:dbs: Split dbs unit per column; naming simplifications; major reorg. 2020-03-25 16:08:17 -07:00

Database Schema

This system provides local storage for all events and related metadata using the events database. The database is divided into several columns.

Writing to the database must only occur through the single write() call, which operates using transactions. In fact, write() itself only builds transactions and does not actually modify the database until the user commits the transaction.

Reading from the database can occur more directly by referencing the columns and using the ircd::db API's.

Note for casual developers: This is low-level API. It is highly likely what
you are looking for has a real interface somewhere in ircd::m.

Column Overview

There are two categories of columns: Direct event data, and indirect metadata.

The only data stored by the server is in the form of matrix events in rooms. No arbitrary application data is stored in the database. For example, there is no "accounts column" storing some user account information; instead this would be implemented as some matrix event with the type like ircd.account and the content containing our arbitrary data.

The only metadata stored in addition to the original event data optimizes and enhances the original event data. Again, no arbitrary application data is stored here. Everything stored here helps to facilitate the service of events inside rooms, for any reasonable purpose, which we then build the application layer on top of.

Direct columns

Direct data consists of original event JSON in addition to several direct property columns. The fundamental event JSON is stored in _event_json. Select properties from the original JSON are also stored in tuned columns (see: event_column.h).

The event_column(s) all store duplicate data from the original _event_json but limited to a single specific property. The index key is an event_idx (just like _event_json). These columns are useful for various optimizations at the cost of the additional space consumed.

When conducting a point lookup of an event property with m::get() or with keys masked in m::event::fetch::opts, these columns handle the query iff all desired properties can be satisfied from these columns; otherwise if a property is sought which does not have an active corresponding column here the _event_json is used transparently to satisfy the query.

Cache and storage details here can be tuned specific to each property. This makes reading faster and cache footprints more compact, holding much larger datasets without eviction; in addition to not disrupting the widely shared _event_json cache during a simple iteration of one property for all events on the server, etc.