0
0
Fork 0
mirror of https://github.com/matrix-construct/construct synced 2024-11-16 15:00:51 +01:00
construct/ircd/m/dbs.cc

1112 lines
22 KiB
C++
Raw Normal View History

// Matrix Construct
//
// Copyright (C) Matrix Construct Developers, Authors & Contributors
// Copyright (C) 2016-2018 Jason Volk <jason@zemos.net>
//
// Permission to use, copy, modify, and/or distribute this software for any
// purpose with or without fee is hereby granted, provided that the above
// copyright notice and this permission notice is present in all copies. The
// full license for this software is available in the LICENSE file.
#include <ircd/m/m.h>
/// Residence of the events database instance pointer.
decltype(ircd::m::dbs::events)
ircd::m::dbs::events
{};
/// Linkage for a cache of the columns of the events database which directly
/// correspond to a property in the matrix event object. This array allows
/// for constant time access to a column the same way one can make constant
/// time access to a property in m::event.
decltype(ircd::m::dbs::event_column)
ircd::m::dbs::event_column
{};
/// Linkage for a reference to the state_node column.
decltype(ircd::m::dbs::state_node)
ircd::m::dbs::state_node
{};
/// Linkage for a reference to the room_events column
decltype(ircd::m::dbs::room_events)
ircd::m::dbs::room_events
{};
//
// init
//
/// Initializes the m::dbs subsystem; sets up the events database. Held/called
/// by m::init. Most of the extern variables in m::dbs are not ready until
/// this call completes.
ircd::m::dbs::init::init()
{
// Open the events database
events = std::make_shared<database>("events"s, ""s, desc::events);
// Cache the columns for the event tuple in order for constant time lookup
assert(event_columns == event::size());
std::array<string_view, event::size()> keys; //TODO: why did this happen?
_key_transform(event{}, begin(keys), end(keys)); //TODO: how did this happen?
for(size_t i(0); i < keys.size(); ++i)
event_column[i] = db::column
{
*events, keys[i]
};
// Cache the columns for the metadata
state_node = db::column{*events, "_state_node"};
room_events = db::index{*events, "_room_events"};
}
/// Shuts down the m::dbs subsystem; closes the events database. The extern
/// variables in m::dbs will no longer be functioning after this call.
ircd::m::dbs::init::~init()
noexcept
{
// Columns should be unrefed before DB closes
state_node = {};
room_events = {};
for(auto &column : event_column)
column = {};
// Unref DB (should close)
events = {};
}
namespace ircd::m::dbs
{
static void _index__room_events(db::txn &, const string_view &state_root, const event &);
static string_view _index_state(db::txn &, const mutable_buffer &root_out, const string_view &state_root, const event &);
static string_view _index_ephem(db::txn &, const string_view &state_root, const event &);
}
ircd::string_view
ircd::m::dbs::write(db::txn &txn,
const mutable_buffer &root_out,
const string_view &root_in,
const event &event)
{
db::txn::append
{
txn, at<"event_id"_>(event), event
};
if(defined(json::get<"state_key"_>(event)))
return _index_state(txn, root_out, root_in, event);
return _index_ephem(txn, root_in, event);
}
ircd::string_view
ircd::m::dbs::_index_ephem(db::txn &txn,
const string_view &state_root_in,
const event &event)
{
_index__room_events(txn, state_root_in, event);
return state_root_in;
}
ircd::string_view
ircd::m::dbs::_index_state(db::txn &txn,
const mutable_buffer &state_root_out,
const string_view &state_root_in,
const event &event)
try
{
const auto &type
{
at<"type"_>(event)
};
const auto &room_id
{
at<"room_id"_>(event)
};
const string_view new_root
{
state::insert(txn, state_root_out, state_root_in, event)
};
_index__room_events(txn, new_root, event);
return new_root;
}
catch(const std::exception &e)
{
log::error
{
"Failed to update state: %s", e.what()
};
throw;
}
/// Adds the entry for the room_events column into the txn.
/// You need find/create the right state_root before this.
void
ircd::m::dbs::_index__room_events(db::txn &txn,
const string_view &state_root,
const event &event)
{
const ctx::critical_assertion ca;
thread_local char buf[768];
const string_view &key
{
room_events_key(buf, at<"room_id"_>(event), at<"depth"_>(event), at<"event_id"_>(event))
};
db::txn::append
{
txn, room_events,
{
key, // key
state_root // val
}
};
}
ircd::string_view
ircd::m::dbs::state_root(const mutable_buffer &out,
const event &event)
{
return state_root(out, at<"room_id"_>(event), at<"event_id"_>(event), at<"depth"_>(event));
}
ircd::string_view
ircd::m::dbs::state_root(const mutable_buffer &out,
const id::event &event_id)
{
static constexpr auto idx
{
json::indexof<event, "room_id"_>()
};
auto &column
{
event_column.at(idx)
};
id::room::buf room_id;
column(event_id, [&room_id](const string_view &val)
{
room_id = val;
});
return state_root(out, room_id, event_id);
}
ircd::string_view
ircd::m::dbs::state_root(const mutable_buffer &out,
const id::room &room_id,
const id::event &event_id)
{
static constexpr auto idx
{
json::indexof<event, "depth"_>()
};
auto &column
{
event_column.at(idx)
};
uint64_t depth;
column(event_id, [&](const string_view &binary)
{
assert(size(binary) == sizeof(depth));
depth = byte_view<uint64_t>(binary);
});
return state_root(out, room_id, event_id, depth);
}
ircd::string_view
ircd::m::dbs::state_root(const mutable_buffer &out,
const id::room &room_id,
const id::event &event_id,
const uint64_t &depth)
{
char keybuf[768]; const auto key
{
room_events_key(keybuf, room_id, depth, event_id)
};
string_view ret;
room_events(key, [&out, &ret](const string_view &val)
{
ret = { data(out), copy(out, val) };
});
return ret;
}
bool
ircd::m::dbs::exists(const event::id &event_id)
{
static constexpr auto idx
{
json::indexof<event, "event_id"_>()
};
auto &column
{
event_column.at(idx)
};
return has(column, event_id);
}
//
// Database descriptors
//
/// State nodes are pieces of the m::state:: b-tree. The key is the hash
/// of the value, which serves as the ID of the node when referenced in
/// the tree. see: m/state.h for details.
///
const ircd::database::descriptor
ircd::m::dbs::desc::events__state_node
{
// name
"_state_node",
// explanation
R"(### developer note:
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
},
// options
{},
// comparator
{},
// prefix transform
{},
};
/// Prefix transform for the events__room_events. The prefix here is a room_id
/// and the suffix is the depth+event_id concatenation.
/// for efficient sequences
///
/// TODO: This needs The Grammar
///
const ircd::db::prefix_transform
ircd::m::dbs::desc::events__room_events__pfx
{
"_room_events",
[](const string_view &key)
{
return key.find(":::") != key.npos;
},
[](const string_view &key)
{
return rsplit(key, ":::").first;
}
};
/// Comparator for the events__room_events. The goal here is to sort the
/// events within a room by their depth from highest to lowest, so the
/// highest depth is hit first when a room is sought from this column.
///
/// TODO: This needs The Grammar
///
const ircd::db::comparator
ircd::m::dbs::desc::events__room_events__cmp
{
"_room_events",
// less
[](const string_view &a, const string_view &b)
{
static const auto &pt
{
events__room_events__pfx
};
// Extract the prefix from the keys
const string_view pre[2]
{
pt.get(a),
pt.get(b),
};
if(pre[0] != pre[1])
return pre[0] < pre[1];
// After the prefix is the depth + event_id
const string_view post[2]
{
a.substr(size(pre[0])),
b.substr(size(pre[1])),
};
if(empty(post[0]))
return true;
if(empty(post[1]))
return false;
// Now want just the depth...
const string_view depths[2]
{
between(post[0], ":::", "$"),
between(post[1], ":::", "$"),
};
// ...as machine words
const int64_t depth[2]
{
lex_cast<int64_t>(depths[0]),
lex_cast<int64_t>(depths[1]),
};
// Highest to lowest sort so highest is first
return depth[1] < depth[0];
},
// equal
[](const string_view &a, const string_view &b)
{
return a == b;
}
};
//TODO: optimize
//TODO: Needs The Gramslam
ircd::string_view
ircd::m::dbs::room_events_key(const mutable_buffer &out,
const id::room &room_id,
const uint64_t &depth,
const id::event &event_id)
{
size_t len{0};
len = strlcpy(out, room_id);
len = strlcat(out, ":::");
len = strlcat(out, lex_cast(depth));
len = strlcat(out, event_id);
return { data(out), len };
}
std::tuple<uint64_t, ircd::string_view>
ircd::m::dbs::room_events_key(const string_view &amalgam)
{
const auto depth
{
between(amalgam, ":::", "$")
};
return
{
lex_cast<uint64_t>(depth),
string_view{end(depth), end(amalgam)}
};
}
/// This column stores events in sequence in a room. Consider the following:
///
/// [room_id | depth + event_id => state_root]
///
/// The key is composed from three parts:
///
/// - `room_id` is the official prefix, bounding the sequence. That means we
/// make a blind query with just a room_id and get to the beginning of the
/// sequence, then iterate until we stop before the next room_id (upper bound).
///
/// - `depth` is the ordering. Within the sequence, all elements are ordered by
/// depth from HIGHEST TO LOWEST. The sequence will start at the highest depth.
///
/// - `event_id` is the key suffix. This column serves to sequence all events
/// within a room ordered by depth. There may be duplicate room_id|depth
/// prefixing but the event_id suffix gives the key total uniqueness.
///
/// The value is then used to store the node ID of the state tree root at this
/// event. Nodes of the state tree are stored in the state_node column. From
/// that root node the state of the room at the time of this event_id can be
/// queried.
///
/// There is one caveat here: we can't directly take a room_id and an event_id
/// and make a trivial query to find the state root, since the depth number
/// gets in the way. Rather than creating yet another column without the depth,
/// for the time being, we pay the cost of an extra query to events_depth and
/// find that missing piece to make the exact query with all three key parts.
///
const ircd::database::descriptor
ircd::m::dbs::desc::events__room_events
{
// name
"_room_events",
// explanation
R"(### developer note:
key is "!room_id$event_id"
the prefix transform is in effect. this column indexes events by
room_id offering an iterable bound of the index prefixed by room_id
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
},
// options
{},
// comparator
events__room_events__cmp,
// prefix transform
events__room_events__pfx,
};
//
// Direct column descriptors
//
const ircd::database::descriptor
ircd::m::dbs::desc::events_event_id
{
// name
"event_id",
// explanation
R"(### protocol note:
10.1
The id of event.
10.4
MUST NOT exceed 255 bytes.
### developer note:
key is event_id. This is redundant data but we have to have it for now.
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
}
};
const ircd::database::descriptor
ircd::m::dbs::desc::events_type
{
// name
"type",
// explanation
R"(### protocol note:
10.1
The type of event. This SHOULD be namespaced similar to Java package naming conventions
e.g. 'com.example.subdomain.event.type'.
10.4
MUST NOT exceed 255 bytes.
### developer note:
key is event_id
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
}
};
const ircd::database::descriptor
ircd::m::dbs::desc::events_content
{
// name
"content",
// explanation
R"(### protocol note:
10.1
The fields in this object will vary depending on the type of event. When interacting
with the REST API, this is the HTTP body.
### developer note:
Since events must not exceed 65 KB the maximum size for the content is the remaining
space after all the other fields for the event are rendered.
key is event_id
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
}
};
const ircd::database::descriptor
ircd::m::dbs::desc::events_room_id
{
// name
"room_id",
// explanation
R"(### protocol note:
10.2 (apropos room events)
Required. The ID of the room associated with this event.
10.4
MUST NOT exceed 255 bytes.
### developer note:
key is event_id
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
}
};
const ircd::database::descriptor
ircd::m::dbs::desc::events_sender
{
// name
"sender",
// explanation
R"(### protocol note:
10.2 (apropos room events)
Required. Contains the fully-qualified ID of the user who sent this event.
10.4
MUST NOT exceed 255 bytes.
### developer note:
key is event_id
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
}
};
const ircd::database::descriptor
ircd::m::dbs::desc::events_state_key
{
// name
"state_key",
// explanation
R"(### protocol note:
10.3 (apropos room state events)
A unique key which defines the overwriting semantics for this piece of room state.
This value is often a zero-length string. The presence of this key makes this event a
State Event. The key MUST NOT start with '_'.
10.4
MUST NOT exceed 255 bytes.
### developer note:
key is event_id
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
}
};
const ircd::database::descriptor
ircd::m::dbs::desc::events_origin
{
// name
"origin",
// explanation
R"(### protocol note:
FEDERATION 4.1
DNS name of homeserver that created this PDU
### developer note:
key is event_id
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
}
};
const ircd::database::descriptor
ircd::m::dbs::desc::events_origin_server_ts
{
// name
"origin_server_ts",
// explanation
R"(### protocol note:
FEDERATION 4.1
Timestamp in milliseconds on origin homeserver when this PDU was created.
### developer note:
key is event_id
value is a machine integer (binary)
TODO: consider unsigned rather than time_t because of millisecond precision
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(time_t)
}
};
const ircd::database::descriptor
ircd::m::dbs::desc::events_signatures
{
// name
"signatures",
// explanation
R"(### protocol note:
### developer note:
key is event_id
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
}
};
const ircd::database::descriptor
ircd::m::dbs::desc::events_auth_events
{
// name
"auth_events",
// explanation
R"(### protocol note:
### developer note:
key is event_id.
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
}
};
const ircd::database::descriptor
ircd::m::dbs::desc::events_depth
{
// name
"depth",
// explanation
R"(### protocol note:
### developer note:
key is event_id value is long integer
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(int64_t)
}
};
const ircd::database::descriptor
ircd::m::dbs::desc::events_hashes
{
// name
"hashes",
// explanation
R"(### protocol note:
### developer note:
key is event_id.
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
}
};
const ircd::database::descriptor
ircd::m::dbs::desc::events_membership
{
// name
"membership",
// explanation
R"(### protocol note:
### developer note:
key is event_id.
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
}
};
const ircd::database::descriptor
ircd::m::dbs::desc::events_prev_events
{
// name
"prev_events",
// explanation
R"(### protocol note:
### developer note:
key is event_id.
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
}
};
const ircd::database::descriptor
ircd::m::dbs::desc::events_prev_state
{
// name
"prev_state",
// explanation
R"(### protocol note:
### developer note:
key is event_id.
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
}
};
/*
const ircd::database::descriptor
event_id_in_sender
{
// name
"event_id in sender",
// explanation
R"(### developer note:
key is "@sender$event_id"
the prefix transform is in effect. this column indexes events by
sender offering an iterable bound of the index prefixed by sender
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
},
// options
{},
// comparator
{},
// prefix transform
event_id_in,
};
/// prefix transform for origin in
///
/// This transform expects a concatenation ending with an origin which means
/// the prefix can be the same for multiple origins; therefor we can find
/// or iterate "origin in X" where X is some repeated prefix
///
/// TODO: strings will have character conflicts. must address
const ircd::db::prefix_transform
origin_in
{
"origin in",
[](const ircd::string_view &key)
{
return has(key, ":::");
//return key.find(':') != key.npos;
},
[](const ircd::string_view &key)
{
return split(key, ":::").first;
//return rsplit(key, ':').first;
}
};
const ircd::database::descriptor
origin_in_room_id
{
// name
"origin in room_id",
// explanation
R"(### developer note:
key is "!room_id:origin"
the prefix transform is in effect. this column indexes origins in a
room_id offering an iterable bound of the index prefixed by room_id
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
},
// options
{},
// comparator - sorts from highest to lowest
{}, //ircd::db::reverse_cmp_string_view{},
// prefix transform
origin_in,
};
const ircd::database::descriptor
origin_joined_in_room_id
{
// name
"origin_joined in room_id",
// explanation
R"(### developer note:
key is "!room_id:origin"
the prefix transform is in effect. this column indexes origins in a
room_id offering an iterable bound of the index prefixed by room_id
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
},
// options
{},
// comparator - sorts from highest to lowest
{}, //ircd::db::reverse_cmp_string_view{},
// prefix transform
origin_in,
};
/// prefix transform for room_id
///
/// This transform expects a concatenation ending with a room_id which means
/// the prefix can be the same for multiple room_id's; therefor we can find
/// or iterate "room_id in X" where X is some repeated prefix
///
const ircd::db::prefix_transform room_id_in
{
"room_id in",
[](const ircd::string_view &key)
{
return key.find('!') != key.npos;
},
[](const ircd::string_view &key)
{
return rsplit(key, '!').first;
}
};
/// prefix transform for type,state_key in room_id
///
/// This transform is special for concatenating room_id with type and state_key
/// in that order with prefix being the room_id (this may change to room_id+
/// type
///
/// TODO: arbitrary type strings will have character conflicts. must address
/// TODO: with grammars.
const ircd::db::prefix_transform type_state_key_in_room_id
{
"type,state_key in room_id",
[](const ircd::string_view &key)
{
return key.find("..") != key.npos;
},
[](const ircd::string_view &key)
{
return split(key, "..").first;
}
};
const ircd::database::descriptor
event_id_for_type_state_key_in_room_id
{
// name
"event_id for type,state_key in room_id",
// explanation
R"(### developer note:
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
},
// options
{},
// comparator
{},
// prefix transform
type_state_key_in_room_id
};
const ircd::database::descriptor
prev_event_id_for_event_id_in_room_id
{
// name
"prev_event_id for event_id in room_id",
// explanation
R"(### developer note:
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
},
// options
{},
// comparator
{},
// prefix transform
event_id_in
};
/// prefix transform for event_id in room_id,type,state_key
///
/// This transform is special for concatenating room_id with type and state_key
/// and event_id in that order with prefix being the room_id,type,state_key. This
/// will index multiple event_ids with the same type,state_key in a room which
/// allows for a temporal depth to the database; event_id for type,state_key only
/// resolves to a single latest event and overwrites itself as per the room state
/// algorithm whereas this can map all of them and then allows for tracing.
///
/// TODO: arbitrary type strings will have character conflicts. must address
/// TODO: with grammars.
const ircd::db::prefix_transform
event_id_in_room_id_type_state_key
{
"event_id in room_id,type_state_key",
[](const ircd::string_view &key)
{
return has(key, '$');
},
[](const ircd::string_view &key)
{
return split(key, '$').first;
}
};
const ircd::database::descriptor
prev_event_id_for_type_state_key_event_id_in_room_id
{
// name
"prev_event_id for type,state_key,event_id in room_id",
// explanation
R"(### developer note:
)",
// typing (key, value)
{
typeid(ircd::string_view), typeid(ircd::string_view)
},
// options
{},
// comparator
{},
// prefix transform
event_id_in_room_id_type_state_key
};
*/
const ircd::database::description
ircd::m::dbs::desc::events
{
// Requirement of RocksDB/LevelDB
{ "default" },
////////
//
// These columns directly represent event fields indexed by event_id and
// the value is the actual event values. Some values may be JSON, like
// content.
//
events_auth_events,
events_content,
events_depth,
events_event_id,
events_hashes,
events_membership,
events_origin,
events_origin_server_ts,
events_prev_events,
events_prev_state,
events_room_id,
events_sender,
events_signatures,
events_state_key,
events_type,
////////
//
// These columns are metadata composed from the event data. Specifically,
// they are designed for fast sequential iterations.
//
// (state tree node id) => (state tree node)
events__state_node,
// (room_id, event_id) => (state_root)
// Sequence of all events for a room, ever.
events__room_events,
};