aboutsummaryrefslogtreecommitdiffstats
path: root/adenosine-pds/plan.txt
blob: c27299e20e272b74ea3350a4784126db0e754cb0 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208

PDS proof of concept:
x ipld sqlite driver importing CAR file
    => simple binary, two args
x skeleton
  x env config: DB paths, port
  x commands: serve, import, inspect
  x integration test
  x atp db wrapper (with methods)
        schema in a single .sql file
        https://docs.rs/rusqlite_migration/latest/rusqlite_migration/
        test version (in memory, per-thread)
        wrap in a mutex, unwrap and make new connection when desired
  x wrap both database in a struct with mutexes; have "get handle" helper that unlocks and returns a connection copy of the given type
  x repo store database wrapper (with methods)
  x response error handling (especially for XRPC endpoints)
x basic crypto and did:plc stuff
  x did:key read/write helpers
  x test that did:plc generated as expected
  x signature read/write helpers
- TODO: why are the multiformat keys so long in did doc?
- MST code to read and mutate tree state
    => mutation batches
    => just read the whole tree and then write the whole tree
    => check that empty tree works (eg, for account creation, and after deletes)
    => with in-memory tests
- service-level config
    domain suffixes (eg, just ".test" for now)
    account registration allowed or not
    CLI account creation (?)
    PDS signing key
- 
- implement basic non-authenticated CRUD on repository, test with CLI
    com.atproto
        createAccount
        repoDescribe
        repoGetRecord
        repoListRecords
        repoBatchWrite
            repoCreateRecord
            repoPutRecord
            repoDeleteRecord
        syncGetRoot
        syncGetRepo
        syncUpdateRepo
- single shared signing key for all users (not what I expected)
- helper web methods

    xrpc_wrap<S: Serialize>(resp: Result<S>) -> Response

    xrpc_get_atproto(endpoint: &str, req) -> Result<Value>
    xrpc_post_atproto(endpoint: &str, req) -> Result<Value>

    xrpc_wrap(xrpc_get_atproto(srv, "asdf", req))

? python test script
- sqlite schema (for application)
- write wrapper which updates MST *and* updates other tables in a transaction
- JSON schema type generation (separate crate?)
- HTTP API handler implementing many endpoints
    com.atproto
        createSession
        getAccountsConfig
        getSession
        resolveName
    app.bsky
        getHomeFeed
        getAuthorFeed
        getLikedBy
        getNotificationCount
        getNotifications
        getPostThread
        getProfile
        getRepostedBy
        getUserFollowers
        getUserFollows
        getUsersSearch
        postNotificationsSeen
        updateProfile
- did:web handler?

other utils/helpers:
- pack/unpack a repo CAR into JSON files in a directory tree (plus a commit.json with sig?)

libraries:
- `jsonschema` to validate requests and records (rich validation)
- `schemafy` to codegen serde types for records (ahead of time?)
- pretty_env_logger
- no good published crate for working with CAR files... could rip out this code?
    https://github.com/n0-computer/iroh/tree/main/iroh-car
- ??? for CBOR (de)serialization of MST, separate from the IPLD stuff?

sync option:
- `rouille` web framework
- `rusqlite` with "bundled" sqlite for datastore
- `rusqlite_migration`
- `ipfs-sqlite-block-store` and `libipld` to parse and persist repo content 

async option:
- `warp` as async HTTP service
- `sqlx` for async pooled sqlite or postgresql db
- `iroh-store` for async rocksdb IPFS blockstore

## concurrency (in warp app)

note that there isn't really any point in having this be async, given that we
just have a single shared sqlite on disk. could try `rouille` instead of
`warp`?

maybe good for remote stuff like did:web resolution?

could try using sqlx instead of rusqlite for natively-async database stuff?

for block store:
- open a single connection at startup, store in mutex
- handlers get a reference to mutex. if they need a connection, they enter a blocking thread then:
    block on the mutex, then create a new connection, unlock the mutex
    do any operations on connection synchronously
    exit the block

## system tables

account
    did (PK)
    username (UNIQUE, indexed)
    email (UNIQUE)
    password_bcrypt
    signing_key

did_doc
    did (PK)
    seen_at (timestamp)

session
    did
    jwt
    ???

repo
    did
    head_commit

record (should this exist? good for queries)
    did
    collection
    tid
    record_cid
    record_cbor (CBOR bytes? JSON?)

password_reset
    did
    token


## atp tables

what actually needs to be indexed?
- post replies (forwards and backwards
- likes (back index)
- follows (back index)
- usernames (as part of profile?)
- mentions? hashtags?

additional state
- notifications

bsky_post
    did
    tid (or timestamp from tid?)
    text
    reply_root (nullable)
    reply_parent (nullable)
    entities: JSON (?)

bsky_profile
    did
    tid
    display_name
    description
    my_state (JSON)

bsky_follow
    did
    tid
    target_did

bsky_like
    did
    tid
    target_uri
    target_cid (what is this? the commit, or record CID?)

bsky_repost
    did
    tid
    target_uri
    target_cid

bsky_notification
    did
    created_at (timestamp)
    seen (boolean)
    reason
    target_uri

TODO:
- bsky_badge (etc)
- bsky_media_embed