Part I — Preparing the ground

02 · Architecture: the whole tree

The Baobáxia baobab: roots, trunk, branches and canopy

Principles

  1. The file system is the source of truth. Each saber is a saber.bbx file (YAML + markdown) versioned in git; binaries go through git-annex. The SQLite index is a cache: delete it and run bbx reindex.
  2. Offline first. Each mucua works on its own. Network is a bonus.
  3. A federation of equals. There is no central mucua. Mucuas run the same code; only the configuration changes.
  4. Modular through environment variables. BBX_AUTH_MODE turns on cryptography; BBX_FED_XMPP/BBX_FED_AP turn on federation; BBX_ENRIQUECER turns on OCR and Whisper.

Backend layers

flowchart TB
    subgraph Entradas["Entry points (branches)"]
        API["api/v3/ — FastAPI<br/>~150 routes"]
        CLI["cli.py — bbx"]
        BOT["federacao/xmpp_bot.py"]
        TIMERS["systemd timers<br/>bbx sincronizar · bbx enriquecer"]
    end
    subgraph Servicos["services/ — business rules"]
        SS["SaberService<br/>(saber_service.py)"]
        PERM["permissions · acl · grupos<br/>acesso_galaxia"]
        SYNC["sincronizacao · links<br/>pedidos · propagacao"]
        OUT["copias · restauro · historico<br/>recovery · challenge · jobs"]
    end
    subgraph Pub["publica/ — operations shared by the CLI and the bot"]
        OPS["operacoes.py · processamento.py"]
    end
    subgraph Nucleo["core/ — the trunk"]
        SAB["saber.py — model"]
        REPO["repository.py — SaberRepository"]
        IDX["index.py — GalaxyIndex (SQLite)"]
        REG["registry.py · relation_graph.py"]
    end
    subgraph Infra["infra/ — the roots"]
        STO["storage.py — .bbx"]
        SANK["sankofa.py + sync/ — git-annex"]
        IDT["identity · envelope · jwt"]
        NET["tor_service · farol_lan · ssh_keys"]
    end
    FED["federacao/ — XMPP · ActivityPub"]
    ENR["enriquecimento/ — text · OCR · Whisper"]
    FS[("disk: ~/.baobaxia/galaxias/*<br/>git + git-annex")]

    API --> SS
    CLI --> OPS --> SS
    BOT --> OPS
    TIMERS --> SYNC
    TIMERS --> ENR
    SS --> PERM
    SS --> SYNC
    SS --> OUT
    SS --> REPO
    FED --> SS
    ENR --> REPO
    REPO --> SAB
    REPO --> IDX
    REPO --> STO
    REPO --> SANK
    SYNC --> SANK
    SYNC --> NET
    PERM --> IDT
    STO --> FS
    SANK --> FS

The dependency rule: top to bottom. core/ does not import services/; infra/ knows nothing about FastAPI. When you see a late import (inside a function), it is almost always there to keep this rule without a cycle.

Deploying a mucua

flowchart LR
    pessoa(["browser"]) -->|":80"| NGINX
    subgraph Maquina["the mucua's machine (baobaxia.deb package)"]
        NGINX["nginx<br/>sites-available/baobaxia"]
        FRONT["baobaxia-front<br/>node build/index.js :3000"]
        APIS["baobaxia-api<br/>bbx api :8000"]
        T1["baobaxia-sincronizar.timer<br/>every 30 min"]
        T2["baobaxia-enriquecer.timer<br/>20 min after each round"]
        T3["baobaxia-healthcheck.timer<br/>every 2 min"]
        TOR["own tor<br/>SOCKS 19050 + .onion service"]
        SSHD["sshd<br/>user baobaxia → git-annex-shell"]
        DADOS[("/var/lib/baobaxia/.baobaxia")]
    end
    NGINX -->|"/"| FRONT
    NGINX -->|"/api/"| APIS
    FRONT -->|"SSR: 127.0.0.1:8000"| APIS
    APIS --> DADOS
    T1 --> DADOS
    T2 --> DADOS
    TOR -->|".onion → :8000"| APIS
    outra(["another mucua"]) -->|"git annex sync over SSH"| SSHD
    SSHD --> DADOS
    APIS -.->|"ssh / tor"| outra

Notice: three processes write to the same galaxy (the API, the sync timer, the enrichment timer). That is why services/trava_galaxia.py exists — a flock between processes — and why the index opens SQLite with a timeout.

One request, from click to disk

sequenceDiagram
    autonumber
    actor P as Person
    participant F as Front (SSR)
    participant A as API /api/v3
    participant D as deps.py
    participant S as SaberService
    participant R as SaberRepository
    participant I as GalaxyIndex
    P->>F: GET /en-US/saber/ab12cd34
    F->>A: GET /saberes/ab12… (Bearer from the cookie)
    A->>D: current_mocambola(token)
    D-->>A: {uuid, role, galaxias, mucua, grupos}
    A->>S: get_saber(uuid, actor)
    S->>R: _find_saber → repo.get(uuid) in each galaxy
    R->>I: get(uuid) — finds rel_path
    R-->>S: Saber (read from saber.bbx + presence on disk)
    S->>S: permissions.can(actor, READ, _resource_of(saber))
    S-->>A: to_public_dict(lang) + context
    A-->>F: JSON
    F-->>P: HTML

Practice

  1. Pick a route in api/v3/saberes.py and follow it down to the disk, noting each file and function. Compare with the sequence diagram.
  2. Look in services/ for an import inside a function. Which cycle does it avoid?
  3. In api/v3/app.py, which middlewares exist? What does infra/vigia_api.py watch, and why?

To talk about

  • What are the advantages and the costs of the disk being the truth, rather than a database?
  • If a mucua goes a year without a network, what happens when it comes back?

vedi questo capitolo come sapere