herocode/herocoordinator
13
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

...

Browse Source
This commit is contained in:
despiegk
2025-08-20 09:56:55 +02:00
parent 4174ff1663
commit 2bf66e83bb
8 changed files with 815 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
specs/model/actor.v Normal file
View File

@@ -0,0 +1,18 @@
module model
// a actor is a participant in the new internet, the one who can ask for work
// user can have more than one actor operating for them, an actor always operates in a context which is hosted by the hero of the user
// stored in the context db at actor:<id> (actor is hset)
@[heap]
pub struct Actor {
pub mut:
id u32
pubkey string
address []Address // address (is to reach the actor back), normally mycelium but doesn't have to be
created_at u32 // epoch
updated_at u32 // epoch
}
pub fn (self Actor) redis_key() string {
return 'actor:${self.id}'
}

20
specs/model/context.v Normal file
View File

@@ -0,0 +1,20 @@
module model
// each job is run in a context, this corresponds to a DB in redis and has specific rights to actors
// context is a redis db and also a locaction on a filesystem which can be used for e.g. logs, temporary files, etc.
// actors create contexts for others to work in
// stored in the context db at context:<id> (context is hset)
@[heap]
pub struct Context {
pub mut:
id u32 // corresponds with the redis db (in our ourdb or other redis)
admins []u32 // actors which have admin rights on this context (means can do everything)
readers []u32 // actors which can read the context info
executors []u32 // actors which can execute jobs in this context
created_at u32 // epoch
updated_at u32 // epoch
}
pub fn (self Context) redis_key() string {
return 'context:${self.id}'
}

41
specs/model/flow.v Normal file
View File

@@ -0,0 +1,41 @@
module model
// what get's executed by an actor and needs to be tracked as a whole, can be represented as a DAG graph
// this is the high level representation of a workflow to execute on work, its fully decentralized and distributed
// only the actor who created the flow can modify it and holds it in DB
// stored in the context db at flow:<id> (flow is hset)
@[heap]
pub struct Flow {
pub mut:
id u32 // this job id is given by the actor who called for it
caller_id u32 // is the actor which called for this job
context_id u32 // each job is executed in a context
jobs []u32 // links to all jobs which make up this flow, this can be dynamically modified
env_vars map[string]string // they are copied to every job done
result map[string]string // the result of the flow
created_at u32 // epoch
updated_at u32 // epoch
status FlowStatus
}
pub fn (self Flow) redis_key() string {
return 'flow:${self.id}'
}
// FlowStatus represents the status of a flow
pub enum FlowStatus {
dispatched
started
error
finished
}
// str returns the string representation of FlowStatus
pub fn (self FlowStatus) str() string {
return match self {
.dispatched { 'dispatched' }
.started { 'started' }
.error { 'error' }
.finished { 'finished' }
}
}

68
specs/model/message.v Normal file
View File

@@ -0,0 +1,68 @@
module model
// Messages is what goes over mycelium (which is our messaging system), they can have a job inside
// stored in the context db at msg:<callerid>:<id> (msg is hset)
// there are 2 queues in the context db: queue: msg_out and msg_in these are generic queues which get all messages from mycelium (in) and the ones who need to be sent (out) are in the outqueue
@[heap]
pub struct Message {
pub mut:
id u32 // is unique id for the message, has been given by the caller
caller_id u32 // is the actor whos send this message
context_id u32 // each message is for a specific context
message string
message_type ScriptType
message_format_type MessageFormatType
timeout u32 // in sec, to arrive destination
timeout_ack u32 // in sec, to acknowledge receipt
timeout_result u32 // in sec, to process result and have it back
job []Job
logs []Log // e.g. for streaming logs back to originator
created_at u32 // epoch
updated_at u32 // epoch
status MessageStatus
}
// MessageType represents the type of message
pub enum MessageType {
job
chat
mail
}
// MessageFormatType represents the format of a message
pub enum MessageFormatType {
html
text
md
}
pub fn (self Message) redis_key() string {
return 'message:${self.caller_id}:${self.id}'
}
// queue_suffix returns the queue suffix for the message type
pub fn (mt MessageType) queue_suffix() string {
return match mt {
.job { 'job' }
.chat { 'chat' }
.mail { 'mail' }
}
}
// MessageStatus represents the status of a message
pub enum MessageStatus {
dispatched
acknowledged
error
processed // e.g. can be something which comes back
}
// str returns the string representation of MessageStatus
pub fn (ms MessageStatus) str() string {
return match ms {
.dispatched { 'dispatched' }
.acknowledged { 'acknowledged' }
.error { 'error' }
.processed { 'processed' }
}
}

27
specs/model/runner.v Normal file
View File

@@ -0,0 +1,27 @@
module model
// a runner executes a job, this can be in VM, in a container or just some processes running somewhere
// the messages always come in over a topic
// stored in the context db at runner:<id> (runner is hset)
@[heap]
pub struct Runner {
pub mut:
id u32
pubkey string // from mycelium
address string // mycelium address
topic string // needs to be set by the runner but often runner<runnerid> e.g. runner20
local bool // if local then goes on redis using the id
created_at u32 // epoch
updated_at u32 // epoch
}
pub enum RunnerType {
v
python
osis
rust
}
pub fn (self Runner) redis_key() string {
return 'runner:${self.id}'
}

64
specs/model/runnerjob.v Normal file
View File

@@ -0,0 +1,64 @@
module model
// Job represents a job, a job is only usable in the context of a runner (which is part of a hero)
// stored in the context db at job:<callerid>:<id> (job is hset)
@[heap]
pub struct RunnerJob {
pub mut:
id u32 // this job id is given by the actor who called for it
caller_id u32 // is the actor which called for this job
context_id u32 // each job is executed in a context
script string
script_type ScriptType
timeout u32 // in sec
retries u8
env_vars map[string]string
result map[string]string
prerequisites []string
dependends []u32
created_at u32 // epoch
updated_at u32 // epoch
status JobStatus
}
// ScriptType represents the type of script
pub enum ScriptType {
osis
sal
v
python
}
pub fn (self RunnerJob) redis_key() string {
return 'job:${self.caller_id}:${self.id}'
}
// queue_suffix returns the queue suffix for the script type
pub fn (st ScriptType) queue_suffix() string {
return match st {
.osis { 'osis' }
.sal { 'sal' }
.v { 'v' }
.python { 'python' }
}
}
// JobStatus represents the status of a job
pub enum JobStatus {
dispatched
waiting_for_prerequisites
started
error
finished
}
// str returns the string representation of JobStatus
pub fn (js JobStatus) str() string {
return match js {
.dispatched { 'dispatched' }
.waiting_for_prerequisites { 'waiting_for_prerequisites' }
.started { 'started' }
.error { 'error' }
.finished { 'finished' }
}
}

314
specs/models.md Normal file
View File

@@ -0,0 +1,314 @@
# Models Specification
*Freeflow Universe – mycojobs*
This document gathers **all data‑models** that exist in the `lib/mycojobs/model/` package, together with a concise purpose description, field semantics, Redis storage layout and the role each model plays in the overall *decentralised workflow* architecture.
## Table of Contents
1. [Actor](#actor)
2. [Context](#context)
3. [Flow](#flow)
4. [Message](#message)
5. [Runner](#runner)
6. [RunnerJob](#runnerjob)
7. [Enums & Shared Types](#enums-shared-types)
8. [Key‑generation helpers](#key-generation-helpers)
---
## <a name="actor"></a>1️⃣ `Actor` – Identity & entry‑point
| Field | Type | Description |
|------|------|-------------|
| `id` | `u32` | Sequential identifier **unique per tenant**. Used as part of the Redis key `actor:<id>`. |
| `pubkey` | `string` | Public key (Mycelium‑compatible) that authenticates the actor when it sends/receives messages. |
| `address` | `[]Address` | One or more reachable addresses (normally Mycelium topics) that other participants can use to contact the actor. |
| `created_at` | `u32` | Unix‑epoch time when the record was created. |
| `updated_at` | `u32` | Unix‑epoch time of the last mutation. |
### Purpose
* An **Actor** is the *human‑or‑service* that **requests work**, receives results and can be an administrator of a **Context**.
* It is the *security principal* – every operation in a context is authorised against the actor’s ID and its public key signature.
### Redis representation
| Key | Example | Storage type | Fields |
|-----|---------|--------------|--------|
| `actor:${id}` | `actor:12` | **hash** (`HSET`) | `id`, `pubkey`, `address` (list), `created_at`, `updated_at` |
---
## <a name="context"></a>2️⃣ `Context` – Tenant & permission container
| Field | Type | Description |
|------|------|-------------|
| `id` | `u32` | Identifier that also selects the underlying **Redis DB** for this tenant. |
| `admins` | `[]u32` | Actor IDs that have **full control** (create/delete any object, manage permissions). |
| `readers` | `[]u32` | Actor IDs that may **read** any object in the context but cannot modify. |
| `executors` | `[]u32` | Actor IDs allowed to **run** `RunnerJob`s and update their status. |
| `created_at` | `u32` | Unix‑epoch of creation. |
| `updated_at` | `u32` | Unix‑epoch of last modification. |
### Purpose
* A **Context** isolates a *tenant* – each tenant gets its own Redis database and a dedicated filesystem area (for logs, temporary files, …).
* It stores **permission lists** that the system consults before any operation (e.g., creating a `Flow`, enqueuing a `RunnerJob`).
### Redis representation
| Key | Example | Storage type | Fields |
|-----|---------|--------------|--------|
| `context:${id}` | `context:7` | **hash** | `id`, `admins`, `readers`, `executors`, `created_at`, `updated_at` |
---
## <a name="flow"></a>3️⃣ `Flow` – High‑level workflow (DAG)
| Field | Type | Description |
|------|------|-------------|
| `id` | `u32` | Flow identifier – *unique inside the creator’s actor space*. |
| `caller_id` | `u32` | Actor that **created** the flow (owner). |
| `context_id` | `u32` | Context in which the flow lives. |
| `jobs` | `[]u32` | List of **RunnerJob** IDs that belong to this flow (the DAG edges are stored in each job’s `dependends`). |
| `env_vars` | `map[string]string` | Global environment variables injected into **every** job of the flow. |
| `result` | `map[string]string` | Aggregated output produced by the flow (filled by the orchestrator when the flow finishes). |
| `created_at` | `u32` | Creation timestamp. |
| `updated_at` | `u32` | Last update timestamp. |
| `status` | `FlowStatus` | Current lifecycle stage (`dispatched`, `started`, `error`, `finished`). |
### Purpose
* A **Flow** is the *public‑facing* representation of a **workflow**.
* It groups many `RunnerJob`s, supplies common env‑vars, tracks overall status and collects the final result.
* Only the *creator* (the `caller_id`) may mutate the flow definition.
### Redis representation
| Key | Example | Storage type | Fields |
|-----|---------|--------------|--------|
| `flow:${id}` | `flow:33` | **hash** | `id`, `caller_id`, `context_id`, `jobs`, `env_vars`, `result`, `created_at`, `updated_at`, `status` |
### `FlowStatus` enum
| Value | Meaning |
|-------|---------|
| `dispatched` | Flow has been stored but not yet started. |
| `started` | At least one job is running. |
| `error` | One or more jobs failed; flow aborted. |
| `finished` | All jobs succeeded, `result` is final. |
---
## <a name="message"></a>4️⃣ `Message` – Transport unit (Mycelium)
| Field | Type | Description |
|------|------|-------------|
| `id` |u32 `_type` | `ScriptType` | *Kind* of the message – currently re‑used for job payloads (`osis`, `sal`, `v`, `python`). |
| `message_format_type` | `MessageFormatType` | Formatting of `message` (`html`, `text`, `md`). |
| `timeout` | `u32` | Seconds before the message is considered *lost* if not delivered. |
| `timeout_ack` | `u32` | Seconds allowed for the receiver to acknowledge. |
| `timeout_result` | `u32` | Seconds allowed for the receiver to send back a result. |
| `job` | `[]Job` | Embedded **RunnerJob** objects (normally a single job). |
| `logs` | `[]Log` | Optional streaming logs attached to the message. |
| `created_at` | `u32` | Timestamp of creation. |
| `updated_at` | `u32` | Timestamp of latest update. |
| `status` | `MessageStatus` | Current lifecycle (`dispatched`, `acknowledged`, `error`, `processed`). |
### Purpose
* `Message` is the **payload carrier** that travels over **Mycelium** (the pub/sub system).
* It can be a **job request**, a **chat line**, an **email**, or any generic data that needs to be routed between actors, runners, or services.
* Every message is persisted as a Redis hash; the system also maintains two *generic* queues:
* `msg_out` – outbound messages waiting to be handed to Mycelium.
* `msg_in` – inbound messages that have already arrived and are awaiting local processing.
### Redis representation
| Key | Example | Storage type | Fields |
|-----|---------|--------------|--------|
| `message:${caller_id}:${id}` | `message:12:101` | **hash** | All fields above (`id`, `caller_id`, `context_id`, …, `status`). |
### `MessageType` enum (legacy – not used in current code but documented)
| Value | Meaning |
|-------|---------|
| `job` | Payload carries a `RunnerJob`. |
| `chat` | Human‑to‑human communication. |
| `mail` | Email‑like message. |
### `MessageFormatType` enum
| Value | Meaning |
|-------|---------|
| `html` | HTML formatted body. |
| `text` | Plain‑text. |
| `md` | Markdown. |
### `MessageStatus` enum
| Value | Meaning |
|-------|---------|
| `dispatched` | Stored, not yet processed. |
| `acknowledged` | Receiver has confirmed receipt. |
| `error` | Delivery or processing failed. |
|` | Message handled (e.g., job result returned). |
---
## <a name="runner"></a>5️⃣ `Runner` – Worker that executes jobs
| Field | Type | Description |
|------|------|-------------|
| `id` | `u32` | Unique runner identifier. |
| `pubkey` | `string` | Public key of the runner (used by Mycelium for auth). |
| `address` | `string` | Mycelium address (e.g., `mycelium://…`). |
| `topic` | `string` | Pub/Sub topic the runner subscribes to; defaults to `runner${id}`. |
| `local` | `bool` | If `true`, the runner also consumes jobs directly from **Redis queues** (e.g., `queue:v`). |
| `created_at` | `u32` | Creation timestamp. |
| `updated_at` | `u32` | Last modification timestamp. |
### Purpose
* A **Runner** is the *execution engine* – it could be a VM, a container, or a process that knows how to run a specific script type (`v`, `python`, `osis`, `rust`).
* It **subscribes** to a Mycelium topic to receive job‑related messages, and, when `local==true`, it also **polls** a Redis list named after the script‑type (`queue:<suffix>`).
### Redis representation
| Key | Example | Storage type |
|-----|---------|--------------|
| `runner:${id}` | `runner:20` | **hash** *(all fields above)* |
### `RunnerType` enum
| Value | Intended runtime |
|-------|------------------|
| `v` | V language VM |
| `python` | CPython / PyPy |
| `osis` | OSIS‑specific runtime |
| `rust` | Native Rust binary |
---
## <a name="runnerjob"></a>6️⃣ `RunnerJob` – Executable unit
| Field | Type | Description |
|------|------|-------------|
| `id` | `u32` | Job identifier **provided by the caller**. |
| `caller_id` | `u32` | Actor that created the job. |
| `context_id` | `u32` | Context in which the job will run. |
| `script` | `string` | Source code / command to be executed. |
| `script_type` | `ScriptType` | Language or runtime of the script (`osis`, `sal`, `v`, `python`). |
| `timeout` | `u32` | Maximum execution time (seconds). |
| `retries` | `u8` | Number of automatic retries on failure. |
| `env_vars` | `map[string]string` | Job‑specific environment variables (merged with `Flow.env_vars`). |
| `result` | `map[string]string` | Key‑value map that the job writes back upon completion. |
| `prerequisites` | `[]string` | Human‑readable IDs of **external** prerequisites (e.g., files, other services). |
| `dependends` | `[]u32` | IDs of **other RunnerJob** objects that must finish before this job can start. |
| `created_at` | `u32` | Creation timestamp. |
| `updated_at` | `u32` | Last update timestamp. |
| `status` | `JobStatus` | Lifecycle status (`dispatched`, `waiting_for_prerequisites`, `started`, `error`, `finished`). |
### Purpose
* A **RunnerJob** is the *atomic piece of work* that a `Runner` executes.
* It lives inside a **Context**, is queued according to its `script_type`, and moves through a well‑defined **state machine**.
* The `dependends` field enables the *DAG* behaviour that the `Flow` model represents at a higher level.
### Redis representation
| Key | Example | Storage type |
|-----|---------|--------------|
| `job:${caller_id}:${id}` | `job:12:2001` | **hash** *(all fields above)* |
### `ScriptType` enum
| Value | Runtime |
|-------|---------|
| `osis` | OSIS interpreter |
| `sal` | SAL DSL (custom) |
| `v` | V language |
| `python`| CPython / PyPy |
*The enum provides a **`queue_suffix()`** helper that maps a script type to the name of the Redis list used for local job dispatch (`queue:python`, `queue:v`, …).*
### `JobStatus` enum
| Value | Meaning |
|-------|---------|
| `dispatched` | Stored, waiting to be examined for prerequisites. |
| `waiting_for_prerequisites` | Has `dependends` that are not yet finished. |
| `started` | Currently executing on a runner. |
| `error` | Execution failed (or exceeded retries). |
| `finished` | Successfully completed, `result` populated. |
---
## <a name="enums-shared-types"></a>7️⃣ Other Enums & Shared Types
| Enum | Location | Values | Note |
|------|----------|--------|------|
| `MessageType` | `message.v` | `job`, `chat`, `mail` | Determines how a `Message` is interpreted. |
| `MessageFormatType` | `message.v` | `html`, `text`, `md` | UI‑layer rendering hint. |
| `MessageStatus` | `message.v` | `dispatched`, `acknowledged`, `error`, `processed` | Life‑cycle of a `Message`. |
| `FlowStatus` | `flow.v` | `dispatched`, `started`, `error`, `finished` | High‑level flow progress. |
| `RunnerType` | `runner.v` | `v`, `python`, `osis`, `rust` | Not currently stored; used by the orchestration layer to pick a runner implementation. |
| `ScriptType` | `runnerjob.v` | `osis`, `sal`, `v`, `python` | Determines queue suffix & runtime. |
| `JobStatus` | `runnerjob.v` | `dispatched`, `waiting_for_prerequisites`, `started`, `error`, `finished` | Per‑job state machine. |
---
## <a name="key-generation-helpers"></a>8️⃣ Key‑generation helpers (methods)
| Model | Method | Returns | Example |
|-------|--------|---------|---------|
| `Actor` | `redis_key()` | `"actor:${self.id}"` | `actor:12` |
| `Context` | `redis_key()` | `"context:${self.id}"` | `context:7` |
| `Flow` | `redis_key()` | `"flow:${self.id}"` | `flow:33` |
| `Message` | `redis_key()` | `"message:${self.caller_id}:${self.id}"` | `message:12:101` |
| `Runner` | `redis_key()` | `"runner:${self.id}"` | `runner:20` |
| `RunnerJob` | `redis_key()` | `"job:${self.caller_id}:${self.id}"` | `job:12:2001` |
| `MessageType` | `queue_suffix()` | `"job"` / `"chat"` / `"mail"` | `MessageType.job.queue_suffix() → "job"` |
| `ScriptType` | `queue_suffix()` | `"osis"` / `"sal"` / `"v"` / `"python"` | `ScriptType.python.queue_suffix() → "python"` |
These helpers guarantee **canonical key naming** throughout the code base and simplify Redis interactions.
---
## 📌 Summary Diagram (quick reference)
```mermaid
%%{init: {"theme":"dark"}}%%
graph TD
%% Actors and what they can create
A[Actor] -->|creates| Ctx[Context]
A -->|creates| Fl[Flow]
A -->|creates| Msg[Message]
A -->|creates| Rnr[Runner]
A -->|creates| Job[RunnerJob]
%% All objects live inside one Redis DB that belongs to a Context
subgraph "Redis DB (per Context)"
Ctx
A
Fl
Msg
Rnr
Job
end
%% Messaging queues (global, outside the Context DB)
Msg -->|pushes key onto| OutQ[msg_out]
OutQ -->|transport via Mycelium| InQ[msg_in]
InQ -->|pulled by| Rnr
%% Local runner queues (only when runner.local == true)
Rnr -->|BRPOP from| QueueV["queue:v"]
Rnr -->|BRPOP from| QueuePy["queue:python"]
Rnr -->|BRPOP from| QueueOSIS["queue:osis"]
```
## context based
* Inside a Context, an **Actor** can create a **Flow** that references many **RunnerJob** IDs (the DAG).
* To *initiate* execution, the Actor packages a **RunnerJob** (or a full Flow) inside a **Message**, pushes it onto `msg_out`, and the system routes it via **Mycelium** to the target Context.
* The remote **Runner** receives the Message, materialises the **RunnerJob**, queues it on a script‑type list, executes it, writes back `result` and status, and optionally sends a *result Message* back to the originator.
All state is persisted as **Redis hashes**, guaranteeing durability and enabling *idempotent* retries. The uniform naming conventions (`actor:<id>`, `job:<caller_id>:<id>`, …) make it trivial to locate any object given its identifiers.

263
specs/specs.md Normal file
View File

@@ -0,0 +1,263 @@
## Objects Used
| Component | What it **stores** | Where it lives (Redis key) | Main responsibilities |
|------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-------------------------------------------|-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| **Actor** | Public key, reachable addresses, timestamps | `actor:<id>` (hash) | An identity that can request work, receive results and act as an administrator of a *Context*. |
| **Context**| Permission lists (`admins`, `readers`, `executors`), timestamps | `context:<id>` (hash) | An isolated “tenant” – a separate Redis DB and filesystem area. All objects (flows, messages, jobs, runners) belonging to a given workflow are stored under this context. The permission lists control who may read, execute or administer the context. |
| **Flow** | DAG of job IDs, env‑vars, result map, status, timestamps | `flow:<id>` (hash) | A high‑level workflow created by a single **Actor**. It groups many **RunnerJob** objects, records their execution order, supplies common environment variables and aggregates the final result. |
| **Message**| Payload, type (`job\|chat\|mail`), format (`html\|text\|md`), time‑outs, embedded **Job** objects, log stream, status, timestamps | `message:<caller_id>:<id>` (hash) | The transport unit that travels over **Mycelium** (the pub/sub/message bus). A message can contain a **RunnerJob** (or a list of jobs) and is queued in two generic Redis lists: `msg_out` (to be sent) and `msg_in` (already received). |
| **Runner** | Public key, Mycelium address, topic name, type (`v\|python\|osis\|rust`), local flag, timestamps | `runner:<id>` (hash) | The *worker* that actually executes **RunnerJob** scripts. It subscribes to a Mycelium topic (normally `runner<id>`). If `local == true` the runner also consumes jobs directly from a Redis queue that is named after the script‑type suffix (`v`, `python`, …). |
| **RunnerJob**| Script source, type (`osis\|sal\|v\|python`), env‑vars, prerequisites, dependencies, status, timestamps, result map | `job:<caller_id>:<id>` (hash) | A single executable unit. It lives inside a **Context**, belongs to a **Runner**, and is queued according to its `script_type` (e.g. `queue:python`). Its status moves through the lifecycle `dispatched → waiting_for_prerequisites → started → finished|error`. |
> **Key idea:** All objects are persisted as *hashes* in a **Redis** database that is dedicated to a *Context*. The system is completely **decentralised** – each actor owns its own context and can spin up as many runners as needed. Communication between actors, runners and the rest of the system happens over **Mycelium**, a message‑bus that uses Redis lists as queues.
---
## Interaction diagram (who talks to who)
### Sequence diagram – “Submit a flow and run it”
```mermaid
%%{init: {"theme":"dark"}}%%
sequenceDiagram
participant A as Actor
participant L as Local‑Context (Redis)
participant M as Mycelium (msg_out / msg_in)
participant R as Remote‑Context (Redis)
participant W as Runner (worker)
%% 1. Actor creates everything locally
A->>L: create Flow + RunnerJob (J)
A->>L: LPUSH msg_out Message{type=job, payload=J, target=Remote}
%% 2. Mycelium transports the message
M->>R: LPUSH msg_in (Message key)
%% 3. Remote context materialises the job
R->>R: HSET Message hash
R->>R: HSET RunnerJob (J') // copy of payload
R->>R: LPUSH queue:v (job key)
%% 4. Runner consumes and executes
W->>R: BRPOP queue:v (job key)
W->>R: HSET job status = started
W->>W: execute script
W->>R: HSET job result + status = finished
%% 5. Result is sent back
W->>M: LPUSH msg_out Message{type=result, payload=result, target=Local}
M->>L: LPUSH msg_in (result Message key)
%% 6. Actor receives the result
A->>L: RPOP msg_in → read result
```
### 2.2 Component diagram – “Static view of objects & links”
```mermaid
%%{init: {"theme":"dark"}}%%
graph LR
subgraph Redis["Redis (per Context)"]
A[Actor] -->|stores| Ctx[Context]
Ctx -->|stores| Fl[Flow]
Ctx -->|stores| Msg[Message]
Ctx -->|stores| Rnr[Runner]
Ctx -->|stores| Job[RunnerJob]
end
subgraph Mycelium["Mycelium (Pub/Sub)"]
MsgOut["queue:msg_out"] -->|outgoing| Mcel[Mycelium Bus]
Mcel -->|incoming| MsgIn["queue:msg_in"]
RnrTopic["topic:runnerX"] -->|subscribed by| Rnr
queueV["queue:v"] -->|local jobs| Rnr
queuePython["queue:python"] -->|local jobs| Rnr
end
A -->|creates / reads| Fl
A -->|creates / reads| Msg
A -->|creates / reads| Rnr
A -->|creates / reads| Job
Fl -->|references| Job
Msg -->|may embed| Job
Rnr -->|executes| Job
Job -->|updates| Fl
Msg -->|carries result back to| A
```
### 2.3 Flow‑status life‑cycle (state diagram)
```mermaid
%%{init: {"theme":"dark"}}%%
stateDiagram-v2
[*] --> dispatched
dispatched --> waiting_for_prerequisites : has prereqs
waiting_for_prerequisites --> started : prereqs met
dispatched --> started : no prereqs
started --> finished : success
started --> error : failure
waiting_for_prerequisites --> error : timeout / impossible
error --> [*]
finished --> [*]
```
---
## 3️⃣ Redis objects – concrete key & data layout
All objects are stored as **hashes** (`HSET`). Below is a concise catalog that can be copied into a design doc.
| Key pattern | Example | Fields (type) | Comments |
|-------------|---------|---------------|----------|
| `actor:${id}` | `actor:12` | `id` u32, `pubkey` str, `address` list\<Address\>, `created_at` u32, `updated_at` u32 | One hash per actor. |
| `context:${id}` | `context:7` | `id` u32, `admins` list\<u32\>, `readers` list\<u32\>, `executors` list\<u32\>, `created_at` u32, `updated_at` u32 | Holds permission lists for a tenant. |
| `flow:${id}` | `flow:33` | `id` u32, `caller_id` u32, `context_id` u32, `jobs` list\<u32\>, `env_vars` map\<str,str\>, `result` map\<str,str\>, `created_at` u32, `updated_at` u32, `status` str (`dispatched|started|error|finished`) |
| `message:${caller_id}:${id}` | `message:12:101` | `id` u32, `caller_id` u32, `context_id` u32, `message` str, `message_type` str (`job|chat|mail`), `message_format_type` str (`html|text|md`), `timeout` u32, `timeout_ack` u32, `timeout_result` u32, `job` list\<RunnerJob\> (serialized), `logs` list\<Log\>, `created_at` u32, `updated_at` u32, `status` str (`dispatched|acknowledged|error|processed`) |
| `runner:${id}` | `runner:20` | `id` u32, `pubkey` str, `address` str, `topic` str, `local` bool, `created_at` u32, `updated_at` u32 |
| `job:${caller_id}:${id}` | `job:12:2001` | `id` u32, `caller_id` u32, `context_id` u32, `script` str, `script_type` str (`osis|sal|v|python`), `timeout` u32, `retries` u8, `env_vars` map\<str,str\>, `result` map\<str,str\>, `prerequisites` list\<str\>, `dependends` list\<u32\>, `created_at` u32, `updated_at` u32, `status` str (`dispatched|waiting_for_prerequisites|started|error|finished`) |
#### Queue objects (lists)
| Queue name | Purpose |
|------------|---------|
| `msg_out` | **Outbound** generic queue – every `Message` that an actor wants to send is pushed here. |
| `msg_in` | **Inbound** generic queue – every message received from Mycelium is placed here for the local consumer to process. |
| `queue:${suffix}` (e.g. `queue:v`, `queue:python`) | Local job queues used by a **Runner** when `local == true`. The suffix comes from `ScriptType.queue_suffix()`. |
---
## 4️⃣ System specification (as a concise “specs” section)
### 4.1 Naming conventions
* All Redis **hashes** are prefixed with the object name (`actor:`, `context:`, …).
* All **queues** are simple Redis lists (`LPUSH` / `RPOP`).
* **Message** keys embed both the *caller* and a locally unique *message id* – this guarantees global uniqueness across contexts.
### 4.2 Permissions & security
* Only IDs present in `Context.admins` may **create** or **delete** any object inside that context.
* `Context.readers` can **GET** any hash but not modify it.
* `Context.executors` are allowed to **update** `RunnerJob.status`, `result` and to **pop** from local job queues.
* Every `Actor` must present a `pubkey` that can be verified by the receiving side (Mycelium uses asymmetric crypto).
### 4.3 Message flow (publish / consume)
Below is a **re‑written “Message flow (publish / consume)”** that reflects the real runtime components:
* **Supervisor daemon** – runs on the node that owns the **Flow** (the *actor’s* side).
It is the only process that ever **RPOP**s from the global `msg_out` queue, adds the proper routing information and hands the message to **Mycelium**.
* **Mycelium** – the pure pub/sub/message‑bus. It never touches Redis directly; it only receives a *payload key* from the coordinator and delivers that key to the remote tenant’s `msg_in` list.
* **Remote‑side runner / service** – consumes from its own `msg_in`, materialises the job and executes it.
The table now uses the exact component names and adds a short note about the permission check that the coordinator performs before it releases a message.
| # | Action (what the system does) | Component that performs it | Redis interaction (exact commands) |
|---|-------------------------------|----------------------------|------------------------------------|
| **1️⃣ Publish** | Actor creates a `Message` hash and **LPUSH**es its key onto the *outbound* queue. | **Actor** (client code) | `HSET message:12:101 …` <br/> `LPUSH msg_out message:12:101` |
| **2️⃣ Coordinate & route** | The **Supervisor daemon** (running at source) **RPOP**s the key, checks the actor’s permissions, adds the *target‑context* and *topic* fields, then forwards the key to Mycelium. | **Supervisor daemon** (per‑actor) | `RPOP msg_out` → (in‑process) → `LPUSH msg_out_coordinator <key>` (internal buffer) |
| **3️⃣ Transport** | Mycelium receives the key, looks at `Message.message_type` (or the explicit `topic`) and pushes the key onto the *inbound* queue of the **remote** tenant. | **Mycelium bus** (network layer) | `LPUSH msg_in:<remote‑ctx> <key>` |
| **4️⃣ Consume** | The **Remote side** (runner or service) **RPOP**s from its `msg_in`, loads the full hash, verifies the actor’s signature and decides what to do based on `message_type`. | **Remote consumer** (runner / service | `RPOP msg_in:<remote‑ctx>` → `HGETALL message:<key>` |
| **5️⃣ Job materialisation** | If `message_type == "job"` the consumer creates a **RunnerJob** entry inside the **remote** context, adds the job **key** to the proper *script‑type* queue (`queue:v`, `queue:python`, …). | **Remote consumer** | `HSET job:<caller_id>:<job_id> …` <br/> `LPUSH queue:<script_type> job:<caller_id>:<job_id>` |
| **6️⃣ Runner execution loop** | A **Runner** attached to that remote context **BRPOP**s from its script‑type queue, sets `status = started`, runs the script, writes `result` and final `status`. | **Runner** | `BRPOP queue:<script_type>` → `HSET job:<…> status started` → … → `HSET job:<…> result … status finished` |
| **7️⃣ Result notification** | The runner builds a new `Message` (type `chat`, `result`, …) and pushes it onto **msg_out** again. The **Supervisor daemon** on the *originating* side will later pick it up and route it back to the original actor. | **Runner** → **Supervisor (remote side)** → **Mycelium** → **Supervisor (origin side)** → **Actor** | `HSET message:<res_key> …` <br/> `LPUSH msg_out message:<res_key>` (steps 2‑3 repeat in reverse direction) |
---
## Tiny end‑to‑end sequence (still simple enough to render)
```mermaid
%%{init: {"theme":"dark"}}%%
sequenceDiagram
participant A as Actor
participant L as Local‑Redis (Flow ctx)
participant C as Supervisor daemon (local)
participant M as Mycelium bus
participant R as Remote‑Redis (target ctx)
participant W as Runner (remote)
%% 1️⃣ publish
A->>L: HSET message:12:101 …
A->>L: LPUSH msg_out message:12:101
%% 2️⃣ coordinate
C->>L: RPOP msg_out
C->>C: check permissions / add routing info
C->>M: push key to Mycelium (msg_out_coordinator)
%% 3️⃣ transport
M->>R: LPUSH msg_in message:12:101
%% 4️⃣ consume
R->>W: RPOP msg_in
R->>R: HGETALL message:12:101
R->>R: verify signature
alt message_type == job
R->>R: HSET job:12:2001 …
R->>R: LPUSH queue:v job:12:2001
end
%% 5️⃣ runner loop
W->>R: BRPOP queue:v (job:12:2001)
W->>R: HSET job:12:2001 status started
W->>W: execute script
W->>R: HSET job:12:2001 result … status finished
%% 6️⃣ result back
W->>R: HSET message:12:900 result …
W->>R: LPUSH msg_out message:12:900
C->>M: (coordinator on remote side) routes back
M->>L: LPUSH msg_in message:12:900
A->>L: RPOP msg_in → read result
```
## 5️⃣ What the **system** is trying to achieve
| Goal | How it is realized |
|------|--------------------|
| **Decentralised execution** | Every *actor* owns a **Context**; any number of **Runners** can be attached to that context, possibly on different machines, and they all talk over the same Mycelium/Redis backend. |
| **Fine‑grained permissions** | `Context.admins/readers/executors` enforce who can create, view or run jobs. |
| **Loose coupling via messages** | All actions (job submission, result propagation, chat, mail …) use the generic `Message` object; the same transport pipeline handles all of them. |
| **Workflow orchestration** | The **Flow** object models a DAG of jobs, tracks collective status and aggregates results, without needing a central scheduler. |
| **Pluggable runtimes** | `ScriptType` and `RunnerType` let a runner choose the proper execution environment (V, Python, OSIS, Rust, …) – adding a new language only means adding a new `ScriptType` and a corresponding worker. |
| **Observability** | `Log` arrays attached to a `Message` and the timestamps on every hash give a complete audit trail. |
| **Resilience** | Jobs are idempotent hash entries; queues are persisted in Redis, and status changes are atomic (`HSET`). Retries and time‑outs guarantee eventual consistency. |
---
## 6️⃣ Diagram summary (quick visual cheat‑sheet)
```mermaid
%%{init: {"theme":"dark"}}%%
graph TD
A[Actor] -->|creates| Ctx[Context]
A -->|creates| Flow
A -->|creates| Msg
A -->|creates| Rnr[Runner]
A -->|creates| Job[RunnerJob]
subgraph Redis["Redis (per Context)"]
Ctx --> A
Ctx --> Flow
Ctx --> Msg
Ctx --> Rnr
Ctx --> Job
end
Msg -->|push to| OutQ[msg_out]
OutQ --> Myc[Mycelium Bus]
Myc -->|deliver| InQ[msg_in]
InQ --> Rnr
Rnr -->|pop from| Qv["queue:v"]
Rnr -->|pop from| Qpy["queue:python"]
Rnr -->|updates| Job
Job -->|updates| Flow
Flow -->|result Message| Msg
```