kobako 0.27.0-aarch64-linux → 0.29.0-aarch64-linux
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- checksums.yaml +4 -4
- data/.release-please-manifest.json +1 -1
- data/CHANGELOG.md +241 -0
- data/README.md +240 -147
- data/ROADMAP.md +28 -22
- data/SECURITY.md +2 -2
- data/crates/kobako-runtime/CHANGELOG.md +117 -0
- data/crates/kobako-runtime/README.md +16 -10
- data/crates/kobako-transport/CHANGELOG.md +45 -0
- data/crates/kobako-transport/README.md +21 -12
- data/crates/kobako-wasmtime/CHANGELOG.md +163 -0
- data/crates/kobako-wasmtime/README.md +6 -4
- data/data/kobako.wasm +0 -0
- data/lib/kobako/3.3/kobako.so +0 -0
- data/lib/kobako/3.4/kobako.so +0 -0
- data/lib/kobako/4.0/kobako.so +0 -0
- data/lib/kobako/capture.rb +1 -4
- data/lib/kobako/catalog/extensions.rb +21 -59
- data/lib/kobako/catalog/flags.rb +29 -0
- data/lib/kobako/catalog/grants.rb +30 -0
- data/lib/kobako/catalog/handles.rb +7 -54
- data/lib/kobako/catalog/mount.rb +27 -0
- data/lib/kobako/catalog/services.rb +17 -60
- data/lib/kobako/catalog/snippets.rb +41 -69
- data/lib/kobako/catalog.rb +13 -10
- data/lib/kobako/codec/decoder.rb +10 -28
- data/lib/kobako/codec/encoder.rb +3 -22
- data/lib/kobako/codec/error.rb +6 -27
- data/lib/kobako/codec/ext_types.rb +72 -94
- data/lib/kobako/codec/handle_walk.rb +65 -130
- data/lib/kobako/codec/nesting.rb +3 -8
- data/lib/kobako/codec/state.rb +21 -11
- data/lib/kobako/codec/utils.rb +6 -34
- data/lib/kobako/codec/value_budget.rb +86 -0
- data/lib/kobako/codec.rb +10 -22
- data/lib/kobako/constant_name.rb +8 -0
- data/lib/kobako/context.rb +83 -117
- data/lib/kobako/errors.rb +57 -132
- data/lib/kobako/execution.rb +16 -21
- data/lib/kobako/extension.rb +23 -40
- data/lib/kobako/filesystem.rb +38 -0
- data/lib/kobako/handle.rb +7 -12
- data/lib/kobako/outcome.rb +51 -80
- data/lib/kobako/payload/arguments.rb +1 -16
- data/lib/kobako/payload.rb +4 -12
- data/lib/kobako/pool.rb +25 -55
- data/lib/kobako/runtime.rb +7 -21
- data/lib/kobako/sandbox.rb +128 -137
- data/lib/kobako/sandbox_options.rb +71 -90
- data/lib/kobako/snippet/bytecode.rb +11 -0
- data/lib/kobako/snippet/source.rb +1 -16
- data/lib/kobako/snippet.rb +2 -10
- data/lib/kobako/transport/block_failure.rb +35 -0
- data/lib/kobako/transport/call.rb +1 -14
- data/lib/kobako/transport/dispatcher.rb +127 -206
- data/lib/kobako/transport/error.rb +6 -20
- data/lib/kobako/transport/exposure.rb +10 -17
- data/lib/kobako/transport/reflection.rb +108 -63
- data/lib/kobako/transport/run.rb +17 -71
- data/lib/kobako/transport/yielder.rb +20 -94
- data/lib/kobako/transport.rb +4 -15
- data/lib/kobako/unresolved.rb +4 -11
- data/lib/kobako/usage.rb +2 -6
- data/lib/kobako/version.rb +1 -1
- data/lib/kobako.rb +1 -0
- data/sig/kobako/catalog/extensions.rbs +2 -2
- data/sig/kobako/catalog/flags.rbs +13 -0
- data/sig/kobako/catalog/grants.rbs +13 -0
- data/sig/kobako/catalog/mount.rbs +28 -0
- data/sig/kobako/catalog/services.rbs +3 -5
- data/sig/kobako/catalog/snippets.rbs +5 -5
- data/sig/kobako/catalog.rbs +1 -0
- data/sig/kobako/codec/ext_types.rbs +10 -8
- data/sig/kobako/codec/handle_walk.rbs +9 -7
- data/sig/kobako/codec/state.rbs +7 -1
- data/sig/kobako/codec/utils.rbs +2 -2
- data/sig/kobako/codec/value_budget.rbs +23 -0
- data/sig/kobako/codec.rbs +4 -2
- data/sig/kobako/constant_name.rbs +3 -0
- data/sig/kobako/context.rbs +15 -8
- data/sig/kobako/errors.rbs +10 -4
- data/sig/kobako/filesystem.rbs +88 -0
- data/sig/kobako/guest_yielder.rbs +9 -0
- data/sig/kobako/outcome.rbs +6 -8
- data/sig/kobako/pool.rbs +1 -1
- data/sig/kobako/runtime.rbs +21 -45
- data/sig/kobako/sandbox.rbs +11 -1
- data/sig/kobako/sandbox_options.rbs +25 -7
- data/sig/kobako/snapshot.rbs +20 -0
- data/sig/kobako/snippet/bytecode.rbs +12 -0
- data/sig/kobako/transport/block_failure.rbs +13 -0
- data/sig/kobako/transport/dispatcher.rbs +15 -19
- data/sig/kobako/transport/exposure.rbs +1 -1
- data/sig/kobako/transport/reflection.rbs +19 -4
- data/sig/kobako/transport/run.rbs +1 -3
- data/sig/kobako/transport/yielder.rbs +5 -9
- data/sig/kobako/transport.rbs +9 -8
- metadata +30 -9
- data/lib/kobako/snippet/binary.rb +0 -26
- data/rust-toolchain.toml +0 -15
- data/sig/kobako/snippet/binary.rbs +0 -12
data/README.md
CHANGED
|
@@ -2,9 +2,9 @@
|
|
|
2
2
|
|
|
3
3
|
[](https://deepwiki.com/elct9620/kobako)
|
|
4
4
|
|
|
5
|
-
Kobako is a Ruby gem that embeds a Wasm-isolated mruby interpreter inside your application
|
|
5
|
+
Kobako is a Ruby gem that embeds a Wasm-isolated mruby interpreter inside your application. It runs untrusted Ruby scripts in-process — LLM-generated code, user formulas, student submissions, third-party plugins. Those scripts get no access to host memory, files, network, or credentials. Its home is the Ruby ecosystem; a Rust SDK offers the same sandbox to hosts written in other languages.
|
|
6
6
|
|
|
7
|
-
The host (`wasmtime`) runs a precompiled `kobako.wasm` guest containing mruby and a Transport proxy.
|
|
7
|
+
The host (`wasmtime`) runs a precompiled `kobako.wasm` guest containing mruby and a Transport proxy. A guest script reaches the outside world only through Host App-declared **Services**, the named host objects you explicitly inject into the sandbox. The guest sees each one as a proxy that forwards calls back to the host over the Transport wire.
|
|
8
8
|
|
|
9
9
|
```
|
|
10
10
|
Host process Wasm guest
|
|
@@ -35,6 +35,8 @@ The precompiled `kobako.wasm` Guest Binary ships inside the gem, so end users do
|
|
|
35
35
|
|
|
36
36
|
## Installation
|
|
37
37
|
|
|
38
|
+
Install the gem; it ships its Guest Binary prebuilt.
|
|
39
|
+
|
|
38
40
|
```bash
|
|
39
41
|
bundle add kobako
|
|
40
42
|
# or
|
|
@@ -43,7 +45,7 @@ gem install kobako
|
|
|
43
45
|
|
|
44
46
|
## Quick Start
|
|
45
47
|
|
|
46
|
-
The same untrusted script runs identically from either host frontend. It executes inside the Wasm guest and
|
|
48
|
+
The same untrusted script runs identically from either host frontend. It executes inside the Wasm guest and reaches no file you have not mounted for it, no socket, and none of your environment.
|
|
47
49
|
|
|
48
50
|
```ruby
|
|
49
51
|
require "kobako"
|
|
@@ -55,12 +57,12 @@ sandbox.eval("1 + 2").value # => 3
|
|
|
55
57
|
```rust
|
|
56
58
|
use kobako::{Options, Sandbox};
|
|
57
59
|
|
|
58
|
-
// Options::default() is secure by default:
|
|
60
|
+
// Options::default() is secure by default: the same caps as Ruby, hermetic isolation.
|
|
59
61
|
let sandbox = Sandbox::new("kobako.wasm", Options::default())?;
|
|
60
62
|
sandbox.eval("1 + 2")?.value()?; // => Value::Int(3)
|
|
61
63
|
```
|
|
62
64
|
|
|
63
|
-
Each invocation returns the record of that one run
|
|
65
|
+
Each invocation returns the record of that one run: `Kobako::Execution` in Ruby, `Execution` in Rust. It carries the guest value alongside the run's output and resource usage. Nothing a run observes is stored on the Sandbox, so the same Sandbox serves concurrent invocations without them seeing each other.
|
|
64
66
|
|
|
65
67
|
The gem bundles its Guest Binary; a Rust host loads one explicitly — see [Frontends & Guest Binaries](#frontends--guest-binaries) for the packages and the Guest Binary variants.
|
|
66
68
|
|
|
@@ -75,31 +77,23 @@ host frontend Guest Binary
|
|
|
75
77
|
your own ──┘ └── kobako+json.wasm · your own
|
|
76
78
|
```
|
|
77
79
|
|
|
78
|
-
They compose freely
|
|
80
|
+
They compose freely, since any frontend loads any Guest Binary. A Ruby host can run a JSON-enabled guest, and a Rust host can run the pure default. How far down those choices go — and which ones a given starting point quietly makes for you — is laid out in [`docs/guides/assembly-levels.md`](docs/guides/assembly-levels.md).
|
|
79
81
|
|
|
80
82
|
### Host frontends
|
|
81
83
|
|
|
82
|
-
The host embeds the sandbox and owns the
|
|
84
|
+
The host embeds the sandbox and owns the wire codec. Choose by your host language and how much of the contract you want pre-assembled.
|
|
83
85
|
|
|
84
86
|
| Frontend | Package | Add it | Best for |
|
|
85
87
|
|----------|---------|--------|----------|
|
|
86
88
|
| Ruby gem | `kobako` (RubyGems) | `gem install kobako` | A Ruby host — Services, Handles, snippets, and pooling out of the box |
|
|
87
|
-
| Rust SDK | `kobako` (crates.io) | `
|
|
88
|
-
| Low-level crates | `kobako-wasmtime` + `kobako-runtime` + `kobako-transport` + `kobako-codec` |
|
|
89
|
+
| Rust SDK | `kobako` (crates.io) | `cargo add kobako` | A Rust host — the same behavior contract behind an idiomatic Rust API |
|
|
90
|
+
| Low-level crates | `kobako-wasmtime` + `kobako-runtime` + `kobako-transport` + `kobako-codec` | `cargo add` each | A custom host, or driving the wire directly in another language |
|
|
89
91
|
|
|
90
|
-
The Rust crates are documented on [crates.io](https://crates.io/crates/kobako); the Ruby gem is this README. Three runnable Rust hosts show the choice
|
|
92
|
+
The Rust crates are documented on [crates.io](https://crates.io/crates/kobako); the Ruby gem is this README. Three runnable Rust hosts show the choice. [`plugin-rs`](examples/plugin-rs) builds on the SDK, and [`wire-rs`](examples/wire-rs) assembles a host by hand on the low-level crates. [`fixed-schema-rs`](examples/fixed-schema-rs) replaces the payload schema both of those keep, with a guest and a host that agree on protobuf.
|
|
91
93
|
|
|
92
94
|
### Pre-built Guest Binaries
|
|
93
95
|
|
|
94
|
-
The gem bundles the pure `kobako.wasm`. Regexp and JSON are opt-in capabilities compiled into separate variants, each attached to every [GitHub Release](https://github.com/elct9620/kobako/releases). Download a variant and point your host at it — `Sandbox.new(wasm_path:)` in Ruby, `Sandbox::new(path, ...)` in Rust.
|
|
95
|
-
|
|
96
|
-
| Variant | File | Adds | Distribution |
|
|
97
|
-
|---------|------|------|--------------|
|
|
98
|
-
| Pure (default) | `kobako.wasm` | mruby + IO | Bundled in the gem, and a Release asset |
|
|
99
|
-
| +regexp | `kobako+regexp.wasm` | Regexp / MatchData (ASCII) | Release asset |
|
|
100
|
-
| +regexp-unicode | `kobako+regexp-unicode.wasm` | Regexp / MatchData (Unicode) | Release asset |
|
|
101
|
-
| +json | `kobako+json.wasm` | JSON | Release asset |
|
|
102
|
-
| +full | `kobako+full.wasm` | JSON + Regexp (ASCII) | Release asset |
|
|
96
|
+
The gem bundles the pure `kobako.wasm`. Regexp and JSON are opt-in capabilities compiled into separate variants, each attached to every [GitHub Release](https://github.com/elct9620/kobako/releases). Download a variant and point your host at it — `Sandbox.new(wasm_path:)` in Ruby, `Sandbox::new(path, ...)` in Rust. [`docs/guides/variants.md`](docs/guides/variants.md) lists every variant and what it adds.
|
|
103
97
|
|
|
104
98
|
```ruby
|
|
105
99
|
sandbox = Kobako::Sandbox.new(wasm_path: "kobako+full.wasm")
|
|
@@ -108,54 +102,40 @@ sandbox.eval('JSON.generate({ n: "42".to_i })').value # => "{\"n\":42}"
|
|
|
108
102
|
|
|
109
103
|
### Custom Guest Binaries
|
|
110
104
|
|
|
111
|
-
When no
|
|
112
|
-
|
|
113
|
-
| Guest crate | Role |
|
|
114
|
-
|-------------|------|
|
|
115
|
-
| `kobako-mruby` | mruby guest harness — the `MrbGuest` trait and provided flows |
|
|
116
|
-
| `kobako-core` | Guest ABI contract — the `Guest` trait and the `export_guest!` macro |
|
|
117
|
-
| `kobako-io` | IO / Kernel capability gem |
|
|
118
|
-
| `kobako-regexp` | Regexp / MatchData capability gem |
|
|
119
|
-
| `kobako-json` | JSON capability gem |
|
|
120
|
-
|
|
121
|
-
```rust
|
|
122
|
-
use beni::{Error, Mrb};
|
|
105
|
+
When no variant matches your capability set, build a Guest Binary of your own in Rust. A shell over `kobako-mruby` installs exactly the capability gems you choose: the shipped `kobako-io`, `kobako-regexp`, and `kobako-json`, or your own `beni::Gem`. `kobako-core` exports the wasm ABI. The Ruby gem drives it unchanged.
|
|
123
106
|
|
|
124
|
-
|
|
107
|
+
| To decide | Read |
|
|
108
|
+
|---|---|
|
|
109
|
+
| which parts you own at each level | [`docs/guides/assembly-levels.md`](docs/guides/assembly-levels.md) |
|
|
110
|
+
| what a guest shell implements | [`docs/guides/customization.md`](docs/guides/customization.md) |
|
|
111
|
+
| a working shell | [`wasm/kobako-wasm/src/guest.rs`](wasm/kobako-wasm/src/guest.rs) |
|
|
125
112
|
|
|
126
|
-
|
|
127
|
-
impl kobako_mruby::MrbGuest for MyGuest {
|
|
128
|
-
fn init_gems(mrb: &Mrb) -> Result<(), Error> {
|
|
129
|
-
mrb.init_gem::<kobako_io::KobakoIo>()?;
|
|
130
|
-
mrb.init_gem::<kobako_json::KobakoJson>()?;
|
|
131
|
-
Ok(())
|
|
132
|
-
}
|
|
133
|
-
}
|
|
113
|
+
Build the shell as a `cdylib` for `wasm32-wasip1`, then bake its boot state into the artifact with `kobako-baker input.wasm output.wasm`.
|
|
134
114
|
|
|
135
|
-
|
|
136
|
-
impl kobako_core::Guest for MyGuest {
|
|
137
|
-
fn eval() { <MyGuest as kobako_mruby::MrbGuest>::eval() }
|
|
138
|
-
fn run(env: &[u8]) { <MyGuest as kobako_mruby::MrbGuest>::run(env) }
|
|
139
|
-
fn yield_to_block(req: &[u8]) -> u64 { <MyGuest as kobako_mruby::MrbGuest>::yield_to_block(req) }
|
|
140
|
-
}
|
|
115
|
+
A shell may also define flags, switches the Host App sets for every invocation. A flag is critical by default, so a Guest Binary that does not recognise it raises `Kobako::UnsupportedFlagError` rather than run without it. Mark it `advisory: true` only when a script cannot tell the flag was skipped.
|
|
141
116
|
|
|
142
|
-
|
|
117
|
+
```ruby
|
|
118
|
+
sandbox.flag(:strict_mode) # critical: refused where unrecognised
|
|
119
|
+
sandbox.flag(:compile_cache, "on", advisory: true) # skipped where unrecognised
|
|
143
120
|
```
|
|
144
121
|
|
|
145
|
-
Build the crate as a `cdylib` for `wasm32-wasip1`, then bake the canonical boot state into the artifact (see [`CLAUDE.md`](CLAUDE.md) § Build Pipeline).
|
|
146
|
-
|
|
147
122
|
## Glossary
|
|
148
123
|
|
|
149
|
-
|
|
150
|
-
|
|
151
|
-
|
|
|
152
|
-
|
|
153
|
-
|
|
|
154
|
-
|
|
|
155
|
-
|
|
|
156
|
-
|
|
|
157
|
-
|
|
|
158
|
-
|
|
|
124
|
+
These terms name the concepts the rest of this README builds on.
|
|
125
|
+
|
|
126
|
+
| Term | Ruby class | Meaning |
|
|
127
|
+
|------|------------|---------|
|
|
128
|
+
| Sandbox | `Kobako::Sandbox` | The unit you configure once and invoke many times. |
|
|
129
|
+
| Invocation | — | One run of guest code, by `#eval` or `#run`. |
|
|
130
|
+
| Execution | `Kobako::Execution` | The frozen record an Invocation leaves. |
|
|
131
|
+
| Service | — | A host object the guest reaches by name, such as `MyService::KV`. |
|
|
132
|
+
| Mount | — | A `Kobako::Filesystem` placed at a guest path by `Sandbox#mount`, for the guest's `File`. |
|
|
133
|
+
| Context | `Kobako::Context` | The per-Invocation object the optional block receives. |
|
|
134
|
+
| Snippet | — | Guest source or bytecode replayed at the start of every Invocation. |
|
|
135
|
+
| Handle | `Kobako::Handle` | An opaque reference to a host object the wire cannot carry by value. |
|
|
136
|
+
| Block | — | Guest code passed alongside a Service call. |
|
|
137
|
+
|
|
138
|
+
A Sandbox keeps nothing from one Invocation to the next. An Execution holds what the run produced, wrote, and consumed: `#value`, output captures, and `#usage`. A failed run raises, carrying its Execution on the error's `#execution`. A Service or a Mount is the guest's only route to a host resource. A Context's `ctx.bind` supplies a Service object for that one run. A Handle names its object only within the Invocation that issued it. A Block stays in the guest; each `yield` to it is a synchronous round-trip.
|
|
159
139
|
|
|
160
140
|
## Usage
|
|
161
141
|
|
|
@@ -164,6 +144,7 @@ Each section below stands on its own; this is the order they build in.
|
|
|
164
144
|
| If you want to | Read |
|
|
165
145
|
|---|---|
|
|
166
146
|
| let guest code call into your host | [Services](#services), [Per-Invocation Bindings](#per-invocation-bindings), [Service Blocks](#service-blocks) |
|
|
147
|
+
| let guest code read and write files | [Mounts](#mounts) |
|
|
167
148
|
| see what a run produced or why it failed | [Output Capture](#output-capture), [Error Handling](#error-handling), [Invocation Lifecycle](#invocation-lifecycle) |
|
|
168
149
|
| bound what a run may consume | [Resource Limits](#resource-limits), [Concurrency](#concurrency), [Pooling](#pooling) |
|
|
169
150
|
| hand host objects across the boundary | [Handle Management](#handle-management) |
|
|
@@ -171,7 +152,7 @@ Each section below stands on its own; this is the order they build in.
|
|
|
171
152
|
|
|
172
153
|
### Services
|
|
173
154
|
|
|
174
|
-
`bind` any Ruby object as a Service at a constant-path name
|
|
155
|
+
`bind` any Ruby object as a Service at a constant-path name. The guest reaches it as a `MyService::KV` (or top-level `Clock`) proxy. Through the Transport wire, the guest invokes the public methods the object's own class defines.
|
|
175
156
|
|
|
176
157
|
```ruby
|
|
177
158
|
class User
|
|
@@ -191,11 +172,11 @@ sandbox.eval(<<~RUBY)
|
|
|
191
172
|
RUBY
|
|
192
173
|
```
|
|
193
174
|
|
|
194
|
-
Each `::`-separated path segment must match `/\A[A-Z]\w
|
|
175
|
+
Each `::`-separated path segment must match `/\A[A-Z]\w{0,65533}\z/`. A path under `Kobako`, which the Guest Binary keeps for itself, raises `ArgumentError`. Symbol kwargs travel transparently to the host method's keyword arguments. The registry seals at the first invocation (see [Invocation Lifecycle](#invocation-lifecycle)); later `#bind` raises `ArgumentError`.
|
|
195
176
|
|
|
196
177
|
### Per-Invocation Bindings
|
|
197
178
|
|
|
198
|
-
A setup-time `bind` fixes one object for the Sandbox's life.
|
|
179
|
+
A setup-time `bind` fixes one object for the Sandbox's life. Some objects belong to a single run instead: the current request, the acting user, a per-tenant store. For those, declare the path at setup and fill it per invocation. `bind(path)` with no object reserves the name as a *fillable*: the guest sees the constant, while an unfilled dispatch fails closed as `Kobako::ServiceError`. The optional `#eval` / `#run` block fills it.
|
|
199
180
|
|
|
200
181
|
```ruby
|
|
201
182
|
sandbox.bind("Req::Current") # declared, unfilled — stands for Kobako::Unresolved
|
|
@@ -203,11 +184,89 @@ sandbox.bind("Req::Current") # declared, unfilled — stands for Kobako::Unreso
|
|
|
203
184
|
sandbox.eval("Req::Current.user_id") { |ctx| ctx.bind("Req::Current", request) }
|
|
204
185
|
```
|
|
205
186
|
|
|
206
|
-
`ctx.bind` also shadows an already-bound path for that one run. The Context is spent once the block returns, so a `ctx` captured out of it raises `ArgumentError`. Choosing the object per invocation
|
|
187
|
+
`ctx.bind` also shadows an already-bound path for that one run. The Context is spent once the block returns, so a `ctx` captured out of it raises `ArgumentError`. Choosing the object per invocation lets concurrent Threads invoke a single Sandbox and still keep their identities apart. The [multi-tenant example](examples/multi-tenant/) runs that shape end to end.
|
|
188
|
+
|
|
189
|
+
### Mounts
|
|
190
|
+
|
|
191
|
+
`mount` places a `Kobako::Filesystem` at an absolute guest path, where the guest's `File` reads it. The default access is `:read`, so a write beneath it raises `Errno::EROFS` in the guest.
|
|
192
|
+
|
|
193
|
+
```ruby
|
|
194
|
+
sandbox.mount("/config", Kobako::Filesystem::Memory.new("app.yml" => yaml))
|
|
195
|
+
sandbox.mount("/tmp", Kobako::Filesystem::Temporary.new, access: :read_write)
|
|
196
|
+
sandbox.mount("/data", Kobako::Filesystem::Directory.new("var/data")) { |path, _access| path.end_with?(".csv") }
|
|
197
|
+
|
|
198
|
+
sandbox.eval('File.read("/config/app.yml")')
|
|
199
|
+
```
|
|
200
|
+
|
|
201
|
+
| Kind | Holds |
|
|
202
|
+
|---|---|
|
|
203
|
+
| `Memory.new(files = {}, limit:, inodes:)` | files in memory, shared by every invocation |
|
|
204
|
+
| `Temporary.new(limit:, inodes:)` | files in memory, empty at every invocation |
|
|
205
|
+
| `Directory.new(path)` | a host directory; nothing outside it is reachable |
|
|
206
|
+
| `Overlay.new.layer(fs, access:)` | Layers stacked into one view, the first added on top |
|
|
207
|
+
|
|
208
|
+
A Memory or Temporary is bounded as a tmpfs mount is. Its files hold 64 MiB unless `limit:` gives another size in bytes. It holds one file or directory for each 4 KiB of that limit, and its root, unless `inodes:` gives another count. `nil` lifts either bound, and the guest raises `Errno::ENOSPC` past one.
|
|
209
|
+
|
|
210
|
+
A block decides each open and each change that reaches the Mount in the access's place, asked at most once for each. A falsy answer or a `StandardError` fails that call as `Errno::EACCES`. Mounts seal with the Sandbox like a `bind`, and `ctx.mount` replaces one for a single run.
|
|
211
|
+
|
|
212
|
+
#### Guest File
|
|
213
|
+
|
|
214
|
+
The guest's `File.open` takes a mode String such as `"r"`, `"w+"`, or `"wx"`, or an Integer of `File::Constants` flags, valued as CRuby's on Linux. An exclusive create lets Sandboxes sharing a directory agree on a lock file. The critical flag `disable_file` removes `File` from the guest altogether.
|
|
215
|
+
|
|
216
|
+
Beneath a mount, `File.open` and `IO.new` answer as CRuby does over a Linux directory. Where they differ, they only refuse more:
|
|
217
|
+
|
|
218
|
+
| When | The guest sees |
|
|
219
|
+
|---|---|
|
|
220
|
+
| nothing is mounted | `NotImplementedError` |
|
|
221
|
+
| a block refuses a change | `Errno::EACCES` where a read-only mount says `Errno::EROFS` |
|
|
222
|
+
| a Ruby Filesystem's file without `append` is asked to append | `Errno::ENOTSUP` on the write |
|
|
223
|
+
| an Integer mode holds a flag kobako does not honor, or `WRONLY` beside `RDWR` | `Errno::EINVAL` |
|
|
224
|
+
| a mode names an encoding, such as `"r:utf-8"` | `ArgumentError` |
|
|
225
|
+
| `IO.new` wraps stdout or stderr in a mode other than `"w"` | `ArgumentError` |
|
|
226
|
+
| `disable_file` is set | no `File` constant |
|
|
227
|
+
|
|
228
|
+
#### Overlays
|
|
229
|
+
|
|
230
|
+
`layer` adds a Filesystem beneath the Layers already there and answers a new Overlay. Each Layer is `:read`, the default, or `:read_write` on its own.
|
|
231
|
+
|
|
232
|
+
```ruby
|
|
233
|
+
overlay = Kobako::Filesystem::Overlay.new
|
|
234
|
+
.layer(Kobako::Filesystem::Temporary.new, access: :read_write) # empty every run
|
|
235
|
+
.layer(Kobako::Filesystem::Directory.new("app")) # never written
|
|
236
|
+
sandbox.mount("/app", overlay, access: :read_write)
|
|
237
|
+
```
|
|
238
|
+
|
|
239
|
+
A change to a file in a writable Layer stays there. A read-only Layer's file is first copied up to the nearest writable Layer above it, and a new or renamed file lands in the topmost one. A lower writable Layer's file does not move, raising `Errno::EXDEV`, as across devices. A removal hides what a lower Layer keeps with an [OCI whiteout](https://github.com/opencontainers/image-spec/blob/main/layer.md#whiteouts), `.wh.<name>`, kept in the topmost writable Layer. Names starting with `.wh.`, in any case, are therefore reserved.
|
|
240
|
+
|
|
241
|
+
#### Filesystems in Ruby
|
|
242
|
+
|
|
243
|
+
Any object answering `open` and `stat` mounts as a Filesystem, directly or as a Layer. Its methods follow Ruby's `File`, `Dir`, and `IO`, so a real `::File` can answer as an open file.
|
|
244
|
+
|
|
245
|
+
| Method | Answers |
|
|
246
|
+
|---|---|
|
|
247
|
+
| `open(path, flags)` | an open file, below |
|
|
248
|
+
| `stat(path)` | an object answering `directory?`, `size`, and `ino`, as `File::Stat` does |
|
|
249
|
+
| `children(path)` | entry names, as `Dir.children` gives them |
|
|
250
|
+
| `mkdir` · `rmdir` · `unlink` · `rename(from, to, directory:)` | optional; without one, that change raises `Errno::EROFS` |
|
|
251
|
+
|
|
252
|
+
A `stat` may also answer `dev`; without one, the Filesystem stands for a single device. An `ino` names one directory while the guest holds it open, and names match exactly, as a block judges them. A path is relative to the mount, the empty path being its root, and `flags` are `File::Constants`. A `rename` told `directory: false` moves only a file and `true` only a directory; `nil` moves either. An `Errno` the guest has a class for reaches the script as itself, and any other `StandardError` as `Errno::EIO`. Every call re-enters Ruby, and its time counts toward `timeout:` without being interrupted.
|
|
253
|
+
|
|
254
|
+
##### Open Files
|
|
255
|
+
|
|
256
|
+
The file `open` answers takes the guest's reads and writes, each method named as `IO` names it.
|
|
257
|
+
|
|
258
|
+
| Method | Answers | Without it |
|
|
259
|
+
|---|---|---|
|
|
260
|
+
| `pread(length, offset)` · `size` | bytes, `nil` or `EOFError` at the end; the length | required |
|
|
261
|
+
| `pwrite(data, offset)` · `truncate(length)` | bytes written; nothing | that change raises `Errno::EROFS` |
|
|
262
|
+
| `append(data)` | where the file then ends, as `IO#pos` does | an append raises `Errno::ENOTSUP` |
|
|
263
|
+
| `fsync` · `close` | nothing | nothing to flush or release |
|
|
264
|
+
|
|
265
|
+
An `append` writes all of `data` at the end in one step, so another writer's append never lands over it.
|
|
207
266
|
|
|
208
267
|
### Output Capture
|
|
209
268
|
|
|
210
|
-
Guest writes through `puts` / `print` / `p` / `$stdout` / `$stderr` are buffered per-channel and read off the run's Execution, independently of its `#value
|
|
269
|
+
Guest writes through `puts` / `print` / `p` / `$stdout` / `$stderr` are buffered per-channel and read off the run's Execution, independently of its `#value`. Each invocation captures its own; overflow is clipped at the cap and flagged by `#stdout_truncated?` / `#stderr_truncated?`.
|
|
211
270
|
|
|
212
271
|
```ruby
|
|
213
272
|
execution = sandbox.eval(<<~RUBY)
|
|
@@ -229,69 +288,96 @@ Every invocation either returns an Execution or raises exactly one of three clas
|
|
|
229
288
|
begin
|
|
230
289
|
sandbox.eval(script)
|
|
231
290
|
rescue Kobako::TrapError
|
|
232
|
-
# Wasm engine fault or cap exhaustion.
|
|
291
|
+
# Wasm engine fault or cap exhaustion. The invocation is over.
|
|
233
292
|
rescue Kobako::ServiceError
|
|
234
293
|
# A host Service call failed and the script did not rescue it.
|
|
235
294
|
rescue Kobako::SandboxError => e
|
|
236
|
-
# The script raised, failed to compile, or returned an unrepresentable value
|
|
295
|
+
# The script raised, failed to compile, or returned an unrepresentable value,
|
|
296
|
+
# or the Guest Binary could not start the run.
|
|
237
297
|
logger.warn(e.execution.stderr) # the failed run's own captures and usage
|
|
238
298
|
end
|
|
239
299
|
```
|
|
240
300
|
|
|
241
|
-
Each of these carries the failed run's Execution on `#execution
|
|
301
|
+
Each of these carries the failed run's Execution on `#execution`. A rescue reads that run's output and usage exactly as a successful caller reads them off the returned one. `#failed?` keeps the two apart when both `#value` are `nil` — a script whose last expression was `nil` versus one that never produced a value. A failure caught before the guest ran at all — a host pre-flight refusal — leaves `#execution` `nil`, since there is no run to report.
|
|
242
302
|
|
|
243
303
|
| Class | Parent | Trigger |
|
|
244
304
|
|---------------------------------|----------------|------------------------------------------------------|
|
|
245
305
|
| `Kobako::TimeoutError` | `TrapError` | Per-invocation `timeout` exhausted |
|
|
246
306
|
| `Kobako::MemoryLimitError` | `TrapError` | Per-invocation `memory_limit` exhausted |
|
|
307
|
+
| `Kobako::NoServiceError` | `ServiceError` | Nothing bound there, a dead Handle, or a method the guest may not call |
|
|
308
|
+
| `Kobako::ServiceArgumentError` | `ServiceError` | The call would not bind, or the Service raised `ArgumentError` |
|
|
247
309
|
| `Kobako::HandleExhaustedError` | `SandboxError` | Handle counter reached its 2³¹ − 1 cap |
|
|
248
310
|
| `Kobako::BytecodeError` | `SandboxError` | `#preload(binary:)` failed RITE validation at replay |
|
|
311
|
+
| `Kobako::UnsupportedFlagError` | `SandboxError` | The Guest Binary does not recognise a critical `#flag`; no script ran |
|
|
249
312
|
| `Kobako::UndefinedEntrypointError` | `SandboxError` | `#run` named a constant no snippet defined; carries `#name` and `#available` |
|
|
250
313
|
|
|
251
314
|
`SandboxError` and `ServiceError` carry structured `origin` / `klass` / `backtrace_lines` fields when the guest produced a panic envelope.
|
|
252
315
|
|
|
316
|
+
#### Host Exceptions
|
|
317
|
+
|
|
318
|
+
A Service, a mount's block, or a Filesystem written in Ruby is your code running inside the run. What it raises splits by class.
|
|
319
|
+
|
|
320
|
+
| Raised by your code | Result |
|
|
321
|
+
|-----------------------------------------|-----------------------------------------------|
|
|
322
|
+
| a `StandardError` | the guest sees a failure it may rescue |
|
|
323
|
+
| `Interrupt`, `exit`, `throw`, `Timeout` | the run ends; the invocation raises it as is |
|
|
324
|
+
|
|
253
325
|
### Resource Limits
|
|
254
326
|
|
|
255
|
-
Each invocation enforces a wall-clock `timeout` and a per-invocation linear-memory `memory_limit`; exhaustion raises a `TrapError` subclass. Pass `nil` to `timeout` / `memory_limit` to disable that cap. Read [`Execution#usage`](lib/kobako/execution.rb) for actual consumption — populated on every outcome, so a rescued trap reports it just as a completed run does
|
|
327
|
+
Each invocation enforces a wall-clock `timeout` and a per-invocation linear-memory `memory_limit`; exhaustion raises a `TrapError` subclass. Pass `nil` to `timeout` / `memory_limit` to disable that cap. Read [`Execution#usage`](lib/kobako/execution.rb) for actual consumption — populated on every outcome, so a rescued trap reports it just as a completed run does.
|
|
256
328
|
|
|
257
329
|
```ruby
|
|
258
330
|
sandbox = Kobako::Sandbox.new(
|
|
259
331
|
timeout: 5.0, # seconds, default 60.0
|
|
260
|
-
|
|
332
|
+
setup_timeout: 2.0, # seconds, default 10.0
|
|
333
|
+
memory_limit: 10 * 1024 * 1024, # bytes, default 4 MiB
|
|
261
334
|
stdout_limit: 64 * 1024, # bytes, default 1 MiB
|
|
262
|
-
stderr_limit: 64 * 1024
|
|
335
|
+
stderr_limit: 64 * 1024,
|
|
336
|
+
open_file_limit: 16 # descriptors, default 1024
|
|
263
337
|
)
|
|
264
338
|
```
|
|
265
339
|
|
|
266
|
-
| Cap
|
|
267
|
-
|
|
268
|
-
| `timeout`
|
|
269
|
-
| `
|
|
270
|
-
| `
|
|
271
|
-
| `
|
|
340
|
+
| Cap | Raises | Default |
|
|
341
|
+
|-------------------|--------------------------------|---------|
|
|
342
|
+
| `timeout` | `Kobako::TimeoutError` | 60.0 s |
|
|
343
|
+
| `setup_timeout` | `Kobako::SetupError` at `new` | 10.0 s |
|
|
344
|
+
| `memory_limit` | `Kobako::MemoryLimitError` | 4 MiB |
|
|
345
|
+
| `stdout_limit` | output clipped (no raise) | 1 MiB |
|
|
346
|
+
| `stderr_limit` | output clipped (no raise) | 1 MiB |
|
|
347
|
+
| `open_file_limit` | `Errno::EMFILE` in the guest | 1024 |
|
|
272
348
|
|
|
273
|
-
`memory_limit` covers the per-invocation `memory.grow` delta from the entry baseline, so a Sandbox reused across invocations does not silently accumulate against a global budget.
|
|
349
|
+
`setup_timeout` bounds instantiating a Guest Binary and reading its ABI version, not compiling it. It applies the first time a path is loaded; later Sandboxes on that path skip the check. A Pool builds each Sandbox at its first checkout, so its `SetupError` raises from `with`. `memory_limit` covers the per-invocation `memory.grow` delta from the entry baseline, so a Sandbox reused across invocations does not silently accumulate against a global budget. `open_file_limit` counts every descriptor one invocation holds, the three standard ones and each Mount's included, so a guest cannot exhaust the host process's own.
|
|
350
|
+
|
|
351
|
+
#### Isolation Profile
|
|
352
|
+
|
|
353
|
+
Beyond the caps, `profile:` requests the Sandbox's isolation posture on a two-rung ladder.
|
|
354
|
+
|
|
355
|
+
```
|
|
356
|
+
:permissive < :hermetic (default)
|
|
357
|
+
live clock frozen clock
|
|
358
|
+
live entropy constant entropy
|
|
359
|
+
```
|
|
274
360
|
|
|
275
|
-
|
|
361
|
+
`:hermetic` denies the guest ambient time and entropy; `:permissive` lets the guest's `wasi:clocks` / `wasi:random` read live host sources, an explicit trade of reproducibility. The guest reaches no ambient filesystem, environment, or network under either. The request is also a floor: construction fails with `Kobako::SetupError` on a runtime that declares a weaker posture than requested. See [`docs/guides/security.md`](docs/guides/security.md) § Isolation Profiles.
|
|
276
362
|
|
|
277
363
|
### Concurrency
|
|
278
364
|
|
|
279
|
-
A Sandbox keeps no state from any run, so concurrent Threads may invoke distinct Sandboxes or share a single one
|
|
365
|
+
A Sandbox keeps no state from any run, so concurrent Threads may invoke distinct Sandboxes or share a single one. Each invocation owns its Handles, captures, and usage either way. One Thread still runs one invocation at a time. Sharing a Sandbox adds a single obligation.
|
|
280
366
|
|
|
281
367
|
| What you bind | Who reaches it | What it owes |
|
|
282
368
|
|---|---|---|
|
|
283
369
|
| Bound once at setup | every Thread sharing the Sandbox | it must itself be thread-safe |
|
|
284
370
|
| Supplied per invocation — `ctx.bind`, or an Extension `provider:` | that invocation alone | nothing |
|
|
285
371
|
|
|
286
|
-
####
|
|
372
|
+
#### GVL Modes
|
|
287
373
|
|
|
288
|
-
By default an invocation holds Ruby's GVL for its whole span, so guest execution across Threads serializes. `gvl: :release` drops the GVL for the guest span and re-acquires it for each guest→host dispatch, running guest code in parallel across Threads
|
|
374
|
+
By default an invocation holds Ruby's GVL for its whole span, so guest execution across Threads serializes. `gvl: :release` drops the GVL for the guest span and re-acquires it for each guest→host dispatch, running guest code in parallel across Threads.
|
|
289
375
|
|
|
290
376
|
```ruby
|
|
291
377
|
sandbox = Kobako::Sandbox.new(gvl: :release)
|
|
292
378
|
```
|
|
293
379
|
|
|
294
|
-
The mode is per-Sandbox and fixed at construction; it changes scheduling only, leaving isolation, Handle lifetimes, captures, and outcomes identical. `:hold` remains the default because releasing pays a handoff cost at every dispatch
|
|
380
|
+
The mode is per-Sandbox and fixed at construction; it changes scheduling only, leaving isolation, Handle lifetimes, captures, and outcomes identical. `:hold` remains the default because releasing pays a handoff cost at every dispatch. Compute-bound scripts scale with Thread count, while dispatch-heavy ones match or trail `:hold`. `rake bench:gvl_scheduling` measures both ends on your own hardware.
|
|
295
381
|
|
|
296
382
|
### Invocation Lifecycle
|
|
297
383
|
|
|
@@ -342,11 +428,11 @@ One Sandbox serves many invocations. Service bindings and preloaded snippets per
|
|
|
342
428
|
Services + snippets persist; invocation N+1 repeats.
|
|
343
429
|
```
|
|
344
430
|
|
|
345
|
-
For workloads that must be isolated from each other
|
|
431
|
+
For workloads that must be isolated from each other, construct a fresh `Kobako::Sandbox` per scope: per tenant, per student submission, per agent session. wasmtime's Engine and the compiled Module are cached at process scope, so additional Sandboxes amortize cold-start cost automatically.
|
|
346
432
|
|
|
347
433
|
### Pooling
|
|
348
434
|
|
|
349
|
-
For hosts that serve many short invocations, `Kobako::Pool` keeps a bounded set of warm, identically set-up Sandboxes
|
|
435
|
+
For hosts that serve many short invocations, `Kobako::Pool` keeps a bounded set of warm, identically set-up Sandboxes. It hands each one to a single exclusive holder at a time. Construction forwards every `Sandbox.new` keyword verbatim; the optional block is the per-Sandbox setup window and runs exactly once per constructed Sandbox.
|
|
350
436
|
|
|
351
437
|
```ruby
|
|
352
438
|
pool = Kobako::Pool.new(slots: 4) do |sandbox|
|
|
@@ -363,19 +449,19 @@ pool.with { |sandbox| sandbox.eval(%(KV::Lookup.call("user_42"))).value }
|
|
|
363
449
|
| `slots:` | Upper bound on constructed Sandboxes | required |
|
|
364
450
|
| `checkout_timeout:` | Seconds `#with` waits for a free Sandbox; `nil` waits indefinitely | 5.0 |
|
|
365
451
|
|
|
366
|
-
Sandboxes construct lazily on first demand. `#with` yields a Sandbox and returns the block's value
|
|
452
|
+
Sandboxes construct lazily on first demand. `#with` yields a Sandbox and returns the block's value, and at block exit the Sandbox returns to the pool, whatever the block raised. A checkout that waits past `checkout_timeout` raises `Kobako::PoolTimeoutError`. There is no teardown verb — a Pool releases everything with its own reachability.
|
|
367
453
|
|
|
368
|
-
####
|
|
454
|
+
#### Pool Trade-offs
|
|
369
455
|
|
|
370
456
|
| It gives you | It does not give you |
|
|
371
457
|
|---|---|
|
|
372
|
-
| warm, pre-configured Sandboxes and exclusive checkout | isolation
|
|
458
|
+
| warm, pre-configured Sandboxes and exclusive checkout | isolation (see [Concurrency](#concurrency)) |
|
|
373
459
|
|
|
374
|
-
`Kobako::Pool` is experimental today and is best treated as a convenience for warm, pre-configured reuse rather than a throughput optimisation. The build bakes the shared boot state into the artifact
|
|
460
|
+
A Sandbox holds no state from any run, so Threads sharing one are equally safe. `Kobako::Pool` is experimental today and is best treated as a convenience for warm, pre-configured reuse rather than a throughput optimisation. The build bakes the shared boot state into the artifact, and every dynamic script still compiles and runs per invocation. All a pool actually saves is the host-side `Sandbox.new` — now about 3 µs, an order of magnitude below the invocation that follows it. For the workload kobako is built for — many small, short-lived Sandboxes running dynamic scripts — that is not a gain worth the coupling.
|
|
375
461
|
|
|
376
462
|
### Service Blocks
|
|
377
463
|
|
|
378
|
-
A Service method can accept a guest-supplied block via `&blk` and `yield` into it. The block body runs inside the Wasm guest
|
|
464
|
+
A Service method can accept a guest-supplied block via `&blk` and `yield` into it. The block body runs inside the Wasm guest. Its `break` / `next` / `return` / exceptions follow normal Ruby semantics: an exit lands where it would through a plain method.
|
|
379
465
|
|
|
380
466
|
```ruby
|
|
381
467
|
sandbox.bind("Seq::Map", ->(items, &blk) { items.map(&blk) })
|
|
@@ -384,9 +470,32 @@ sandbox.eval('Seq::Map.call([1, 2, 3]) { |x| x * 2 }').value
|
|
|
384
470
|
# => [2, 4, 6]
|
|
385
471
|
```
|
|
386
472
|
|
|
473
|
+
A `yield` that cannot complete raises at the Service's call site, so the Service may rescue it. Left unrescued, a `BlockError` hands the guest its own exception back; the others fail the Service call.
|
|
474
|
+
|
|
475
|
+
| Class | Trigger at `yield` |
|
|
476
|
+
|----------------------------|---------------------------------------|
|
|
477
|
+
| `Kobako::BlockError` | The guest block raised |
|
|
478
|
+
| `Kobako::YieldValueError` | An argument cannot cross to the guest |
|
|
479
|
+
| `Kobako::UnsupportedError` | The Guest Binary serves no block |
|
|
480
|
+
| `ThreadError` | The block runs off the calling Fiber |
|
|
481
|
+
|
|
482
|
+
#### Block Fiber
|
|
483
|
+
|
|
484
|
+
A block runs only on the Fiber that called the Service. The guest's Wasm stack lives on that Fiber's native stack, so entering it from another raises `ThreadError` before the block runs. CRuby itself lets a block run anywhere; kobako narrows that to keep the guest's stack checks sound.
|
|
485
|
+
|
|
486
|
+
```
|
|
487
|
+
Service on the calling Fiber
|
|
488
|
+
├─ blk.call / yield ─> runs in the guest
|
|
489
|
+
├─ Thread.new { blk.call } ─> ThreadError
|
|
490
|
+
└─ Enumerator.new { |y| y << blk.call }
|
|
491
|
+
.next (runs its body on a Fiber) ─> ThreadError
|
|
492
|
+
```
|
|
493
|
+
|
|
494
|
+
`Enumerator#next` counts as another Fiber, because it runs the enumerator's body on one. Internal iteration such as `each` or `map` stays on the calling Fiber and runs the block normally.
|
|
495
|
+
|
|
387
496
|
### Handle Management
|
|
388
497
|
|
|
389
|
-
A
|
|
498
|
+
A host object the wire cannot represent may be returned from a Service, passed to `#run`, or handed back from the guest. It crosses the boundary as an opaque `Kobako::Handle` proxy and is restored to the original object before host code sees it. Any other unrepresentable value raises `Kobako::SandboxError`. Handles are scoped to a single invocation.
|
|
390
499
|
|
|
391
500
|
```ruby
|
|
392
501
|
class Greeter
|
|
@@ -400,11 +509,11 @@ sandbox.eval('Factory::Make.call("Bob").greet').value # => "hi, Bob" (Handle r
|
|
|
400
509
|
sandbox.eval('Factory::Make.call("Bob")').value # => #<Greeter @name="Bob"> (Handle restoration)
|
|
401
510
|
```
|
|
402
511
|
|
|
403
|
-
A `break` value from a guest block
|
|
512
|
+
A `break` value from a guest block never reaches host code. It stays in the guest while the Service unwinds, so a Handle in it stays a Handle.
|
|
404
513
|
|
|
405
|
-
####
|
|
514
|
+
#### Per-dispatch Handles
|
|
406
515
|
|
|
407
|
-
Each dispatch that hands back a non-wire-representable object allocates a *new* Handle — kobako never deduplicates by object identity
|
|
516
|
+
Each dispatch that hands back a non-wire-representable object allocates a *new* Handle — kobako never deduplicates by object identity. This is most visible with fluent / builder APIs. An `ActiveRecord::Relation` chain `spawn`s a fresh relation at each step, so every hop is an independent dispatch that binds its own Handle:
|
|
408
517
|
|
|
409
518
|
```
|
|
410
519
|
guest chain host (Catalog::Handles, one invocation)
|
|
@@ -420,11 +529,11 @@ Each dispatch that hands back a non-wire-representable object allocates a *new*
|
|
|
420
529
|
all stay live until the invocation ends, then reset together
|
|
421
530
|
```
|
|
422
531
|
|
|
423
|
-
This is deliberate, not a leak. Handle IDs run to 2³¹ − 1 per invocation and reset between invocations, so even deep chains stay far inside the range. Two consequences are worth keeping in mind
|
|
532
|
+
This is deliberate, not a leak. Handle IDs run to 2³¹ − 1 per invocation and reset between invocations, so even deep chains stay far inside the range. Two consequences are worth keeping in mind. The same host object handed back twice yields two *different* Handles, and the guest cannot tell they alias. Every intermediate Handle stays live until the invocation ends, since there is no per-Handle release.
|
|
424
533
|
|
|
425
534
|
### Snippets & Entrypoints
|
|
426
535
|
|
|
427
|
-
`Sandbox#preload` registers named mruby snippets that replay into every invocation's canonical boot state; `Sandbox#run(:Target, *args, **kwargs)` dispatches into a top-level `Object` constant defined by those snippets
|
|
536
|
+
`Sandbox#preload` registers named mruby snippets that replay into every invocation's canonical boot state; `Sandbox#run(:Target, *args, **kwargs)` dispatches into a top-level `Object` constant defined by those snippets.
|
|
428
537
|
|
|
429
538
|
```ruby
|
|
430
539
|
sandbox = Kobako::Sandbox.new
|
|
@@ -435,9 +544,9 @@ sandbox.run(:Adder, 2, 3).value # => 5
|
|
|
435
544
|
sandbox.run(:Greeter, name: "world").value # => "hello, world"
|
|
436
545
|
```
|
|
437
546
|
|
|
438
|
-
An entrypoint's `kwargs` arrive as a trailing positional Hash
|
|
547
|
+
An entrypoint's `kwargs` arrive as a trailing positional Hash, because mruby's C-side call path carries no keyword arguments. Declare a Hash parameter and unpack it yourself.
|
|
439
548
|
|
|
440
|
-
A target no snippet defined raises `Kobako::UndefinedEntrypointError
|
|
549
|
+
A target no snippet defined raises `Kobako::UndefinedEntrypointError`. Its `#available` lists the top-level constants the snippets did contribute, so the name is corrected from the error rather than by reading the guest source.
|
|
441
550
|
|
|
442
551
|
```
|
|
443
552
|
per-invocation replay (every #eval / #run, snippets in insertion order):
|
|
@@ -454,7 +563,7 @@ A target no snippet defined raises `Kobako::UndefinedEntrypointError`, whose `#a
|
|
|
454
563
|
return the Execution, then discard the instance
|
|
455
564
|
```
|
|
456
565
|
|
|
457
|
-
####
|
|
566
|
+
#### Payload Forms
|
|
458
567
|
|
|
459
568
|
`#preload` accepts two payload forms:
|
|
460
569
|
|
|
@@ -463,15 +572,15 @@ A target no snippet defined raises `Kobako::UndefinedEntrypointError`, whose `#a
|
|
|
463
572
|
| Source | `preload(code: "...", name: :Const)` | The `name:` keyword | First invocation's replay; compile errors raise `Kobako::SandboxError` |
|
|
464
573
|
| Bytecode | `preload(binary: bytes)` | Read from the bytecode's `debug_info` | First invocation's replay; structural failure raises `Kobako::BytecodeError` |
|
|
465
574
|
|
|
466
|
-
Use the source form for snippets authored in your repo; use the bytecode form when snippets ship as build artifacts from a separate `mrbc` pipeline. Both replay through the same per-invocation path, so no snippet content failure surfaces at `#preload` — force the first replay with a no-op invocation (e.g. `sandbox.eval("nil")`) when you want validation before real traffic.
|
|
575
|
+
Use the source form for snippets authored in your repo; use the bytecode form when snippets ship as build artifacts from a separate `mrbc` pipeline. Source is compiled once, before the first invocation, and replays as bytecode from then on, so the bytecode form saves only that one compile. Both replay through the same per-invocation path, so no snippet content failure surfaces at `#preload` — force the first replay with a no-op invocation (e.g. `sandbox.eval("nil")`) when you want validation before real traffic.
|
|
467
576
|
|
|
468
577
|
### Extensions
|
|
469
578
|
|
|
470
|
-
An Extension teaches the guest a native-style constant by pairing a guest idiom (`source`) with an optional host `backend`. `Sandbox#install` composes the two through the existing `#preload` and `#bind` verbs, adding no wire or Guest Binary surface
|
|
579
|
+
An Extension teaches the guest a native-style constant by pairing a guest idiom (`source`) with an optional host `backend`. `Sandbox#install` composes the two through the existing `#preload` and `#bind` verbs, adding no wire or Guest Binary surface. Pure operations run in-guest with no round-trip. The rest dispatch to the backend under the same isolation and reflection guarantees as any bound Service ([`docs/guides/extensions.md`](docs/guides/extensions.md)).
|
|
471
580
|
|
|
472
581
|
```ruby
|
|
473
|
-
|
|
474
|
-
class
|
|
582
|
+
VFS = <<~'MRUBY'
|
|
583
|
+
class Vfs
|
|
475
584
|
extend Kobako::Proxy
|
|
476
585
|
def self.basename(path) = path.split("/").last || "" # pure: runs in-guest
|
|
477
586
|
# read / write are undefined here, so they dispatch to the backend
|
|
@@ -480,16 +589,16 @@ MRUBY
|
|
|
480
589
|
|
|
481
590
|
sandbox.install(
|
|
482
591
|
Kobako::Extension.new(
|
|
483
|
-
name: :
|
|
484
|
-
source:
|
|
592
|
+
name: :Vfs,
|
|
593
|
+
source: VFS,
|
|
485
594
|
backend: Kobako::Extension::Backend.new(
|
|
486
|
-
path: "
|
|
487
|
-
provider: -> {
|
|
595
|
+
path: "Vfs",
|
|
596
|
+
provider: -> { CopyOnWriteFiles.new(root) } # invoked once per invocation
|
|
488
597
|
)
|
|
489
598
|
)
|
|
490
599
|
)
|
|
491
600
|
|
|
492
|
-
sandbox.eval('
|
|
601
|
+
sandbox.eval('Vfs.read("sample.txt")') # dispatches to the backend's #read
|
|
493
602
|
```
|
|
494
603
|
|
|
495
604
|
A backend declares the bound object's lifetime by keyword, never by inference — so a static object that happens to be callable stays unambiguous.
|
|
@@ -497,17 +606,17 @@ A backend declares the bound object's lifetime by keyword, never by inference
|
|
|
497
606
|
| Keyword | Lifetime |
|
|
498
607
|
|-------------|-----------------------------------------------------------------------------|
|
|
499
608
|
| `object:` | One object, shared by every invocation |
|
|
500
|
-
| `provider:` | A no-argument callable invoked per invocation
|
|
609
|
+
| `provider:` | A no-argument callable invoked per invocation |
|
|
501
610
|
| neither | A fillable, standing for `Kobako::Unresolved` until `ctx.bind` supplies the run's object |
|
|
502
611
|
|
|
503
|
-
kobako ships no concrete Extension; the idiom and backend are yours. The [overlay VFS example](examples/vfs/) is a worked
|
|
612
|
+
A writable backend takes `provider:`, so its state cannot leak across runs. kobako ships no concrete Extension; the idiom and backend are yours. The [overlay VFS example](examples/vfs/) is a worked virtual file that reads through to disk while protecting it from guest writes.
|
|
504
613
|
|
|
505
614
|
## Security
|
|
506
615
|
|
|
507
|
-
kobako isolates the guest, but **what it may reach is whatever you `bind
|
|
508
|
-
exposes every public method the object's own class defines. Inherited, mixed-in, and built-in
|
|
509
|
-
methods stay out of reach, but the class's own surface does not
|
|
510
|
-
|
|
616
|
+
kobako isolates the guest, but **what it may reach is whatever you `bind` or `mount`**.
|
|
617
|
+
`bind` exposes every public method the object's own class defines. Inherited, mixed-in, and built-in
|
|
618
|
+
methods stay out of reach, but the class's own surface does not. Bind a purpose-built object
|
|
619
|
+
scoped to the task, not a capable one whose other methods leak more than you intend.
|
|
511
620
|
|
|
512
621
|
```ruby
|
|
513
622
|
class ThemeReader # only #color is reachable; AppConfig.secret_key is not
|
|
@@ -517,30 +626,14 @@ end
|
|
|
517
626
|
sandbox = Kobako::Sandbox.new
|
|
518
627
|
sandbox.bind("Cfg::Settings", ThemeReader.new) # not: bind("Cfg::Settings", AppConfig)
|
|
519
628
|
|
|
520
|
-
sandbox.eval('Cfg::Settings.color').value # => "#3366ff" —
|
|
521
|
-
```
|
|
522
|
-
|
|
523
|
-
#### Gating an object's own surface
|
|
524
|
-
|
|
525
|
-
When a purpose-built wrapper is more than you need, an object can gate its own surface in
|
|
526
|
-
place: a private `respond_to_guest?(name)` answers, per method, whether the guest may call
|
|
527
|
-
it. Returning `false` for every name makes the object opaque — a credential the guest
|
|
528
|
-
forwards to another Service but never reads — while permitting a named subset exposes exactly those.
|
|
529
|
-
|
|
530
|
-
```ruby
|
|
531
|
-
class Credential
|
|
532
|
-
def initialize(token) = @token = token
|
|
533
|
-
def to_s = @token
|
|
534
|
-
|
|
535
|
-
private def respond_to_guest?(_name) = false # forwardable, never readable
|
|
536
|
-
end
|
|
629
|
+
sandbox.eval('Cfg::Settings.color').value # => "#3366ff" — any other method raises Kobako::NoServiceError
|
|
537
630
|
```
|
|
538
631
|
|
|
539
632
|
Guest code can name any `MyService::KV` path, but a forged name only resolves to
|
|
540
|
-
something you bound
|
|
541
|
-
|
|
542
|
-
|
|
543
|
-
and
|
|
633
|
+
something you bound. What you bind is the real authorization gate. An object can also
|
|
634
|
+
narrow its own surface with a private `respond_to_guest?`. Give each trust context its own
|
|
635
|
+
Sandbox. [`docs/guides/security.md`](docs/guides/security.md) covers the rest: self-gating
|
|
636
|
+
objects, validating untrusted input, default-deny external effects, and the return surface.
|
|
544
637
|
|
|
545
638
|
## Performance
|
|
546
639
|
|
|
@@ -549,25 +642,25 @@ Order-of-magnitude figures on macOS arm64, Ruby 3.4.7, YJIT off. Absolute values
|
|
|
549
642
|
| Phase | Cost |
|
|
550
643
|
|--------------------------------------------------------------|-----------------------|
|
|
551
644
|
| First `Sandbox.new` ever for a Guest Binary (Module JIT, then disk-cached) | ~500 ms once per machine |
|
|
552
|
-
| First `Sandbox.new` in a fresh process (`.cwasm` cache warm) | ~
|
|
645
|
+
| First `Sandbox.new` in a fresh process (`.cwasm` cache warm) | ~1.4 ms one-time |
|
|
553
646
|
| Subsequent `Sandbox.new` (caches warm) | ~3 µs |
|
|
554
|
-
| Warm `#eval` with one Service call on a reused Sandbox | ~
|
|
555
|
-
| Warm `#run(:Entrypoint, ...)` dispatch | ~
|
|
556
|
-
| Service call amortized inside one invocation | ~
|
|
557
|
-
| Snippet replay per invocation | ~7
|
|
558
|
-
| Per additional idle Sandbox (RSS) |
|
|
647
|
+
| Warm `#eval` with one Service call on a reused Sandbox | ~62 µs |
|
|
648
|
+
| Warm `#run(:Entrypoint, ...)` dispatch | ~57 µs |
|
|
649
|
+
| Service call amortized inside one invocation | ~4.9 µs |
|
|
650
|
+
| Snippet replay per invocation | ~2.7 µs each |
|
|
651
|
+
| Per additional idle Sandbox (RSS) | < 1 KB |
|
|
559
652
|
|
|
560
|
-
The Cranelift JIT runs once per machine and
|
|
653
|
+
The Cranelift JIT runs once per machine and wasmtime release — the compiled artifact persists in a `.cwasm` disk cache, so later processes deserialize in milliseconds. An idle Sandbox holds no wasm instance, which is why a thousand idle tenants cost ~37 MB total. The canonical boot state is baked into the artifact and instantiated per invocation. A +10% regression on any gated benchmark blocks release.
|
|
561
654
|
|
|
562
|
-
####
|
|
655
|
+
#### Throughput Levers
|
|
563
656
|
|
|
564
657
|
| Choice | What it changes |
|
|
565
658
|
|---|---|
|
|
566
|
-
| `gvl: :hold` (default) |
|
|
659
|
+
| `gvl: :hold` (default) | ~15–20k `#eval`/s in aggregate, at any Thread count |
|
|
567
660
|
| `gvl: :release` | lifts that ceiling for compute-bound scripts (see [Concurrency](#concurrency)) |
|
|
568
|
-
| `match?` over `=~` | `=~` (~5 µs/match) costs about
|
|
661
|
+
| `match?` over `=~` | `=~` (~5.6 µs/match) costs about 6× `match?` (~0.9 µs) |
|
|
569
662
|
|
|
570
|
-
Regexp is an opt-in capability gem, excluded from the default binary and the gated set
|
|
663
|
+
Under `:hold` the GVL serializes wasm work, though Ruby-side `#eval` setup still overlaps. `=~` eagerly builds the `MatchData` and match globals, so prefer `match?` for boolean tests. Regexp is an opt-in capability gem, excluded from the default binary and the gated set. Its throughput is tracked in a separate non-gated characterization (`#11` in [`benchmark/README.md`](benchmark/README.md)).
|
|
571
664
|
|
|
572
665
|
```bash
|
|
573
666
|
bundle exec rake bench # every gated regression benchmark (~5-8 min)
|
|
@@ -579,10 +672,10 @@ After checking out the repo:
|
|
|
579
672
|
|
|
580
673
|
```bash
|
|
581
674
|
bin/setup # install dependencies
|
|
582
|
-
bundle exec rake #
|
|
675
|
+
bundle exec rake # the release gate: compile, tests, rubocop, steep, gate checks
|
|
583
676
|
```
|
|
584
677
|
|
|
585
|
-
Building from source requires a WASI-capable Rust toolchain in addition to the standard host toolchain
|
|
678
|
+
Building from source requires a WASI-capable Rust toolchain in addition to the standard host toolchain. The first compile walks the full chain. The [beni](https://github.com/elct9620/beni) gem vendors wasi-sdk + mruby and builds `libmruby.a` (`rake beni:build`), then `rake wasm:build` produces the Guest Binary. See [`AGENTS.md`](AGENTS.md) for the rake task map and pipeline layout. `bin/console` opens an IRB session with the gem preloaded; `bundle exec rake install` installs the local checkout as a gem.
|
|
586
679
|
|
|
587
680
|
## Contributing
|
|
588
681
|
|