676c45bc93
`swarm/agents/<agent>/bao-mtls` did not exist, and neither did any per-agent identity at the secret store: `policy::agent_object_name`, `render_agent` and `render_agent_with_queue` had been written and never called outside their own tests. An agent's only "per-agent" secret today is read under the HIVE's certificate, through a wide grant on `swarm/agents/*` — so "per-agent" was presentational. The swarm now mints the certificate, so no hive ever needs the capability to mint one. `swarm-controller` is the service that does it: it already logs in to the store, and its existing grant already covers exactly the three objects written here (`create/update` on `secret/data/swarm/agents/*`, `sys/policies/acl/hive-*` and `auth/cert/certs/hive-*`). No new bao grant, and nothing co-located — a cert-auth role pins its authority by value, per role, so the controller issues from its own CA on its own host and pins that CA in the role it writes. No existing role changes. The mint node does not report success on a write. After publishing it connects again, with the leaf it just issued and under the role it just wrote, and reads the path back — so the policy, the role, the common name and the leaf are exercised in production on every agent creation. A certificate this code mints that the role this code writes will not accept turns the job node red at creation time instead of surfacing later as an agent container that cannot start. `TriggerDeploy` gains an `after_any` edge on the mint, not `after_ok`: a hive cannot pass down a certificate the swarm has not published, but a host with no authority configured must still create agents exactly as it does today. The private key is generated in memory and never written to disk on the controller — `SecretStore::connect_with_identity` takes the PEM the minter is already holding, so nothing is written out purely to be logged in with. Refs #4137
240 lines
9.1 KiB
TOML
240 lines
9.1 KiB
TOML
[workspace]
|
|
resolver = "3"
|
|
members = [
|
|
"hive-agent",
|
|
"hive-agent-mcp",
|
|
"hive-agent-sock",
|
|
"hive-core-agent-sock",
|
|
"hive-bash-mcp",
|
|
"hive-c0re",
|
|
"hive-screen-mcp",
|
|
"hive-forge",
|
|
"hive-forge-notify",
|
|
"hive-host-sock",
|
|
"hive-jobq",
|
|
"hive-jobq-metrics",
|
|
"hive-jobq-wire",
|
|
"hive-matrix-mcp",
|
|
"hive-metric",
|
|
"hive-priv",
|
|
"hive-priv-sock",
|
|
"hive-sh4re",
|
|
"hive-sock-client",
|
|
"hive-subagent-mcp",
|
|
"hive-types",
|
|
"hivectl",
|
|
"swarm-authelia-bridge",
|
|
"swarm-authelia-bridge-sock",
|
|
"swarm-controller",
|
|
"swarm-nats-auth",
|
|
"swarm-queue-client",
|
|
"swarm-logs",
|
|
"swarm-secret-client",
|
|
"swarmctl",
|
|
]
|
|
|
|
[workspace.package]
|
|
edition = "2024"
|
|
version = "0.1.0"
|
|
|
|
[workspace.lints.clippy]
|
|
# Pedantic is a hard error (locally + in CI): we want pedantic lints
|
|
# gated, so a toolchain bump that adds a new one reds the build until the
|
|
# code is updated rather than sliding in unnoticed. Priority -1 keeps the
|
|
# specific allows below winning over the group.
|
|
pedantic = { level = "deny", priority = -1 }
|
|
# Tolerated stylistic pedantic lints (noisy, not actionable).
|
|
missing_errors_doc = "allow"
|
|
missing_panics_doc = "allow"
|
|
module_name_repetitions = "allow"
|
|
|
|
[workspace.lints.rustdoc]
|
|
# Doc-link rot has no other discoverer: clippy does not read intra-doc
|
|
# links, `cargo test` does not, and nothing else builds docs. A `[`Foo`]`
|
|
# pointing at a renamed, moved or deleted item renders as plain text and
|
|
# misleads the next reader — worse than no link, since it names something
|
|
# and so sends them looking.
|
|
#
|
|
# Here rather than in `RUSTDOCFLAGS` on the CI check, so a plain local
|
|
# `cargo doc` fails the same way CI does. A gate you only meet in CI is a
|
|
# gate you meet too late.
|
|
broken_intra_doc_links = "deny"
|
|
private_intra_doc_links = "deny"
|
|
invalid_html_tags = "deny"
|
|
redundant_explicit_links = "deny"
|
|
bare_urls = "deny"
|
|
unescaped_backticks = "deny"
|
|
|
|
[workspace.dependencies]
|
|
anyhow = "1"
|
|
libc = "0.2"
|
|
axum = { version = "0.8", features = ["ws"] }
|
|
base64 = "0.22"
|
|
bcrypt = "0.19"
|
|
chrono = { version = "0.4", default-features = false, features = [
|
|
"clock",
|
|
"serde",
|
|
"std",
|
|
] }
|
|
clap = { version = "4", features = ["derive"] }
|
|
clap_complete = "4"
|
|
enumflags2 = { version = "0.7.12", features = ["serde"] }
|
|
indicatif = "0.18"
|
|
hive-sh4re = { path = "hive-sh4re" }
|
|
hive-agent-sock = { path = "hive-agent-sock" }
|
|
hive-jobq = { path = "hive-jobq" }
|
|
hive-jobq-metrics = { path = "hive-jobq-metrics" }
|
|
hive-jobq-wire = { path = "hive-jobq-wire" }
|
|
hive-core-agent-sock = { path = "hive-core-agent-sock" }
|
|
hive-claude = "0.1.1"
|
|
hive-host-sock = { path = "hive-host-sock" }
|
|
hive-priv-sock = { path = "hive-priv-sock" }
|
|
hive-sock-client = { path = "hive-sock-client" }
|
|
hive-types = { path = "hive-types" }
|
|
swarm-authelia-bridge-sock = { path = "swarm-authelia-bridge-sock" }
|
|
swarm-queue-client = { path = "swarm-queue-client" }
|
|
swarm-secret-client = { path = "swarm-secret-client" }
|
|
thiserror = "2"
|
|
vaultrs = "0.8"
|
|
# vaultrs's transport crates. Direct dependencies because one endpoint is
|
|
# defined here rather than by it; both versions must stay the ones vaultrs
|
|
# resolves, since its `exec_with_empty` takes *its* `Endpoint` trait.
|
|
rustify = "0.7"
|
|
rustify_derive = "0.5"
|
|
# Signs the per-agent client leaf `swarm-controller::agent_identity` mints.
|
|
# A library rather than an `openssl` shellout, which is what every other CA
|
|
# in this tree is (`glue-bao-tls.nix`, `hive-tls.nix`, `swarm-ca.nix`): those
|
|
# run once at deploy time and write to disk, this one runs per agent
|
|
# creation and must keep the private key it generates in memory long enough
|
|
# to log in with it and no longer. `ring` over `aws_lc_rs` because `ring` is
|
|
# already in the lock; the crypto backend is not otherwise load-bearing.
|
|
#
|
|
# `x509-parser`: `Issuer::from_ca_cert_pem` is gated behind it, and reading
|
|
# the authority back out of its own PEM is how a leaf inherits the subject and
|
|
# key identifier that make it chain. The ungated constructors take a
|
|
# `CertificateParams` the caller would have to restate by hand — two spellings
|
|
# of one authority, agreeing until the day they don't.
|
|
rcgen = { version = "0.14", features = ["x509-parser"] }
|
|
tower-http = { version = "0.7", features = ["fs"] }
|
|
uuid = { version = "1", features = ["v4"] }
|
|
rmcp = { version = "2", default-features = false, features = [
|
|
"server",
|
|
"macros",
|
|
"transport-io",
|
|
"transport-streamable-http-server",
|
|
] }
|
|
rusqlite = { version = "0.37" }
|
|
schemars = "1.0"
|
|
serde = { version = "1", features = ["derive"] }
|
|
serde_json = "1"
|
|
similar = "2"
|
|
# `derive`-only: `Kind`'s segment strings via `#[strum(serialize = "...")]`
|
|
# instead of a hand-written match, so the derive macro is the single source
|
|
# of truth (see `swarm-secret-client::path::Kind`).
|
|
strum = { version = "0.28.0", features = ["derive"] }
|
|
tokio = { version = "1", features = [
|
|
"fs",
|
|
"io-util",
|
|
"macros",
|
|
"net",
|
|
"process",
|
|
"rt-multi-thread",
|
|
"signal",
|
|
"sync",
|
|
"time",
|
|
] }
|
|
tokio-stream = { version = "0.1", features = ["sync"] }
|
|
tracing = "0.1"
|
|
tracing-subscriber = { version = "0.3", features = ["env-filter"] }
|
|
reqwest = { version = "0.13", default-features = false, features = [
|
|
# RFC 7662 introspection posts an urlencoded body; without this,
|
|
# `.form()` does not exist and the alternative is percent-encoding a
|
|
# credential by hand.
|
|
"form",
|
|
"json",
|
|
"rustls",
|
|
] }
|
|
hyper = { version = "1", features = ["client", "http1"] }
|
|
hyper-util = { version = "0.1", features = ["tokio"] }
|
|
# OTEL SDK, shared by every crate that pushes metrics (hive-c0re, hive-metric,
|
|
# swarm-controller). The blocking OTLP client is deliberate everywhere it's
|
|
# used: the metrics SDK's `PeriodicReader` drives its export from a
|
|
# background thread with no Tokio reactor, where the async client panics.
|
|
# `default-features = false` on the OTLP crate drops its own diagnostics
|
|
# ("internal-logs") unless a member opts back in — hive-c0re does, via
|
|
# `{ workspace = true, features = ["internal-logs"] }`.
|
|
opentelemetry = "0.32"
|
|
opentelemetry_sdk = { version = "0.32", features = ["metrics"] }
|
|
opentelemetry-otlp = { version = "0.32", default-features = false, features = [
|
|
"metrics",
|
|
"http-json",
|
|
"reqwest-blocking-client",
|
|
"reqwest-rustls",
|
|
] }
|
|
# The trait an OTLP exporter's HTTP transport is built on
|
|
# (`opentelemetry_otlp::WithHttpConfig::with_http_client`) — pulled in
|
|
# directly (not just transitively via `opentelemetry-otlp` above) by
|
|
# `swarm-controller`'s own `otel_http_client` module, which implements the
|
|
# trait itself rather than using the crate's own blanket `reqwest`/
|
|
# `reqwest-blocking` impls (this workspace's implementation needs to mint
|
|
# a fresh bearer token per request, which no blanket impl can do). No
|
|
# extra cargo features requested here for exactly that reason — the
|
|
# blanket impls live behind `reqwest`/`reqwest-blocking` features this
|
|
# workspace never turns on for this crate.
|
|
opentelemetry-http = "0.32"
|
|
# `async-trait`: `opentelemetry_http::HttpClient` predates stable
|
|
# `async fn` in traits and is still expressed with this macro upstream —
|
|
# implementing the trait (rather than only consuming the crate's own
|
|
# blanket impls) means writing `#[async_trait::async_trait]` on the impl
|
|
# block to match its expansion.
|
|
async-trait = "0.1"
|
|
# `bytes`/`http`: the payload and request/response types
|
|
# `opentelemetry_http::HttpClient::send_bytes` is expressed in terms of —
|
|
# needed to implement the trait, not just call something that already
|
|
# does.
|
|
bytes = "1"
|
|
http = "1"
|
|
http-body-util = "0.1"
|
|
# ⚠️ Keep at 0.11.1 or newer, and keep it on the SAME reqwest as everything
|
|
# else. 0.11.0 links reqwest 0.12 while the workspace is on 0.13, and cargo
|
|
# resolves features per (crate, VERSION): that older client got neither the
|
|
# workspace's TLS features nor the system trust store, so every https call to
|
|
# a gateway name died with `invalid peer certificate: UnknownIssuer` while
|
|
# clients built here worked (#3391). One version of reqwest in the tree is
|
|
# what keeps that class impossible rather than merely fixed.
|
|
forgejo-api = { version = "0.11.1", default-features = false, features = [
|
|
"rustls-tls",
|
|
] }
|
|
url = "2"
|
|
time = { version = "0.3", default-features = false, features = [
|
|
"formatting",
|
|
"parsing",
|
|
] }
|
|
petgraph = { version = "0.8", default-features = false, features = ["std"] }
|
|
matrix-sdk = { version = "0.18", default-features = false, features = [
|
|
"sqlite",
|
|
"markdown",
|
|
"e2e-encryption",
|
|
] }
|
|
futures-util = "0.3"
|
|
hmac = "0.13"
|
|
sha2 = "0.11"
|
|
# The NATS protocol client, for the swarm queue's auth-callout responder.
|
|
# `default-features = false` because the default set is broad - jetstream, kv,
|
|
# object-store, websockets, service - and a callout responder speaks none of
|
|
# them. What is named is the whole requirement: the server generation we
|
|
# deploy, nkey auth, and a TLS backend.
|
|
async-nats = { version = "0.50", default-features = false, features = [
|
|
"server_2_14",
|
|
"nkeys",
|
|
"ring",
|
|
] }
|
|
data-encoding = "2"
|
|
# The nkey *format* - ed25519 + base32 + CRC16. The primitives are already in
|
|
# the tree; the format is not, and hand-rolling a key format on an auth path
|
|
# is how you get a CRC bug nobody reviews.
|
|
nkeys = "0.4"
|
|
# A TEST ORACLE, not a runtime dependency - see swarm-nats-auth's respond.rs.
|
|
nats-jwt = "0.3"
|
|
utoipa = { version = "5", features = ["axum_extras", "chrono"] }
|
|
utoipa-axum = "0.2"
|