hyperhive/hive-c0re/src/server.rs

410 lines
16 KiB
Rust

use std::path::Path;
use std::sync::Arc;
use anyhow::{Context, Result};
use hive_sh4re::priv_proto::InfraAction;
use hive_sh4re::{HostRequest, HostResponse, LifecycleScope};
use tokio::io::{AsyncBufReadExt, AsyncWriteExt, BufReader};
use tokio::net::{UnixListener, UnixStream};
use crate::actions;
use crate::coordinator::Coordinator;
use crate::lifecycle;
pub async fn serve(socket: &Path, coord: Arc<Coordinator>) -> Result<()> {
// Prefer a socket passed by systemd socket-activation (LISTEN_FDS).
// When running under a `.socket` unit, systemd has already created,
// bound, and chmod-ed the socket for us — we just accept on it.
// Fall back to the traditional bind path when not socket-activated
// (direct invocation, dev, tests).
let listener = {
let mut listenfd = listenfd::ListenFd::from_env();
if let Some(std_listener) = listenfd
.take_unix_listener(0)
.context("take socket-activated unix listener")?
{
std_listener.set_nonblocking(true)?;
UnixListener::from_std(std_listener)
.context("convert socket-activated listener to tokio")?
} else {
// Standalone: create parent dir, remove any stale socket, bind.
if let Some(parent) = socket.parent() {
std::fs::create_dir_all(parent)
.with_context(|| format!("create socket parent {}", parent.display()))?;
}
if socket.exists() {
std::fs::remove_file(socket).context("remove stale socket")?;
}
UnixListener::bind(socket)
.with_context(|| format!("bind admin socket {}", socket.display()))?
}
};
tracing::info!(socket = %socket.display(), hyperhive_flake = %coord.hyperhive_flake, "hive-c0re admin listening");
loop {
let (stream, _) = listener.accept().await.context("accept connection")?;
let coord = coord.clone();
tokio::spawn(async move {
if let Err(e) = handle(stream, coord).await {
tracing::warn!(error = ?e, "connection failed");
}
});
}
}
async fn handle(stream: UnixStream, coord: Arc<Coordinator>) -> Result<()> {
let (read, mut write) = stream.into_split();
let mut reader = BufReader::new(read);
let mut line = String::new();
loop {
line.clear();
let n = reader.read_line(&mut line).await?;
if n == 0 {
return Ok(());
}
let resp = match serde_json::from_str::<HostRequest>(line.trim()) {
Ok(req) => dispatch(&req, coord.clone()).await,
Err(e) => HostResponse::error(format!("parse error: {e}")),
};
let mut payload = serde_json::to_string(&resp)?;
payload.push('\n');
write.write_all(payload.as_bytes()).await?;
write.flush().await?;
}
}
async fn dispatch(req: &HostRequest, coord: Arc<Coordinator>) -> HostResponse {
let result: anyhow::Result<HostResponse> = async {
Ok(match req {
HostRequest::Spawn { name } => handle_spawn(&coord, name).await?,
HostRequest::RequestSpawn { name } => {
tracing::info!(%name, "request_spawn");
let id =
coord
.approvals
.submit_kind(name, hive_sh4re::ApprovalKind::Spawn, "", None)?;
tracing::info!(%id, %name, "spawn approval queued");
HostResponse::success()
}
HostRequest::Kill { name } => handle_kill(&coord, name).await?,
HostRequest::Restart { name } => {
tracing::info!(%name, "restart");
lifecycle::restart(name).await?;
HostResponse::success()
}
HostRequest::RestartAll => handle_restart_all().await?,
HostRequest::Stop { scope, graceful } => {
// Resolve the scope to explicit container names at the entry
// point, then operate on names — never pass the bare "all
// agents" flag deeper (it'd force every consumer, incl. the
// graceful-stop queue, to re-expand it).
let agents = scoped_agents(scope).await?;
let infra = scoped_infra(scope);
handle_stop(&coord, &agents, &infra, *graceful).await?
}
HostRequest::Start { scope } => {
let agents = scoped_agents(scope).await?;
let infra = scoped_infra(scope);
handle_start(&agents, &infra).await?
}
HostRequest::Destroy { name, purge } => {
actions::destroy(&coord, name, *purge).await?;
HostResponse::success()
}
HostRequest::Rebuild { name } => handle_rebuild(&coord, name).await?,
HostRequest::List => HostResponse::list(lifecycle::list().await?),
HostRequest::Pending => HostResponse::pending(coord.approvals.pending()?),
HostRequest::Approve { id } => {
actions::approve(coord.clone(), *id).await?;
HostResponse::success()
}
HostRequest::Deny { id } => {
actions::deny(&coord, *id, None).await?;
HostResponse::success()
}
HostRequest::SetParent { child, new_parent } => {
tracing::info!(%child, ?new_parent, "set_parent");
// `reparent_with_notify` wraps `topology::set_parent`
// with the three notification messages + the
// ContainerView rescan. Idempotent same-parent calls
// skip both the messages and the disk write per the
// topology fast-path.
coord
.reparent_with_notify(child, new_parent.as_deref())
.await
.map_err(anyhow::Error::msg)?;
HostResponse::success()
}
})
}
.await;
match result {
Ok(r) => r,
Err(e) => HostResponse::error(format!("{e:#}")),
}
}
/// Create + start the container for `name`, rolling back socket
/// registration and notifying the manager on failure.
async fn handle_spawn(coord: &Arc<Coordinator>, name: &str) -> Result<HostResponse> {
tracing::info!(%name, "spawn");
let agent_dir = coord.ensure_runtime(name)?;
let hive = coord.hive_env();
let paths = Coordinator::agent_paths(name, agent_dir);
match lifecycle::spawn(name, &hive, &paths).await {
Ok(()) => {
coord.notify_manager(&hive_sh4re::HelperEvent::Spawned {
agent: name.to_owned(),
ok: true,
note: None,
sha: None,
});
}
Err(e) => {
// Roll back socket registration if container creation failed.
coord.unregister_agent(name);
coord.notify_manager(&hive_sh4re::HelperEvent::Spawned {
agent: name.to_owned(),
ok: false,
note: Some(format!("{e:#}")),
sha: None,
});
return Err(e);
}
}
Ok(HostResponse::success())
}
/// Kill `name`'s container, unregister its socket, notify the manager.
async fn handle_kill(coord: &Arc<Coordinator>, name: &str) -> Result<HostResponse> {
tracing::info!(%name, "kill");
lifecycle::kill(name).await?;
coord.unregister_agent(name);
coord.notify_manager(&hive_sh4re::HelperEvent::Killed {
agent: name.to_owned(),
});
Ok(HostResponse::success())
}
/// Restart every container, aggregating per-agent failures into one
/// response rather than aborting on the first error.
async fn handle_restart_all() -> Result<HostResponse> {
tracing::info!("restart-all");
let agents = lifecycle::list().await?;
let mut ok_agents: Vec<String> = Vec::new();
let mut errors: Vec<String> = Vec::new();
for agent in &agents {
if let Err(e) = lifecycle::restart(agent).await {
tracing::warn!(%agent, error = ?e, "restart-all: failed to restart agent");
errors.push(format!("{agent}: {e:#}"));
} else {
ok_agents.push(agent.clone());
}
}
if errors.is_empty() {
Ok(HostResponse::list(ok_agents))
} else {
Ok(HostResponse {
ok: false,
error: Some(errors.join("; ")),
agents: Some(ok_agents),
approvals: None,
})
}
}
/// Stop the given `agents` (resolved logical names) then `infra` containers
/// (`hivectl stop`). Agents go down before infra so they're not mid-request
/// against a forge/matrix that's already gone. Per-target failures are
/// aggregated rather than aborting on the first error, mirroring
/// `handle_restart_all`. Callers resolve the [`LifecycleScope`] to these
/// explicit name lists up front — this never sees the "all" flag.
///
/// A `graceful` stop enqueues a `QueueKind::GracefulStop` per agent (signal the
/// harness, run one stop-checkpoint turn, drain, then container stop, with a
/// timeout fallback to a hard stop), mirroring the dashboard `?graceful=1`
/// path. `graceful` applies to agents only - infra containers have no harness
/// turn loop, so they're always hard-stopped.
async fn handle_stop(
coord: &Arc<Coordinator>,
agents: &[String],
infra: &[&str],
graceful: bool,
) -> Result<HostResponse> {
tracing::info!(?agents, ?infra, graceful, "stop");
let mut ok_items: Vec<String> = Vec::new();
let mut errors: Vec<String> = Vec::new();
let mut enqueued_graceful = false;
for agent in agents {
if graceful {
// Graceful stop: enqueue the quiesce orchestration rather than a
// hard kill. Serialised through the rebuild queue so it can't race
// an in-flight rebuild for the same agent, and its per-step
// progress surfaces on the queue snapshot + build log.
coord.rebuild_queue.enqueue(
crate::rebuild_queue::QueueKind::GracefulStop,
agent.clone(),
crate::rebuild_queue::QueueSource::Manual,
"manual via hivectl graceful stop".to_owned(),
None,
);
ok_items.push(agent.clone());
enqueued_graceful = true;
continue;
}
match lifecycle::kill(agent).await {
Ok(()) => ok_items.push(agent.clone()),
Err(e) => {
tracing::warn!(%agent, error = ?e, "stop: agent kill failed");
errors.push(format!("{agent}: {e:#}"));
}
}
}
if enqueued_graceful {
coord.emit_rebuild_queue_snapshot();
}
for &container in infra {
match crate::priv_client::control_infra_container(container, InfraAction::Stop).await {
Ok(()) => ok_items.push(container.to_owned()),
Err(e) => {
tracing::warn!(%container, error = ?e, "stop: infra stop failed");
errors.push(format!("{container}: {e:#}"));
}
}
}
Ok(finish_lifecycle(ok_items, &errors))
}
/// Start the given `infra` containers then `agents` (`hivectl start`) — the
/// inverse of [`handle_stop`]. Infra comes up before agents so the agents
/// find forge/matrix/gateway ready. Per-target failures aggregated. Callers
/// resolve the [`LifecycleScope`] to these explicit name lists up front.
async fn handle_start(agents: &[String], infra: &[&str]) -> Result<HostResponse> {
tracing::info!(?agents, ?infra, "start");
let mut ok_items: Vec<String> = Vec::new();
let mut errors: Vec<String> = Vec::new();
for &container in infra {
match crate::priv_client::control_infra_container(container, InfraAction::Start).await {
Ok(()) => ok_items.push(container.to_owned()),
Err(e) => {
tracing::warn!(%container, error = ?e, "start: infra start failed");
errors.push(format!("{container}: {e:#}"));
}
}
}
for agent in agents {
match lifecycle::start(agent).await {
Ok(()) => ok_items.push(agent.clone()),
Err(e) => {
tracing::warn!(%agent, error = ?e, "start: agent start failed");
errors.push(format!("{agent}: {e:#}"));
}
}
}
Ok(finish_lifecycle(ok_items, &errors))
}
/// Resolve which sub-agent logical names a scope targets: every live
/// container (from `lifecycle::list`) when `agents` is set or the scope is
/// "everything", plus any explicit `agent_names`. Returns de-duplicated
/// logical names with the `h-` container prefix stripped.
async fn scoped_agents(scope: &LifecycleScope) -> Result<Vec<String>> {
use std::collections::BTreeSet;
let mut set: BTreeSet<String> = BTreeSet::new();
if scope.agents || scope.is_everything() {
for c in lifecycle::list().await? {
let logical = c
.strip_prefix(lifecycle::AGENT_PREFIX)
.unwrap_or(&c)
.to_owned();
set.insert(logical);
}
}
for n in &scope.agent_names {
set.insert(n.clone());
}
Ok(set.into_iter().collect())
}
/// Resolve which infra container names a scope targets. An "everything"
/// scope (no flags set) selects all controllable infra; otherwise each set
/// flag maps to its container. Fixed order for deterministic output.
fn scoped_infra(scope: &LifecycleScope) -> Vec<&'static str> {
let everything = scope.is_everything();
let mut out = Vec::new();
if everything || scope.ci {
out.push("hive-ci");
}
if everything || scope.forge {
out.push("hive-forge");
}
if everything || scope.gateway {
out.push("hive-gateway");
}
if everything || scope.matrix {
out.push("hive-matrix");
}
out
}
/// Build the aggregated lifecycle response: `ok` with the touched names when
/// every target succeeded, otherwise `ok: false` with the joined errors and
/// the partial success list (matches `handle_restart_all`).
fn finish_lifecycle(ok_items: Vec<String>, errors: &[String]) -> HostResponse {
if errors.is_empty() {
HostResponse::list(ok_items)
} else {
HostResponse {
ok: false,
error: Some(errors.join("; ")),
agents: Some(ok_items),
approvals: None,
}
}
}
/// Rebuild `name`'s container, notifying the manager of the outcome
/// (success or failure) and kicking the agent's next turn on success.
async fn handle_rebuild(coord: &Arc<Coordinator>, name: &str) -> Result<HostResponse> {
tracing::info!(%name, "rebuild");
let agent_dir = coord.ensure_runtime(name)?;
let hive = coord.hive_env();
let paths = Coordinator::agent_paths(name, agent_dir);
let result = lifecycle::rebuild(name, &hive, &paths, &|_| (), &|_| ()).await;
// Mirror auto_update::rebuild_agent — the manager wants to know
// about every rebuild attempt regardless of which surface triggered
// it, especially failures (build error → manager can adjust the
// agent's agent.nix). Without this the admin-socket CLI was a
// notify-gap.
match &result {
Ok(()) => {
coord.notify_manager(&hive_sh4re::HelperEvent::Rebuilt {
agent: name.to_owned(),
ok: true,
note: None,
sha: None,
tag: None,
});
// Wake the agent's next turn with the "you were rebuilt"
// hint. Same pattern as auto_update::rebuild_agent and the
// dashboard rebuild path — this is the CLI's equivalent.
coord.kick_agent(name, "container rebuilt");
}
Err(e) => coord.notify_manager(&hive_sh4re::HelperEvent::Rebuilt {
agent: name.to_owned(),
ok: false,
note: Some(format!("{e:#}")),
sha: None,
tag: None,
}),
}
result?;
Ok(HostResponse::success())
}