auto-update: surface pending updates in dashboard + include manager

This commit is contained in:
müde 2026-05-15 13:31:33 +02:00
parent a4e1556f90
commit e777576528
3 changed files with 149 additions and 51 deletions

View file

@ -11,22 +11,25 @@
use std::path::{Path, PathBuf};
use std::sync::Arc;
use anyhow::Result;
use anyhow::{Context, Result, bail};
use tokio::process::Command;
use crate::coordinator::Coordinator;
use crate::lifecycle::{self, AGENT_PREFIX};
use crate::lifecycle::{self, AGENT_PREFIX, MANAGER_NAME};
/// Marker file recording the hyperhive rev a sub-agent's container was last
/// built against. Sibling of `applied/<name>/` (rather than inside it) to
/// keep it out of the applied repo's git history.
fn rev_marker_path(name: &str) -> PathBuf {
/// keep it out of the applied repo's git history. Uses a leading dot so a
/// glob over `applied/*` doesn't include it.
pub fn rev_marker_path(name: &str) -> PathBuf {
PathBuf::from(format!("/var/lib/hyperhive/applied/.{name}.hyperhive-rev"))
}
/// Resolve the current rev of `hyperhive_flake`. For a path on disk we
/// canonicalize (following symlinks) so a /etc/hyperhive → /nix/store/...
/// update yields a different string. For anything else we return None.
fn current_flake_rev(hyperhive_flake: &str) -> Option<String> {
#[must_use]
pub fn current_flake_rev(hyperhive_flake: &str) -> Option<String> {
let path = Path::new(hyperhive_flake);
if !path.exists() {
return None;
@ -36,6 +39,62 @@ fn current_flake_rev(hyperhive_flake: &str) -> Option<String> {
.map(|p| p.display().to_string())
}
/// Read the marker for `name` and return whether the recorded rev matches
/// `current_rev`. Missing/unreadable marker counts as out-of-date.
#[must_use]
pub fn agent_needs_update(name: &str, current_rev: &str) -> bool {
let prev = std::fs::read_to_string(rev_marker_path(name))
.ok()
.map(|s| s.trim().to_owned());
prev.as_deref() != Some(current_rev)
}
/// Rebuild one sub-agent and refresh its marker. Used by both the startup
/// scanner and the dashboard's manual "update" button so the two paths
/// can't diverge.
pub async fn rebuild_agent(coord: &Arc<Coordinator>, name: &str, current_rev: &str) -> Result<()> {
tracing::info!(%name, rev = %current_rev, "rebuild agent");
let agent_dir = coord
.register_agent(name)
.with_context(|| format!("register_agent {name}"))?;
let applied_dir = Coordinator::agent_applied_dir(name);
let claude_dir = Coordinator::agent_claude_dir(name);
lifecycle::rebuild(
name,
&coord.hyperhive_flake,
&agent_dir,
&applied_dir,
&claude_dir,
)
.await?;
std::fs::write(rev_marker_path(name), current_rev)
.with_context(|| format!("write rev marker for {name}"))?;
Ok(())
}
/// Apply the manager's host-declared config: `nixos-container update hm1nd`
/// (no `--flake`) re-reads `/etc/nixos-containers/hm1nd.conf`, which the
/// host's `nixos-rebuild switch` rewrites to point at the new `SYSTEM_PATH`.
/// Idempotent when nothing has changed.
pub async fn rebuild_manager(current_rev: &str) -> Result<()> {
tracing::info!(rev = %current_rev, "rebuild manager (nixos-container update hm1nd)");
let out = Command::new("nixos-container")
.args(["update", MANAGER_NAME])
.output()
.await
.context("invoke nixos-container update hm1nd")?;
if !out.status.success() {
bail!(
"nixos-container update {MANAGER_NAME} failed ({}): {}",
out.status,
String::from_utf8_lossy(&out.stderr).trim()
);
}
std::fs::write(rev_marker_path(MANAGER_NAME), current_rev)
.with_context(|| format!("write rev marker for {MANAGER_NAME}"))?;
Ok(())
}
/// Rebuild every sub-agent whose marker differs from the current rev. Logs
/// per-agent outcomes and continues past failures. Returns Ok even if some
/// rebuilds failed — startup shouldn't be blocked by a broken agent.
@ -58,55 +117,38 @@ pub async fn run(coord: Arc<Coordinator>) -> Result<()> {
};
let mut tasks = Vec::new();
let mut manager_present = false;
for container in containers {
if container == MANAGER_NAME {
manager_present = true;
continue;
}
let Some(name) = container.strip_prefix(AGENT_PREFIX) else {
continue;
};
let name = name.to_owned();
let marker = rev_marker_path(&name);
let prev = std::fs::read_to_string(&marker).ok();
if prev.as_deref().map(str::trim) == Some(current_rev.as_str()) {
if !agent_needs_update(&name, &current_rev) {
tracing::debug!(%name, "auto-update: up-to-date");
continue;
}
let coord = coord.clone();
let current_rev = current_rev.clone();
tasks.push(tokio::spawn(async move {
tracing::info!(
%name,
prev = ?prev,
rev = %current_rev,
"auto-update: rebuilding agent",
);
let agent_dir = match coord.register_agent(&name) {
Ok(d) => d,
Err(e) => {
tracing::warn!(%name, error = ?e, "auto-update: register_agent failed");
return;
}
};
let applied_dir = Coordinator::agent_applied_dir(&name);
let claude_dir = Coordinator::agent_claude_dir(&name);
match lifecycle::rebuild(
&name,
&coord.hyperhive_flake,
&agent_dir,
&applied_dir,
&claude_dir,
)
.await
{
Ok(()) => {
if let Err(e) = std::fs::write(&marker, &current_rev) {
tracing::warn!(%name, error = ?e, "auto-update: write rev marker failed");
} else {
tracing::info!(%name, "auto-update: agent rebuilt");
}
}
Err(e) => {
tracing::warn!(%name, error = ?e, "auto-update: rebuild failed");
}
if let Err(e) = rebuild_agent(&coord, &name, &current_rev).await {
tracing::warn!(%name, error = ?e, "auto-update: rebuild failed");
}
}));
}
// Manager runs unconditionally when its marker differs: even if the host
// hasn't been rebuilt yet, `nixos-container update hm1nd` is a no-op, so
// there's no harm. The host's own activation already updates declarative
// containers — this is belt-and-braces for hive-c0re restarts.
if manager_present && agent_needs_update(MANAGER_NAME, &current_rev) {
let current_rev = current_rev.clone();
tasks.push(tokio::spawn(async move {
if let Err(e) = rebuild_manager(&current_rev).await {
tracing::warn!(error = ?e, "auto-update: manager rebuild failed");
}
}));
}