393 lines
14 KiB
Rust
Raw Normal View History

refactor(container): move companion UIs to systemd via Quadlet Companion UI containers (archy-bitcoin-ui, archy-lnd-ui, archy-electrs-ui) used to be launched as fire-and-forget tokio::spawn blocks from install.rs. If archipelago crashed mid-spawn or the container's cgroup was reaped, companions vanished from podman ps -a and only a manual rm/run could bring them back (the .228 incident). Now each companion is rendered as a Quadlet .container unit under ~/.config/containers/systemd/, daemon-reloaded, and started via systemctl --user. systemd owns supervision from that point on: - archipelago can crash, restart, or be uninstalled without touching any companion. - Quadlet's Restart=always + RestartSec=10 handles container exits. - A 30s reconcile tick in boot_reconciler enumerates expected companion units and re-installs any whose unit file or service vanished — defense-in-depth against external tampering. New module layout: - container/quadlet.rs: pure unit renderer + atomic write_if_changed + systemctl helpers (daemon_reload_user / enable_now / disable_remove / is_active). 6 unit tests, no I/O in the renderer. - container/companion.rs: per-app companion specs, install/remove/ reconcile, image presence (build local first, fall back to insecure registry only via image_uses_insecure_registry whitelist). 2 tests. install.rs handle_package_install now ends with a single call to companion::install_for(package_id), replacing 287 lines of spawn-and- hope shellouts plus a ~120-line nginx auth-injector helper that worked around per-node RPC password baking. The helper is gone too — the pre-start hook renders the per-node nginx.conf to /var/lib/archipelago/ bitcoin-ui/nginx.conf and the Quadlet unit bind-mounts it read-only. runtime.rs handle_package_uninstall now disables companions before the container rm loop. Otherwise systemd's Restart=always would respawn each companion within ~10s of removal. Tests: 53 container tests pass, including 6 quadlet renderer tests (host network, bridge network, capability set, atomic write idempotence) and 2 companion specs (per-app companion lookup, build_unit shape). boot_reconciler tests gain a #[cfg(test)] without_companion_stage() flag so the paused-clock fixtures don't race the real systemctl I/O. A bats regression test (companion-survives-archipelago-restart.bats, gated on ARCHY_ALLOW_DESTRUCTIVE=1) asserts the .228 failure mode cannot recur: every installed companion has a unit file, services stay active across systemctl --user restart archipelago, and a deleted unit file is recreated within one reconcile tick. Net delta: +941 / -363, but the +941 is mostly tests (~440 lines) and the new declarative layer; the imperative tokio::spawn block and its nginx-auth helper are gone, removing two failure classes (orphan companions on archipelago crash, and post-start exec races under tightly-confined cgroups) that previously needed manual SSH recovery. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 10:45:07 -04:00
//! Companion UI container lifecycle, entirely Quadlet-managed.
//!
//! A "companion" is a small nginx-based container that exposes a
//! browser-friendly UI on top of a headless backend service:
//!
//! | Backend | Companion | Purpose |
//! |------------------|--------------------|--------------------------|
//! | bitcoin-knots | archy-bitcoin-ui | RPC viewer |
//! | bitcoin-core | archy-bitcoin-ui | RPC viewer |
//! | lnd | archy-lnd-ui | wallet/channel UI |
//! | electrumx | archy-electrs-ui | indexer status UI |
//!
//! Lifecycle: `install` writes a Quadlet `.container` unit to
//! `~/.config/containers/systemd/`, daemon-reloads, then starts the
//! generated `.service`. systemd owns supervision from that point on
//! — archipelago can crash, restart, or be uninstalled without
//! touching the companion.
//!
//! This replaces the old `tokio::spawn { podman run }` block in
//! `install.rs` (~165 lines of fire-and-forget shellouts) with a
//! single declarative call.
use anyhow::{Context, Result};
use std::path::PathBuf;
use tokio::fs;
use tokio::process::Command;
use tracing::{info, warn};
refactor(install): route orchestrator-managed apps through orchestrator first Phase 3a of the install path consolidation. Two coupled changes: 1. install.rs handle_package_install: gate the legacy "container exists → adopt + return" probe on !orchestrator_managed. Apps the orchestrator knows about (bitcoin-knots, bitcoin-core, lnd, electrumx, fedimint, filebrowser, btcpay-server stack apps, mempool stack apps, plus the companion UIs that just moved to Quadlet) skip the legacy probe and fall straight into the orchestrator branch. The legacy adopt block was returning success on a bare `podman start` exit-0 — even when the process inside the container crashed seconds later. That's the .228 "running but unreachable" failure mode. The orchestrator's ensure_running honors the manifest's health check and pre-start hooks (e.g. re-renders bitcoin-ui's nginx.conf if the RPC password rotated), so this is a behavioral upgrade, not just a refactor. 2. ProdContainerOrchestrator::install: make idempotent. Previously it blindly called install_fresh which would fail on `podman create` if the container name already existed. Now it delegates to ensure_running: - Container Running + healthy → no-op (refresh hooks, restart if config rewritten) - Container Stopped/Exited → start (with hook refresh) - Container missing → install_fresh - Container in wedged state (Created/Paused/Unknown) → force-recreate Without this, change #1 would regress every "container already exists" case for the 18 orchestrator-managed app IDs. With it, install becomes the single source of truth for "make app X be in the desired state." Tests: 654 passed across the workspace (614 unit + 37 orchestration + 3 rpc), 0 failures. The 20 prod_orchestrator tests cover the install / ensure_running / reconcile paths the new install delegates through. Net delta: install.rs grows by ~30 lines (gating wrapper + comments), prod_orchestrator.rs grows by ~30 lines (idempotent install body). Both are temporary — the larger deletions (~1700 lines) come once every app has been verified through the orchestrator path in subsequent phases. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 11:12:52 -04:00
use crate::container::quadlet::{self, BindMount, NetworkMode, QuadletUnit};
refactor(container): move companion UIs to systemd via Quadlet Companion UI containers (archy-bitcoin-ui, archy-lnd-ui, archy-electrs-ui) used to be launched as fire-and-forget tokio::spawn blocks from install.rs. If archipelago crashed mid-spawn or the container's cgroup was reaped, companions vanished from podman ps -a and only a manual rm/run could bring them back (the .228 incident). Now each companion is rendered as a Quadlet .container unit under ~/.config/containers/systemd/, daemon-reloaded, and started via systemctl --user. systemd owns supervision from that point on: - archipelago can crash, restart, or be uninstalled without touching any companion. - Quadlet's Restart=always + RestartSec=10 handles container exits. - A 30s reconcile tick in boot_reconciler enumerates expected companion units and re-installs any whose unit file or service vanished — defense-in-depth against external tampering. New module layout: - container/quadlet.rs: pure unit renderer + atomic write_if_changed + systemctl helpers (daemon_reload_user / enable_now / disable_remove / is_active). 6 unit tests, no I/O in the renderer. - container/companion.rs: per-app companion specs, install/remove/ reconcile, image presence (build local first, fall back to insecure registry only via image_uses_insecure_registry whitelist). 2 tests. install.rs handle_package_install now ends with a single call to companion::install_for(package_id), replacing 287 lines of spawn-and- hope shellouts plus a ~120-line nginx auth-injector helper that worked around per-node RPC password baking. The helper is gone too — the pre-start hook renders the per-node nginx.conf to /var/lib/archipelago/ bitcoin-ui/nginx.conf and the Quadlet unit bind-mounts it read-only. runtime.rs handle_package_uninstall now disables companions before the container rm loop. Otherwise systemd's Restart=always would respawn each companion within ~10s of removal. Tests: 53 container tests pass, including 6 quadlet renderer tests (host network, bridge network, capability set, atomic write idempotence) and 2 companion specs (per-app companion lookup, build_unit shape). boot_reconciler tests gain a #[cfg(test)] without_companion_stage() flag so the paused-clock fixtures don't race the real systemctl I/O. A bats regression test (companion-survives-archipelago-restart.bats, gated on ARCHY_ALLOW_DESTRUCTIVE=1) asserts the .228 failure mode cannot recur: every installed companion has a unit file, services stay active across systemctl --user restart archipelago, and a deleted unit file is recreated within one reconcile tick. Net delta: +941 / -363, but the +941 is mostly tests (~440 lines) and the new declarative layer; the imperative tokio::spawn block and its nginx-auth helper are gone, removing two failure classes (orphan companions on archipelago crash, and post-start exec races under tightly-confined cgroups) that previously needed manual SSH recovery. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 10:45:07 -04:00
use archipelago_container::image_uses_insecure_registry;
const COMPANION_REGISTRY: &str = "146.59.87.168:3000/lfg2025";
/// Static description of one companion. The full list per backend
/// app_id lives in `companions_for`.
#[derive(Debug, Clone)]
pub struct CompanionSpec {
/// Container + unit name (e.g. "archy-bitcoin-ui").
pub name: &'static str,
/// Image base name in the lfg2025 registry namespace
/// (e.g. "bitcoin-ui" → "146.59.87.168:3000/lfg2025/bitcoin-ui:latest").
pub image_base: &'static str,
/// Filesystem locations to look for a local Dockerfile (build wins
/// over registry pull). Searched in order; first hit wins.
pub build_dir_candidates: &'static [&'static str],
/// Optional pre-start hook that renders config files referenced
/// by `bind_mounts`. Returns Ok(()) on success; bind-mount must
/// be present at start time or the companion will 502.
pub pre_start: Option<PreStartHook>,
/// Bind mounts. Always read-only — companions don't write to
/// host paths.
pub bind_mounts: &'static [(&'static str, &'static str)],
2026-05-05 11:29:18 -04:00
/// Host-to-container TCP ports for non-host-network companions.
pub ports: &'static [(u16, u16)],
/// Whether the companion must share the host network namespace.
pub host_network: bool,
refactor(container): move companion UIs to systemd via Quadlet Companion UI containers (archy-bitcoin-ui, archy-lnd-ui, archy-electrs-ui) used to be launched as fire-and-forget tokio::spawn blocks from install.rs. If archipelago crashed mid-spawn or the container's cgroup was reaped, companions vanished from podman ps -a and only a manual rm/run could bring them back (the .228 incident). Now each companion is rendered as a Quadlet .container unit under ~/.config/containers/systemd/, daemon-reloaded, and started via systemctl --user. systemd owns supervision from that point on: - archipelago can crash, restart, or be uninstalled without touching any companion. - Quadlet's Restart=always + RestartSec=10 handles container exits. - A 30s reconcile tick in boot_reconciler enumerates expected companion units and re-installs any whose unit file or service vanished — defense-in-depth against external tampering. New module layout: - container/quadlet.rs: pure unit renderer + atomic write_if_changed + systemctl helpers (daemon_reload_user / enable_now / disable_remove / is_active). 6 unit tests, no I/O in the renderer. - container/companion.rs: per-app companion specs, install/remove/ reconcile, image presence (build local first, fall back to insecure registry only via image_uses_insecure_registry whitelist). 2 tests. install.rs handle_package_install now ends with a single call to companion::install_for(package_id), replacing 287 lines of spawn-and- hope shellouts plus a ~120-line nginx auth-injector helper that worked around per-node RPC password baking. The helper is gone too — the pre-start hook renders the per-node nginx.conf to /var/lib/archipelago/ bitcoin-ui/nginx.conf and the Quadlet unit bind-mounts it read-only. runtime.rs handle_package_uninstall now disables companions before the container rm loop. Otherwise systemd's Restart=always would respawn each companion within ~10s of removal. Tests: 53 container tests pass, including 6 quadlet renderer tests (host network, bridge network, capability set, atomic write idempotence) and 2 companion specs (per-app companion lookup, build_unit shape). boot_reconciler tests gain a #[cfg(test)] without_companion_stage() flag so the paused-clock fixtures don't race the real systemctl I/O. A bats regression test (companion-survives-archipelago-restart.bats, gated on ARCHY_ALLOW_DESTRUCTIVE=1) asserts the .228 failure mode cannot recur: every installed companion has a unit file, services stay active across systemctl --user restart archipelago, and a deleted unit file is recreated within one reconcile tick. Net delta: +941 / -363, but the +941 is mostly tests (~440 lines) and the new declarative layer; the imperative tokio::spawn block and its nginx-auth helper are gone, removing two failure classes (orphan companions on archipelago crash, and post-start exec races under tightly-confined cgroups) that previously needed manual SSH recovery. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 10:45:07 -04:00
}
pub type PreStartHook = fn() -> futures_util::future::BoxFuture<'static, Result<()>>;
/// Companions to install when `package_id` lands. Empty for apps
/// without a companion UI.
pub fn companions_for(package_id: &str) -> &'static [CompanionSpec] {
match package_id {
"bitcoin" | "bitcoin-core" | "bitcoin-knots" => BITCOIN_UI,
"lnd" => LND_UI,
"electrumx" | "electrs" | "mempool-electrs" => ELECTRS_UI,
_ => &[],
}
}
const BITCOIN_UI: &[CompanionSpec] = &[CompanionSpec {
name: "archy-bitcoin-ui",
image_base: "bitcoin-ui",
build_dir_candidates: &[
"/opt/archipelago/docker/bitcoin-ui",
"/home/archipelago/archy/docker/bitcoin-ui",
"/home/archipelago/Projects/archy/docker/bitcoin-ui",
],
pre_start: Some(render_bitcoin_ui),
bind_mounts: &[(
"/var/lib/archipelago/bitcoin-ui/nginx.conf",
"/etc/nginx/conf.d/default.conf",
)],
2026-05-05 11:29:18 -04:00
ports: &[],
host_network: true,
refactor(container): move companion UIs to systemd via Quadlet Companion UI containers (archy-bitcoin-ui, archy-lnd-ui, archy-electrs-ui) used to be launched as fire-and-forget tokio::spawn blocks from install.rs. If archipelago crashed mid-spawn or the container's cgroup was reaped, companions vanished from podman ps -a and only a manual rm/run could bring them back (the .228 incident). Now each companion is rendered as a Quadlet .container unit under ~/.config/containers/systemd/, daemon-reloaded, and started via systemctl --user. systemd owns supervision from that point on: - archipelago can crash, restart, or be uninstalled without touching any companion. - Quadlet's Restart=always + RestartSec=10 handles container exits. - A 30s reconcile tick in boot_reconciler enumerates expected companion units and re-installs any whose unit file or service vanished — defense-in-depth against external tampering. New module layout: - container/quadlet.rs: pure unit renderer + atomic write_if_changed + systemctl helpers (daemon_reload_user / enable_now / disable_remove / is_active). 6 unit tests, no I/O in the renderer. - container/companion.rs: per-app companion specs, install/remove/ reconcile, image presence (build local first, fall back to insecure registry only via image_uses_insecure_registry whitelist). 2 tests. install.rs handle_package_install now ends with a single call to companion::install_for(package_id), replacing 287 lines of spawn-and- hope shellouts plus a ~120-line nginx auth-injector helper that worked around per-node RPC password baking. The helper is gone too — the pre-start hook renders the per-node nginx.conf to /var/lib/archipelago/ bitcoin-ui/nginx.conf and the Quadlet unit bind-mounts it read-only. runtime.rs handle_package_uninstall now disables companions before the container rm loop. Otherwise systemd's Restart=always would respawn each companion within ~10s of removal. Tests: 53 container tests pass, including 6 quadlet renderer tests (host network, bridge network, capability set, atomic write idempotence) and 2 companion specs (per-app companion lookup, build_unit shape). boot_reconciler tests gain a #[cfg(test)] without_companion_stage() flag so the paused-clock fixtures don't race the real systemctl I/O. A bats regression test (companion-survives-archipelago-restart.bats, gated on ARCHY_ALLOW_DESTRUCTIVE=1) asserts the .228 failure mode cannot recur: every installed companion has a unit file, services stay active across systemctl --user restart archipelago, and a deleted unit file is recreated within one reconcile tick. Net delta: +941 / -363, but the +941 is mostly tests (~440 lines) and the new declarative layer; the imperative tokio::spawn block and its nginx-auth helper are gone, removing two failure classes (orphan companions on archipelago crash, and post-start exec races under tightly-confined cgroups) that previously needed manual SSH recovery. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 10:45:07 -04:00
}];
const LND_UI: &[CompanionSpec] = &[CompanionSpec {
name: "archy-lnd-ui",
image_base: "lnd-ui",
build_dir_candidates: &[
"/opt/archipelago/docker/lnd-ui",
"/home/archipelago/archy/docker/lnd-ui",
"/home/archipelago/Projects/archy/docker/lnd-ui",
],
pre_start: None,
bind_mounts: &[],
2026-05-05 11:29:18 -04:00
ports: &[(18083, 80)],
host_network: false,
refactor(container): move companion UIs to systemd via Quadlet Companion UI containers (archy-bitcoin-ui, archy-lnd-ui, archy-electrs-ui) used to be launched as fire-and-forget tokio::spawn blocks from install.rs. If archipelago crashed mid-spawn or the container's cgroup was reaped, companions vanished from podman ps -a and only a manual rm/run could bring them back (the .228 incident). Now each companion is rendered as a Quadlet .container unit under ~/.config/containers/systemd/, daemon-reloaded, and started via systemctl --user. systemd owns supervision from that point on: - archipelago can crash, restart, or be uninstalled without touching any companion. - Quadlet's Restart=always + RestartSec=10 handles container exits. - A 30s reconcile tick in boot_reconciler enumerates expected companion units and re-installs any whose unit file or service vanished — defense-in-depth against external tampering. New module layout: - container/quadlet.rs: pure unit renderer + atomic write_if_changed + systemctl helpers (daemon_reload_user / enable_now / disable_remove / is_active). 6 unit tests, no I/O in the renderer. - container/companion.rs: per-app companion specs, install/remove/ reconcile, image presence (build local first, fall back to insecure registry only via image_uses_insecure_registry whitelist). 2 tests. install.rs handle_package_install now ends with a single call to companion::install_for(package_id), replacing 287 lines of spawn-and- hope shellouts plus a ~120-line nginx auth-injector helper that worked around per-node RPC password baking. The helper is gone too — the pre-start hook renders the per-node nginx.conf to /var/lib/archipelago/ bitcoin-ui/nginx.conf and the Quadlet unit bind-mounts it read-only. runtime.rs handle_package_uninstall now disables companions before the container rm loop. Otherwise systemd's Restart=always would respawn each companion within ~10s of removal. Tests: 53 container tests pass, including 6 quadlet renderer tests (host network, bridge network, capability set, atomic write idempotence) and 2 companion specs (per-app companion lookup, build_unit shape). boot_reconciler tests gain a #[cfg(test)] without_companion_stage() flag so the paused-clock fixtures don't race the real systemctl I/O. A bats regression test (companion-survives-archipelago-restart.bats, gated on ARCHY_ALLOW_DESTRUCTIVE=1) asserts the .228 failure mode cannot recur: every installed companion has a unit file, services stay active across systemctl --user restart archipelago, and a deleted unit file is recreated within one reconcile tick. Net delta: +941 / -363, but the +941 is mostly tests (~440 lines) and the new declarative layer; the imperative tokio::spawn block and its nginx-auth helper are gone, removing two failure classes (orphan companions on archipelago crash, and post-start exec races under tightly-confined cgroups) that previously needed manual SSH recovery. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 10:45:07 -04:00
}];
const ELECTRS_UI: &[CompanionSpec] = &[CompanionSpec {
name: "archy-electrs-ui",
image_base: "electrs-ui",
build_dir_candidates: &[
"/opt/archipelago/docker/electrs-ui",
"/home/archipelago/archy/docker/electrs-ui",
"/home/archipelago/Projects/archy/docker/electrs-ui",
],
pre_start: None,
bind_mounts: &[],
2026-05-05 11:29:18 -04:00
ports: &[],
host_network: true,
refactor(container): move companion UIs to systemd via Quadlet Companion UI containers (archy-bitcoin-ui, archy-lnd-ui, archy-electrs-ui) used to be launched as fire-and-forget tokio::spawn blocks from install.rs. If archipelago crashed mid-spawn or the container's cgroup was reaped, companions vanished from podman ps -a and only a manual rm/run could bring them back (the .228 incident). Now each companion is rendered as a Quadlet .container unit under ~/.config/containers/systemd/, daemon-reloaded, and started via systemctl --user. systemd owns supervision from that point on: - archipelago can crash, restart, or be uninstalled without touching any companion. - Quadlet's Restart=always + RestartSec=10 handles container exits. - A 30s reconcile tick in boot_reconciler enumerates expected companion units and re-installs any whose unit file or service vanished — defense-in-depth against external tampering. New module layout: - container/quadlet.rs: pure unit renderer + atomic write_if_changed + systemctl helpers (daemon_reload_user / enable_now / disable_remove / is_active). 6 unit tests, no I/O in the renderer. - container/companion.rs: per-app companion specs, install/remove/ reconcile, image presence (build local first, fall back to insecure registry only via image_uses_insecure_registry whitelist). 2 tests. install.rs handle_package_install now ends with a single call to companion::install_for(package_id), replacing 287 lines of spawn-and- hope shellouts plus a ~120-line nginx auth-injector helper that worked around per-node RPC password baking. The helper is gone too — the pre-start hook renders the per-node nginx.conf to /var/lib/archipelago/ bitcoin-ui/nginx.conf and the Quadlet unit bind-mounts it read-only. runtime.rs handle_package_uninstall now disables companions before the container rm loop. Otherwise systemd's Restart=always would respawn each companion within ~10s of removal. Tests: 53 container tests pass, including 6 quadlet renderer tests (host network, bridge network, capability set, atomic write idempotence) and 2 companion specs (per-app companion lookup, build_unit shape). boot_reconciler tests gain a #[cfg(test)] without_companion_stage() flag so the paused-clock fixtures don't race the real systemctl I/O. A bats regression test (companion-survives-archipelago-restart.bats, gated on ARCHY_ALLOW_DESTRUCTIVE=1) asserts the .228 failure mode cannot recur: every installed companion has a unit file, services stay active across systemctl --user restart archipelago, and a deleted unit file is recreated within one reconcile tick. Net delta: +941 / -363, but the +941 is mostly tests (~440 lines) and the new declarative layer; the imperative tokio::spawn block and its nginx-auth helper are gone, removing two failure classes (orphan companions on archipelago crash, and post-start exec races under tightly-confined cgroups) that previously needed manual SSH recovery. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 10:45:07 -04:00
}];
fn render_bitcoin_ui() -> futures_util::future::BoxFuture<'static, Result<()>> {
Box::pin(async {
let paths = crate::container::bitcoin_ui::RenderPaths::default();
crate::container::bitcoin_ui::render(&paths)
.await
.map(|_| ())
.context("render bitcoin-ui nginx.conf")
})
}
/// Provision and start every companion for `package_id`. Each
/// companion is independent — a failure in one is logged but does
/// not abort the others.
pub async fn install_for(package_id: &str) -> Vec<(String, anyhow::Error)> {
let mut failures = Vec::new();
for spec in companions_for(package_id) {
if let Err(e) = install_one(spec).await {
warn!(companion = spec.name, error = %e, "companion install failed");
failures.push((spec.name.to_string(), e));
}
}
failures
}
/// Stop and remove every companion for `package_id`. Best effort:
/// errors are logged but do not abort the sequence.
pub async fn remove_for(package_id: &str) {
let dir = match quadlet::unit_dir().await {
Ok(d) => d,
Err(e) => {
warn!("companion remove: cannot resolve quadlet dir: {e:#}");
return;
}
};
for spec in companions_for(package_id) {
if let Err(e) = quadlet::disable_remove(spec.name, &dir).await {
warn!(companion = spec.name, error = %e, "companion remove failed");
}
}
}
/// Provision one companion: pre-start hook → image present → write
/// quadlet → daemon-reload → start.
pub async fn install_one(spec: &CompanionSpec) -> Result<()> {
if let Some(hook) = spec.pre_start {
hook().await.with_context(|| {
refactor(install): route orchestrator-managed apps through orchestrator first Phase 3a of the install path consolidation. Two coupled changes: 1. install.rs handle_package_install: gate the legacy "container exists → adopt + return" probe on !orchestrator_managed. Apps the orchestrator knows about (bitcoin-knots, bitcoin-core, lnd, electrumx, fedimint, filebrowser, btcpay-server stack apps, mempool stack apps, plus the companion UIs that just moved to Quadlet) skip the legacy probe and fall straight into the orchestrator branch. The legacy adopt block was returning success on a bare `podman start` exit-0 — even when the process inside the container crashed seconds later. That's the .228 "running but unreachable" failure mode. The orchestrator's ensure_running honors the manifest's health check and pre-start hooks (e.g. re-renders bitcoin-ui's nginx.conf if the RPC password rotated), so this is a behavioral upgrade, not just a refactor. 2. ProdContainerOrchestrator::install: make idempotent. Previously it blindly called install_fresh which would fail on `podman create` if the container name already existed. Now it delegates to ensure_running: - Container Running + healthy → no-op (refresh hooks, restart if config rewritten) - Container Stopped/Exited → start (with hook refresh) - Container missing → install_fresh - Container in wedged state (Created/Paused/Unknown) → force-recreate Without this, change #1 would regress every "container already exists" case for the 18 orchestrator-managed app IDs. With it, install becomes the single source of truth for "make app X be in the desired state." Tests: 654 passed across the workspace (614 unit + 37 orchestration + 3 rpc), 0 failures. The 20 prod_orchestrator tests cover the install / ensure_running / reconcile paths the new install delegates through. Net delta: install.rs grows by ~30 lines (gating wrapper + comments), prod_orchestrator.rs grows by ~30 lines (idempotent install body). Both are temporary — the larger deletions (~1700 lines) come once every app has been verified through the orchestrator path in subsequent phases. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 11:12:52 -04:00
format!(
"pre-start hook failed for {} — companion will not start",
spec.name
)
refactor(container): move companion UIs to systemd via Quadlet Companion UI containers (archy-bitcoin-ui, archy-lnd-ui, archy-electrs-ui) used to be launched as fire-and-forget tokio::spawn blocks from install.rs. If archipelago crashed mid-spawn or the container's cgroup was reaped, companions vanished from podman ps -a and only a manual rm/run could bring them back (the .228 incident). Now each companion is rendered as a Quadlet .container unit under ~/.config/containers/systemd/, daemon-reloaded, and started via systemctl --user. systemd owns supervision from that point on: - archipelago can crash, restart, or be uninstalled without touching any companion. - Quadlet's Restart=always + RestartSec=10 handles container exits. - A 30s reconcile tick in boot_reconciler enumerates expected companion units and re-installs any whose unit file or service vanished — defense-in-depth against external tampering. New module layout: - container/quadlet.rs: pure unit renderer + atomic write_if_changed + systemctl helpers (daemon_reload_user / enable_now / disable_remove / is_active). 6 unit tests, no I/O in the renderer. - container/companion.rs: per-app companion specs, install/remove/ reconcile, image presence (build local first, fall back to insecure registry only via image_uses_insecure_registry whitelist). 2 tests. install.rs handle_package_install now ends with a single call to companion::install_for(package_id), replacing 287 lines of spawn-and- hope shellouts plus a ~120-line nginx auth-injector helper that worked around per-node RPC password baking. The helper is gone too — the pre-start hook renders the per-node nginx.conf to /var/lib/archipelago/ bitcoin-ui/nginx.conf and the Quadlet unit bind-mounts it read-only. runtime.rs handle_package_uninstall now disables companions before the container rm loop. Otherwise systemd's Restart=always would respawn each companion within ~10s of removal. Tests: 53 container tests pass, including 6 quadlet renderer tests (host network, bridge network, capability set, atomic write idempotence) and 2 companion specs (per-app companion lookup, build_unit shape). boot_reconciler tests gain a #[cfg(test)] without_companion_stage() flag so the paused-clock fixtures don't race the real systemctl I/O. A bats regression test (companion-survives-archipelago-restart.bats, gated on ARCHY_ALLOW_DESTRUCTIVE=1) asserts the .228 failure mode cannot recur: every installed companion has a unit file, services stay active across systemctl --user restart archipelago, and a deleted unit file is recreated within one reconcile tick. Net delta: +941 / -363, but the +941 is mostly tests (~440 lines) and the new declarative layer; the imperative tokio::spawn block and its nginx-auth helper are gone, removing two failure classes (orphan companions on archipelago crash, and post-start exec races under tightly-confined cgroups) that previously needed manual SSH recovery. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 10:45:07 -04:00
})?;
}
let image = ensure_image_present(spec).await?;
let unit = build_unit(spec, &image);
let dir = quadlet::unit_dir().await?;
let changed = quadlet::write_if_changed(&unit, &dir).await?;
if changed {
info!(companion = spec.name, "wrote quadlet unit");
quadlet::daemon_reload_user().await?;
}
// Start is idempotent — if already running, systemctl returns 0.
quadlet::enable_now(&unit.service_name()).await?;
info!(companion = spec.name, "companion started");
Ok(())
}
/// Build companion image locally if a Dockerfile exists, otherwise
/// pull from the lfg2025 registry. Returns the image ref the quadlet
/// should reference (`localhost/<base>:latest` for build, registry
/// URL for pull).
async fn ensure_image_present(spec: &CompanionSpec) -> Result<String> {
let local_image = format!("localhost/{}:latest", spec.image_base);
let registry_image = format!("{}/{}:latest", COMPANION_REGISTRY, spec.image_base);
// Prefer local build — companions can carry build-time customizations
// (e.g. nginx.conf templates baked in). Search known candidates.
for dir in spec.build_dir_candidates {
let dockerfile = PathBuf::from(dir).join("Dockerfile");
if fs::try_exists(&dockerfile).await.unwrap_or(false) {
2026-05-05 11:29:18 -04:00
if image_exists(&local_image).await {
return Ok(local_image);
}
refactor(container): move companion UIs to systemd via Quadlet Companion UI containers (archy-bitcoin-ui, archy-lnd-ui, archy-electrs-ui) used to be launched as fire-and-forget tokio::spawn blocks from install.rs. If archipelago crashed mid-spawn or the container's cgroup was reaped, companions vanished from podman ps -a and only a manual rm/run could bring them back (the .228 incident). Now each companion is rendered as a Quadlet .container unit under ~/.config/containers/systemd/, daemon-reloaded, and started via systemctl --user. systemd owns supervision from that point on: - archipelago can crash, restart, or be uninstalled without touching any companion. - Quadlet's Restart=always + RestartSec=10 handles container exits. - A 30s reconcile tick in boot_reconciler enumerates expected companion units and re-installs any whose unit file or service vanished — defense-in-depth against external tampering. New module layout: - container/quadlet.rs: pure unit renderer + atomic write_if_changed + systemctl helpers (daemon_reload_user / enable_now / disable_remove / is_active). 6 unit tests, no I/O in the renderer. - container/companion.rs: per-app companion specs, install/remove/ reconcile, image presence (build local first, fall back to insecure registry only via image_uses_insecure_registry whitelist). 2 tests. install.rs handle_package_install now ends with a single call to companion::install_for(package_id), replacing 287 lines of spawn-and- hope shellouts plus a ~120-line nginx auth-injector helper that worked around per-node RPC password baking. The helper is gone too — the pre-start hook renders the per-node nginx.conf to /var/lib/archipelago/ bitcoin-ui/nginx.conf and the Quadlet unit bind-mounts it read-only. runtime.rs handle_package_uninstall now disables companions before the container rm loop. Otherwise systemd's Restart=always would respawn each companion within ~10s of removal. Tests: 53 container tests pass, including 6 quadlet renderer tests (host network, bridge network, capability set, atomic write idempotence) and 2 companion specs (per-app companion lookup, build_unit shape). boot_reconciler tests gain a #[cfg(test)] without_companion_stage() flag so the paused-clock fixtures don't race the real systemctl I/O. A bats regression test (companion-survives-archipelago-restart.bats, gated on ARCHY_ALLOW_DESTRUCTIVE=1) asserts the .228 failure mode cannot recur: every installed companion has a unit file, services stay active across systemctl --user restart archipelago, and a deleted unit file is recreated within one reconcile tick. Net delta: +941 / -363, but the +941 is mostly tests (~440 lines) and the new declarative layer; the imperative tokio::spawn block and its nginx-auth helper are gone, removing two failure classes (orphan companions on archipelago crash, and post-start exec races under tightly-confined cgroups) that previously needed manual SSH recovery. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 10:45:07 -04:00
info!(companion = spec.name, "building locally from {dir}");
let out = Command::new("podman")
2026-05-05 11:29:18 -04:00
.args(["build", "-t", &local_image, dir])
refactor(container): move companion UIs to systemd via Quadlet Companion UI containers (archy-bitcoin-ui, archy-lnd-ui, archy-electrs-ui) used to be launched as fire-and-forget tokio::spawn blocks from install.rs. If archipelago crashed mid-spawn or the container's cgroup was reaped, companions vanished from podman ps -a and only a manual rm/run could bring them back (the .228 incident). Now each companion is rendered as a Quadlet .container unit under ~/.config/containers/systemd/, daemon-reloaded, and started via systemctl --user. systemd owns supervision from that point on: - archipelago can crash, restart, or be uninstalled without touching any companion. - Quadlet's Restart=always + RestartSec=10 handles container exits. - A 30s reconcile tick in boot_reconciler enumerates expected companion units and re-installs any whose unit file or service vanished — defense-in-depth against external tampering. New module layout: - container/quadlet.rs: pure unit renderer + atomic write_if_changed + systemctl helpers (daemon_reload_user / enable_now / disable_remove / is_active). 6 unit tests, no I/O in the renderer. - container/companion.rs: per-app companion specs, install/remove/ reconcile, image presence (build local first, fall back to insecure registry only via image_uses_insecure_registry whitelist). 2 tests. install.rs handle_package_install now ends with a single call to companion::install_for(package_id), replacing 287 lines of spawn-and- hope shellouts plus a ~120-line nginx auth-injector helper that worked around per-node RPC password baking. The helper is gone too — the pre-start hook renders the per-node nginx.conf to /var/lib/archipelago/ bitcoin-ui/nginx.conf and the Quadlet unit bind-mounts it read-only. runtime.rs handle_package_uninstall now disables companions before the container rm loop. Otherwise systemd's Restart=always would respawn each companion within ~10s of removal. Tests: 53 container tests pass, including 6 quadlet renderer tests (host network, bridge network, capability set, atomic write idempotence) and 2 companion specs (per-app companion lookup, build_unit shape). boot_reconciler tests gain a #[cfg(test)] without_companion_stage() flag so the paused-clock fixtures don't race the real systemctl I/O. A bats regression test (companion-survives-archipelago-restart.bats, gated on ARCHY_ALLOW_DESTRUCTIVE=1) asserts the .228 failure mode cannot recur: every installed companion has a unit file, services stay active across systemctl --user restart archipelago, and a deleted unit file is recreated within one reconcile tick. Net delta: +941 / -363, but the +941 is mostly tests (~440 lines) and the new declarative layer; the imperative tokio::spawn block and its nginx-auth helper are gone, removing two failure classes (orphan companions on archipelago crash, and post-start exec races under tightly-confined cgroups) that previously needed manual SSH recovery. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 10:45:07 -04:00
.output()
.await
.context("spawn podman build")?;
if out.status.success() {
return Ok(local_image);
}
warn!(
companion = spec.name,
"local build failed: {}",
String::from_utf8_lossy(&out.stderr).trim()
);
// Fall through to registry pull rather than fail outright.
break;
}
}
// Registry pull. Use insecure flag only for whitelisted hosts.
let mut cmd = Command::new("podman");
cmd.arg("pull");
if image_uses_insecure_registry(&registry_image) {
cmd.arg("--tls-verify=false");
}
cmd.arg(&registry_image);
let out = cmd.output().await.context("spawn podman pull")?;
if !out.status.success() {
anyhow::bail!(
"no local Dockerfile and registry pull failed for {}: {}",
spec.name,
String::from_utf8_lossy(&out.stderr).trim()
);
}
Ok(registry_image)
}
2026-05-05 11:29:18 -04:00
async fn image_exists(image: &str) -> bool {
Command::new("podman")
.args(["image", "exists", image])
.status()
.await
.is_ok_and(|status| status.success())
}
refactor(container): move companion UIs to systemd via Quadlet Companion UI containers (archy-bitcoin-ui, archy-lnd-ui, archy-electrs-ui) used to be launched as fire-and-forget tokio::spawn blocks from install.rs. If archipelago crashed mid-spawn or the container's cgroup was reaped, companions vanished from podman ps -a and only a manual rm/run could bring them back (the .228 incident). Now each companion is rendered as a Quadlet .container unit under ~/.config/containers/systemd/, daemon-reloaded, and started via systemctl --user. systemd owns supervision from that point on: - archipelago can crash, restart, or be uninstalled without touching any companion. - Quadlet's Restart=always + RestartSec=10 handles container exits. - A 30s reconcile tick in boot_reconciler enumerates expected companion units and re-installs any whose unit file or service vanished — defense-in-depth against external tampering. New module layout: - container/quadlet.rs: pure unit renderer + atomic write_if_changed + systemctl helpers (daemon_reload_user / enable_now / disable_remove / is_active). 6 unit tests, no I/O in the renderer. - container/companion.rs: per-app companion specs, install/remove/ reconcile, image presence (build local first, fall back to insecure registry only via image_uses_insecure_registry whitelist). 2 tests. install.rs handle_package_install now ends with a single call to companion::install_for(package_id), replacing 287 lines of spawn-and- hope shellouts plus a ~120-line nginx auth-injector helper that worked around per-node RPC password baking. The helper is gone too — the pre-start hook renders the per-node nginx.conf to /var/lib/archipelago/ bitcoin-ui/nginx.conf and the Quadlet unit bind-mounts it read-only. runtime.rs handle_package_uninstall now disables companions before the container rm loop. Otherwise systemd's Restart=always would respawn each companion within ~10s of removal. Tests: 53 container tests pass, including 6 quadlet renderer tests (host network, bridge network, capability set, atomic write idempotence) and 2 companion specs (per-app companion lookup, build_unit shape). boot_reconciler tests gain a #[cfg(test)] without_companion_stage() flag so the paused-clock fixtures don't race the real systemctl I/O. A bats regression test (companion-survives-archipelago-restart.bats, gated on ARCHY_ALLOW_DESTRUCTIVE=1) asserts the .228 failure mode cannot recur: every installed companion has a unit file, services stay active across systemctl --user restart archipelago, and a deleted unit file is recreated within one reconcile tick. Net delta: +941 / -363, but the +941 is mostly tests (~440 lines) and the new declarative layer; the imperative tokio::spawn block and its nginx-auth helper are gone, removing two failure classes (orphan companions on archipelago crash, and post-start exec races under tightly-confined cgroups) that previously needed manual SSH recovery. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 10:45:07 -04:00
fn build_unit(spec: &CompanionSpec, image: &str) -> QuadletUnit {
QuadletUnit {
name: spec.name.into(),
description: format!("Archipelago companion UI: {}", spec.name),
image: image.into(),
2026-05-05 11:29:18 -04:00
network: if spec.host_network {
NetworkMode::Host
} else {
NetworkMode::Bridge("bridge".into())
},
refactor(container): move companion UIs to systemd via Quadlet Companion UI containers (archy-bitcoin-ui, archy-lnd-ui, archy-electrs-ui) used to be launched as fire-and-forget tokio::spawn blocks from install.rs. If archipelago crashed mid-spawn or the container's cgroup was reaped, companions vanished from podman ps -a and only a manual rm/run could bring them back (the .228 incident). Now each companion is rendered as a Quadlet .container unit under ~/.config/containers/systemd/, daemon-reloaded, and started via systemctl --user. systemd owns supervision from that point on: - archipelago can crash, restart, or be uninstalled without touching any companion. - Quadlet's Restart=always + RestartSec=10 handles container exits. - A 30s reconcile tick in boot_reconciler enumerates expected companion units and re-installs any whose unit file or service vanished — defense-in-depth against external tampering. New module layout: - container/quadlet.rs: pure unit renderer + atomic write_if_changed + systemctl helpers (daemon_reload_user / enable_now / disable_remove / is_active). 6 unit tests, no I/O in the renderer. - container/companion.rs: per-app companion specs, install/remove/ reconcile, image presence (build local first, fall back to insecure registry only via image_uses_insecure_registry whitelist). 2 tests. install.rs handle_package_install now ends with a single call to companion::install_for(package_id), replacing 287 lines of spawn-and- hope shellouts plus a ~120-line nginx auth-injector helper that worked around per-node RPC password baking. The helper is gone too — the pre-start hook renders the per-node nginx.conf to /var/lib/archipelago/ bitcoin-ui/nginx.conf and the Quadlet unit bind-mounts it read-only. runtime.rs handle_package_uninstall now disables companions before the container rm loop. Otherwise systemd's Restart=always would respawn each companion within ~10s of removal. Tests: 53 container tests pass, including 6 quadlet renderer tests (host network, bridge network, capability set, atomic write idempotence) and 2 companion specs (per-app companion lookup, build_unit shape). boot_reconciler tests gain a #[cfg(test)] without_companion_stage() flag so the paused-clock fixtures don't race the real systemctl I/O. A bats regression test (companion-survives-archipelago-restart.bats, gated on ARCHY_ALLOW_DESTRUCTIVE=1) asserts the .228 failure mode cannot recur: every installed companion has a unit file, services stay active across systemctl --user restart archipelago, and a deleted unit file is recreated within one reconcile tick. Net delta: +941 / -363, but the +941 is mostly tests (~440 lines) and the new declarative layer; the imperative tokio::spawn block and its nginx-auth helper are gone, removing two failure classes (orphan companions on archipelago crash, and post-start exec races under tightly-confined cgroups) that previously needed manual SSH recovery. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 10:45:07 -04:00
// Run as root inside the container so nginx can chown its
// worker dirs. Rootless podman maps this to a high host UID,
// so it is unprivileged on the host.
user: Some("0:0".into()),
memory_mb: Some(128),
cap_drop_all: true,
cap_add: vec![
"CHOWN".into(),
"DAC_OVERRIDE".into(),
"NET_BIND_SERVICE".into(),
"SETUID".into(),
"SETGID".into(),
],
bind_mounts: spec
.bind_mounts
.iter()
.map(|(host, container)| BindMount {
host: PathBuf::from(*host),
container: PathBuf::from(*container),
read_only: true,
})
.collect(),
2026-05-05 11:29:18 -04:00
ports: spec
.ports
.iter()
.map(|(host, container)| (*host, *container, "tcp".into()))
.collect(),
refactor(container): move companion UIs to systemd via Quadlet Companion UI containers (archy-bitcoin-ui, archy-lnd-ui, archy-electrs-ui) used to be launched as fire-and-forget tokio::spawn blocks from install.rs. If archipelago crashed mid-spawn or the container's cgroup was reaped, companions vanished from podman ps -a and only a manual rm/run could bring them back (the .228 incident). Now each companion is rendered as a Quadlet .container unit under ~/.config/containers/systemd/, daemon-reloaded, and started via systemctl --user. systemd owns supervision from that point on: - archipelago can crash, restart, or be uninstalled without touching any companion. - Quadlet's Restart=always + RestartSec=10 handles container exits. - A 30s reconcile tick in boot_reconciler enumerates expected companion units and re-installs any whose unit file or service vanished — defense-in-depth against external tampering. New module layout: - container/quadlet.rs: pure unit renderer + atomic write_if_changed + systemctl helpers (daemon_reload_user / enable_now / disable_remove / is_active). 6 unit tests, no I/O in the renderer. - container/companion.rs: per-app companion specs, install/remove/ reconcile, image presence (build local first, fall back to insecure registry only via image_uses_insecure_registry whitelist). 2 tests. install.rs handle_package_install now ends with a single call to companion::install_for(package_id), replacing 287 lines of spawn-and- hope shellouts plus a ~120-line nginx auth-injector helper that worked around per-node RPC password baking. The helper is gone too — the pre-start hook renders the per-node nginx.conf to /var/lib/archipelago/ bitcoin-ui/nginx.conf and the Quadlet unit bind-mounts it read-only. runtime.rs handle_package_uninstall now disables companions before the container rm loop. Otherwise systemd's Restart=always would respawn each companion within ~10s of removal. Tests: 53 container tests pass, including 6 quadlet renderer tests (host network, bridge network, capability set, atomic write idempotence) and 2 companion specs (per-app companion lookup, build_unit shape). boot_reconciler tests gain a #[cfg(test)] without_companion_stage() flag so the paused-clock fixtures don't race the real systemctl I/O. A bats regression test (companion-survives-archipelago-restart.bats, gated on ARCHY_ALLOW_DESTRUCTIVE=1) asserts the .228 failure mode cannot recur: every installed companion has a unit file, services stay active across systemctl --user restart archipelago, and a deleted unit file is recreated within one reconcile tick. Net delta: +941 / -363, but the +941 is mostly tests (~440 lines) and the new declarative layer; the imperative tokio::spawn block and its nginx-auth helper are gone, removing two failure classes (orphan companions on archipelago crash, and post-start exec races under tightly-confined cgroups) that previously needed manual SSH recovery. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 10:45:07 -04:00
extra_podman_args: vec![],
depends_on: vec![],
feat(quadlet): backend-manifest renderer (Phase 3.1 of v1.7.52) The QuadletUnit struct now covers everything a backend manifest needs (ports, environment, devices, add_hosts, entrypoint+command, read-only root, no_new_privileges, cpu_quota, restart policy choice). Adds QuadletUnit::from_manifest(&AppManifest, name) that translates a parsed manifest into a unit, plus parse_memory_mib for "1g"/"512m"/raw-MiB forms. The renderer skips empty/false directives so existing companion units render byte-identically — no behavior change for shipping companions; the backend renderer is dead code until Phase 3.2 wires it into the orchestrator. Eight new unit tests cover: * parse_memory_mib forms (1024, 512m, 2g, garbage) * shell_join quoting (whitespace, embedded quotes) * RestartPolicy → systemd string mapping * render emits backend directives when set * render skips them when defaulted (companion regression gate) * from_manifest happy path on a bitcoin-knots-shaped manifest * from_manifest read-only volume detection * from_manifest tmpfs filtering * end-to-end manifest → render bytes assertion Tests: 615 → 624 (+9 net; one pre-existing parse_memory_mib path was implicitly covered before but is now explicit). Cargo warnings: 0. `from_manifest`, `parse_memory_mib`, and `RestartPolicy::OnFailure` are marked allow(dead_code) with explicit references to Phase 3.2 — if 3.2 doesn't wire them, the dead-code warning resurfaces. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 17:09:50 -04:00
// Companions don't use the backend-manifest extension fields;
// the renderer skips empty/false directives so the rendered
// bytes are unchanged from before quadlet.rs grew the new fields.
..QuadletUnit::default()
refactor(container): move companion UIs to systemd via Quadlet Companion UI containers (archy-bitcoin-ui, archy-lnd-ui, archy-electrs-ui) used to be launched as fire-and-forget tokio::spawn blocks from install.rs. If archipelago crashed mid-spawn or the container's cgroup was reaped, companions vanished from podman ps -a and only a manual rm/run could bring them back (the .228 incident). Now each companion is rendered as a Quadlet .container unit under ~/.config/containers/systemd/, daemon-reloaded, and started via systemctl --user. systemd owns supervision from that point on: - archipelago can crash, restart, or be uninstalled without touching any companion. - Quadlet's Restart=always + RestartSec=10 handles container exits. - A 30s reconcile tick in boot_reconciler enumerates expected companion units and re-installs any whose unit file or service vanished — defense-in-depth against external tampering. New module layout: - container/quadlet.rs: pure unit renderer + atomic write_if_changed + systemctl helpers (daemon_reload_user / enable_now / disable_remove / is_active). 6 unit tests, no I/O in the renderer. - container/companion.rs: per-app companion specs, install/remove/ reconcile, image presence (build local first, fall back to insecure registry only via image_uses_insecure_registry whitelist). 2 tests. install.rs handle_package_install now ends with a single call to companion::install_for(package_id), replacing 287 lines of spawn-and- hope shellouts plus a ~120-line nginx auth-injector helper that worked around per-node RPC password baking. The helper is gone too — the pre-start hook renders the per-node nginx.conf to /var/lib/archipelago/ bitcoin-ui/nginx.conf and the Quadlet unit bind-mounts it read-only. runtime.rs handle_package_uninstall now disables companions before the container rm loop. Otherwise systemd's Restart=always would respawn each companion within ~10s of removal. Tests: 53 container tests pass, including 6 quadlet renderer tests (host network, bridge network, capability set, atomic write idempotence) and 2 companion specs (per-app companion lookup, build_unit shape). boot_reconciler tests gain a #[cfg(test)] without_companion_stage() flag so the paused-clock fixtures don't race the real systemctl I/O. A bats regression test (companion-survives-archipelago-restart.bats, gated on ARCHY_ALLOW_DESTRUCTIVE=1) asserts the .228 failure mode cannot recur: every installed companion has a unit file, services stay active across systemctl --user restart archipelago, and a deleted unit file is recreated within one reconcile tick. Net delta: +941 / -363, but the +941 is mostly tests (~440 lines) and the new declarative layer; the imperative tokio::spawn block and its nginx-auth helper are gone, removing two failure classes (orphan companions on archipelago crash, and post-start exec races under tightly-confined cgroups) that previously needed manual SSH recovery. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 10:45:07 -04:00
}
}
/// Is a user systemd manager reachable? In production archipelago.service
/// inherits XDG_RUNTIME_DIR from systemd; in unit tests / CI sandboxes it
/// is unset, in which case `systemctl --user` would fail and write to
/// HOME would be an unwanted side effect. The reconciler skips its
/// companion stage when this is false.
fn user_systemd_available() -> bool {
std::env::var_os("XDG_RUNTIME_DIR")
.map(|v| !v.is_empty())
.unwrap_or(false)
}
/// Reconcile companion presence: every expected companion for the
/// given installed apps must have its quadlet unit on disk and its
/// service active. Returns a list of (companion, error) for anything
/// that needed correction and failed.
///
/// Called from `boot_reconciler` so a deleted unit file or a stopped
/// service is repaired within one tick. No-ops if the user systemd
/// manager is not reachable (CI / test environments).
pub async fn reconcile(installed_apps: &[String]) -> Vec<(String, anyhow::Error)> {
if !user_systemd_available() {
return Vec::new();
}
let mut failures = Vec::new();
for app_id in installed_apps {
for spec in companions_for(app_id) {
match needs_repair(spec).await {
Ok(false) => {}
Ok(true) => {
refactor(install): route orchestrator-managed apps through orchestrator first Phase 3a of the install path consolidation. Two coupled changes: 1. install.rs handle_package_install: gate the legacy "container exists → adopt + return" probe on !orchestrator_managed. Apps the orchestrator knows about (bitcoin-knots, bitcoin-core, lnd, electrumx, fedimint, filebrowser, btcpay-server stack apps, mempool stack apps, plus the companion UIs that just moved to Quadlet) skip the legacy probe and fall straight into the orchestrator branch. The legacy adopt block was returning success on a bare `podman start` exit-0 — even when the process inside the container crashed seconds later. That's the .228 "running but unreachable" failure mode. The orchestrator's ensure_running honors the manifest's health check and pre-start hooks (e.g. re-renders bitcoin-ui's nginx.conf if the RPC password rotated), so this is a behavioral upgrade, not just a refactor. 2. ProdContainerOrchestrator::install: make idempotent. Previously it blindly called install_fresh which would fail on `podman create` if the container name already existed. Now it delegates to ensure_running: - Container Running + healthy → no-op (refresh hooks, restart if config rewritten) - Container Stopped/Exited → start (with hook refresh) - Container missing → install_fresh - Container in wedged state (Created/Paused/Unknown) → force-recreate Without this, change #1 would regress every "container already exists" case for the 18 orchestrator-managed app IDs. With it, install becomes the single source of truth for "make app X be in the desired state." Tests: 654 passed across the workspace (614 unit + 37 orchestration + 3 rpc), 0 failures. The 20 prod_orchestrator tests cover the install / ensure_running / reconcile paths the new install delegates through. Net delta: install.rs grows by ~30 lines (gating wrapper + comments), prod_orchestrator.rs grows by ~30 lines (idempotent install body). Both are temporary — the larger deletions (~1700 lines) come once every app has been verified through the orchestrator path in subsequent phases. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 11:12:52 -04:00
info!(
companion = spec.name,
"reconcile: companion not active, repairing"
);
refactor(container): move companion UIs to systemd via Quadlet Companion UI containers (archy-bitcoin-ui, archy-lnd-ui, archy-electrs-ui) used to be launched as fire-and-forget tokio::spawn blocks from install.rs. If archipelago crashed mid-spawn or the container's cgroup was reaped, companions vanished from podman ps -a and only a manual rm/run could bring them back (the .228 incident). Now each companion is rendered as a Quadlet .container unit under ~/.config/containers/systemd/, daemon-reloaded, and started via systemctl --user. systemd owns supervision from that point on: - archipelago can crash, restart, or be uninstalled without touching any companion. - Quadlet's Restart=always + RestartSec=10 handles container exits. - A 30s reconcile tick in boot_reconciler enumerates expected companion units and re-installs any whose unit file or service vanished — defense-in-depth against external tampering. New module layout: - container/quadlet.rs: pure unit renderer + atomic write_if_changed + systemctl helpers (daemon_reload_user / enable_now / disable_remove / is_active). 6 unit tests, no I/O in the renderer. - container/companion.rs: per-app companion specs, install/remove/ reconcile, image presence (build local first, fall back to insecure registry only via image_uses_insecure_registry whitelist). 2 tests. install.rs handle_package_install now ends with a single call to companion::install_for(package_id), replacing 287 lines of spawn-and- hope shellouts plus a ~120-line nginx auth-injector helper that worked around per-node RPC password baking. The helper is gone too — the pre-start hook renders the per-node nginx.conf to /var/lib/archipelago/ bitcoin-ui/nginx.conf and the Quadlet unit bind-mounts it read-only. runtime.rs handle_package_uninstall now disables companions before the container rm loop. Otherwise systemd's Restart=always would respawn each companion within ~10s of removal. Tests: 53 container tests pass, including 6 quadlet renderer tests (host network, bridge network, capability set, atomic write idempotence) and 2 companion specs (per-app companion lookup, build_unit shape). boot_reconciler tests gain a #[cfg(test)] without_companion_stage() flag so the paused-clock fixtures don't race the real systemctl I/O. A bats regression test (companion-survives-archipelago-restart.bats, gated on ARCHY_ALLOW_DESTRUCTIVE=1) asserts the .228 failure mode cannot recur: every installed companion has a unit file, services stay active across systemctl --user restart archipelago, and a deleted unit file is recreated within one reconcile tick. Net delta: +941 / -363, but the +941 is mostly tests (~440 lines) and the new declarative layer; the imperative tokio::spawn block and its nginx-auth helper are gone, removing two failure classes (orphan companions on archipelago crash, and post-start exec races under tightly-confined cgroups) that previously needed manual SSH recovery. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 10:45:07 -04:00
if let Err(e) = install_one(spec).await {
failures.push((spec.name.to_string(), e));
}
}
Err(e) => {
warn!(companion = spec.name, error = %e, "reconcile probe failed");
failures.push((spec.name.to_string(), e));
}
}
}
}
failures
}
/// Does this companion need install_one to be re-run? Returns true if
/// the unit file is missing OR the service is not active.
async fn needs_repair(spec: &CompanionSpec) -> Result<bool> {
let dir = quadlet::unit_dir().await?;
let unit_path = dir.join(format!("{}.container", spec.name));
if !fs::try_exists(&unit_path).await.unwrap_or(false) {
return Ok(true);
}
let svc = format!("{}.service", spec.name);
Ok(!quadlet::is_active(&svc).await)
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn companions_for_known_apps_returns_expected_set() {
assert_eq!(companions_for("bitcoin-knots").len(), 1);
assert_eq!(companions_for("bitcoin-core").len(), 1);
assert_eq!(companions_for("bitcoin").len(), 1);
assert_eq!(companions_for("lnd").len(), 1);
assert_eq!(companions_for("electrumx").len(), 1);
assert_eq!(companions_for("electrs").len(), 1);
assert_eq!(companions_for("mempool-electrs").len(), 1);
assert_eq!(companions_for("nextcloud").len(), 0);
assert_eq!(companions_for("not-a-real-app").len(), 0);
}
#[test]
fn build_unit_uses_host_network_and_drops_caps() {
let spec = &BITCOIN_UI[0];
let u = build_unit(spec, "localhost/bitcoin-ui:latest");
assert_eq!(u.name, "archy-bitcoin-ui");
assert!(matches!(u.network, NetworkMode::Host));
assert!(u.cap_drop_all);
assert!(u.cap_add.iter().any(|c| c == "NET_BIND_SERVICE"));
assert_eq!(u.user.as_deref(), Some("0:0"));
assert_eq!(u.memory_mb, Some(128));
assert_eq!(u.bind_mounts.len(), 1);
assert_eq!(
u.bind_mounts[0].container,
PathBuf::from("/etc/nginx/conf.d/default.conf")
);
assert!(u.bind_mounts[0].read_only);
}
2026-05-05 11:29:18 -04:00
#[test]
fn lnd_ui_uses_port_mapping_not_host_port_80() {
let spec = &LND_UI[0];
let u = build_unit(spec, "localhost/lnd-ui:latest");
assert_eq!(u.name, "archy-lnd-ui");
assert!(matches!(u.network, NetworkMode::Bridge(ref n) if n == "bridge"));
assert_eq!(u.ports, vec![(18083, 80, "tcp".into())]);
}
refactor(container): move companion UIs to systemd via Quadlet Companion UI containers (archy-bitcoin-ui, archy-lnd-ui, archy-electrs-ui) used to be launched as fire-and-forget tokio::spawn blocks from install.rs. If archipelago crashed mid-spawn or the container's cgroup was reaped, companions vanished from podman ps -a and only a manual rm/run could bring them back (the .228 incident). Now each companion is rendered as a Quadlet .container unit under ~/.config/containers/systemd/, daemon-reloaded, and started via systemctl --user. systemd owns supervision from that point on: - archipelago can crash, restart, or be uninstalled without touching any companion. - Quadlet's Restart=always + RestartSec=10 handles container exits. - A 30s reconcile tick in boot_reconciler enumerates expected companion units and re-installs any whose unit file or service vanished — defense-in-depth against external tampering. New module layout: - container/quadlet.rs: pure unit renderer + atomic write_if_changed + systemctl helpers (daemon_reload_user / enable_now / disable_remove / is_active). 6 unit tests, no I/O in the renderer. - container/companion.rs: per-app companion specs, install/remove/ reconcile, image presence (build local first, fall back to insecure registry only via image_uses_insecure_registry whitelist). 2 tests. install.rs handle_package_install now ends with a single call to companion::install_for(package_id), replacing 287 lines of spawn-and- hope shellouts plus a ~120-line nginx auth-injector helper that worked around per-node RPC password baking. The helper is gone too — the pre-start hook renders the per-node nginx.conf to /var/lib/archipelago/ bitcoin-ui/nginx.conf and the Quadlet unit bind-mounts it read-only. runtime.rs handle_package_uninstall now disables companions before the container rm loop. Otherwise systemd's Restart=always would respawn each companion within ~10s of removal. Tests: 53 container tests pass, including 6 quadlet renderer tests (host network, bridge network, capability set, atomic write idempotence) and 2 companion specs (per-app companion lookup, build_unit shape). boot_reconciler tests gain a #[cfg(test)] without_companion_stage() flag so the paused-clock fixtures don't race the real systemctl I/O. A bats regression test (companion-survives-archipelago-restart.bats, gated on ARCHY_ALLOW_DESTRUCTIVE=1) asserts the .228 failure mode cannot recur: every installed companion has a unit file, services stay active across systemctl --user restart archipelago, and a deleted unit file is recreated within one reconcile tick. Net delta: +941 / -363, but the +941 is mostly tests (~440 lines) and the new declarative layer; the imperative tokio::spawn block and its nginx-auth helper are gone, removing two failure classes (orphan companions on archipelago crash, and post-start exec races under tightly-confined cgroups) that previously needed manual SSH recovery. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-01 10:45:07 -04:00
}