B: B2/B3 health probes, loops and the in-memory status model
This commit is contained in:
1 parent
bd5b21bcc4
commit
d94c2b3a1d
3 files changed
+994
-126
No files matched your search
+397
-40
@@ -2,11 +2,16 @@
|
||||
//! side never awaits on a peer and is cheap enough for a metrics scrape.
|
||||
|
||||
use std::collections::BTreeMap;
|
||||
use std::sync::{Arc, RwLock};
|
||||
use std::sync::{Arc, Mutex, RwLock};
|
||||
use std::time::Duration;
|
||||
|
||||
use tokio::task::JoinHandle;
|
||||
use tokio_util::sync::CancellationToken;
|
||||
|
||||
use crate::config::{CliOverrides, Config};
|
||||
use crate::error::AppError;
|
||||
use crate::peers::error::PeerConfigError;
|
||||
use crate::peers::health::{self, LoopConfig, StatusSink};
|
||||
use crate::peers::transport::{PeerResponse, PeerTransport};
|
||||
use crate::peers::types::{ApiStatus, PeerConfig, PeerName, PeerStatus, PeersConfig, ServerStatus};
|
||||
|
||||
@@ -29,9 +34,39 @@ struct RegistryState {
|
||||
statuses: BTreeMap<PeerName, PeerStatus>,
|
||||
}
|
||||
|
||||
/// Status storage shared with the loops; this is the only mutable state a
|
||||
/// loop needs, so the registry can be dropped without leaking a task.
|
||||
struct Shared {
|
||||
state: RwLock<RegistryState>,
|
||||
}
|
||||
|
||||
impl StatusSink for Shared {
|
||||
fn apply_server(&self, name: &PeerName, status: ServerStatus) {
|
||||
if let Some(entry) = lock_write(&self.state).statuses.get_mut(name) {
|
||||
entry.server = status;
|
||||
}
|
||||
}
|
||||
|
||||
fn apply_api(&self, name: &PeerName, status: ApiStatus) {
|
||||
if let Some(entry) = lock_write(&self.state).statuses.get_mut(name) {
|
||||
entry.api = status;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
struct LoopEntry {
|
||||
peer: PeerConfig,
|
||||
params: LoopConfig,
|
||||
token: CancellationToken,
|
||||
join: JoinHandle<()>,
|
||||
}
|
||||
|
||||
pub struct PeerRegistry {
|
||||
transport: Arc<dyn PeerTransport>,
|
||||
state: RwLock<RegistryState>,
|
||||
shared: Arc<Shared>,
|
||||
loops: Mutex<BTreeMap<PeerName, LoopEntry>>,
|
||||
loops_enabled: bool,
|
||||
shutdown: CancellationToken,
|
||||
}
|
||||
|
||||
fn initial_status(peer: &PeerConfig) -> PeerStatus {
|
||||
@@ -52,8 +87,22 @@ fn lock_write<T>(lock: &RwLock<T>) -> std::sync::RwLockWriteGuard<'_, T> {
|
||||
lock.write().unwrap_or_else(|e| e.into_inner())
|
||||
}
|
||||
|
||||
fn lock_mutex<T>(lock: &Mutex<T>) -> std::sync::MutexGuard<'_, T> {
|
||||
lock.lock().unwrap_or_else(|e| e.into_inner())
|
||||
}
|
||||
|
||||
impl PeerRegistry {
|
||||
/// Registry without loops, for tests that drive the transport directly.
|
||||
pub(crate) fn new(config: PeersConfig, transport: Arc<dyn PeerTransport>) -> Self {
|
||||
Self::build(config, transport, CancellationToken::new(), false)
|
||||
}
|
||||
|
||||
fn build(
|
||||
config: PeersConfig,
|
||||
transport: Arc<dyn PeerTransport>,
|
||||
shutdown: CancellationToken,
|
||||
loops_enabled: bool,
|
||||
) -> Self {
|
||||
let statuses = config
|
||||
.peer
|
||||
.iter()
|
||||
@@ -61,10 +110,15 @@ impl PeerRegistry {
|
||||
.collect();
|
||||
Self {
|
||||
transport,
|
||||
state: RwLock::new(RegistryState {
|
||||
config: Arc::new(config),
|
||||
statuses,
|
||||
shared: Arc::new(Shared {
|
||||
state: RwLock::new(RegistryState {
|
||||
config: Arc::new(config),
|
||||
statuses,
|
||||
}),
|
||||
}),
|
||||
loops: Mutex::new(BTreeMap::new()),
|
||||
loops_enabled,
|
||||
shutdown,
|
||||
}
|
||||
}
|
||||
|
||||
@@ -73,40 +127,114 @@ impl PeerRegistry {
|
||||
config: PeersConfig,
|
||||
transport: Arc<dyn PeerTransport>,
|
||||
) -> Self {
|
||||
Self {
|
||||
transport,
|
||||
state: RwLock::new(RegistryState {
|
||||
config: Arc::new(config),
|
||||
statuses: statuses.into_iter().map(|s| (s.name.clone(), s)).collect(),
|
||||
}),
|
||||
let reg = Self::build(config, transport, CancellationToken::new(), false);
|
||||
lock_write(®.shared.state).statuses =
|
||||
statuses.into_iter().map(|s| (s.name.clone(), s)).collect();
|
||||
reg
|
||||
}
|
||||
|
||||
/// Real registry for the server: builds the outbound transport and starts
|
||||
/// one loop per enabled peer. `shutdown` is the process token; cancelling
|
||||
/// it stops the loops.
|
||||
pub fn start(
|
||||
config: PeersConfig,
|
||||
shutdown: CancellationToken,
|
||||
) -> Result<Arc<Self>, PeerConfigError> {
|
||||
let transport = Arc::new(ReqwestTransport::new(&config)?);
|
||||
Ok(Self::start_with_transport(config, transport, shutdown))
|
||||
}
|
||||
|
||||
/// Loop-enabled registry over a supplied transport, for tests.
|
||||
pub fn start_with_transport(
|
||||
config: PeersConfig,
|
||||
transport: Arc<dyn PeerTransport>,
|
||||
shutdown: CancellationToken,
|
||||
) -> Arc<Self> {
|
||||
let registry = Arc::new(Self::build(config, transport, shutdown.child_token(), true));
|
||||
registry.spawn_all();
|
||||
registry
|
||||
}
|
||||
|
||||
fn spawn_all(&self) {
|
||||
let peers: Vec<PeerConfig> = if self.config().enabled {
|
||||
self.config()
|
||||
.peer
|
||||
.iter()
|
||||
.filter(|peer| peer.enabled)
|
||||
.cloned()
|
||||
.collect()
|
||||
} else {
|
||||
Vec::new()
|
||||
};
|
||||
for peer in peers {
|
||||
self.spawn_peer(&peer);
|
||||
}
|
||||
}
|
||||
|
||||
fn spawn_peer(&self, peer: &PeerConfig) {
|
||||
let params = LoopConfig::from_config(&self.config());
|
||||
let initial = self.status_internal(&peer.name);
|
||||
let token = self.shutdown.child_token();
|
||||
let sink: Arc<dyn StatusSink> = self.shared.clone();
|
||||
let transport = self.transport.clone();
|
||||
let peer_copy = peer.clone();
|
||||
let name = peer.name.clone();
|
||||
let task_token = token.clone();
|
||||
let join = tokio::spawn(async move {
|
||||
health::run_peer(transport, peer_copy, params, initial, sink, task_token).await;
|
||||
});
|
||||
lock_mutex(&self.loops).insert(
|
||||
name,
|
||||
LoopEntry {
|
||||
peer: peer.clone(),
|
||||
params,
|
||||
token,
|
||||
join,
|
||||
},
|
||||
);
|
||||
}
|
||||
|
||||
fn stop_loop(&self, entry: LoopEntry) {
|
||||
// Cancel, then let the task finish its bounded probe on its own.
|
||||
entry.token.cancel();
|
||||
tokio::spawn(async move {
|
||||
let _ = entry.join.await;
|
||||
});
|
||||
}
|
||||
|
||||
/// Sorted by name; disabled peers included.
|
||||
pub fn snapshot(&self) -> Vec<PeerStatus> {
|
||||
lock_read(&self.state).statuses.values().cloned().collect()
|
||||
lock_read(&self.shared.state)
|
||||
.statuses
|
||||
.values()
|
||||
.cloned()
|
||||
.collect()
|
||||
}
|
||||
|
||||
pub fn status(&self, name: &str) -> Result<PeerStatus, AppError> {
|
||||
let Ok(key) = PeerName::parse(name) else {
|
||||
return Err(AppError::NotFound(format!("peer {name}: unknown peer")));
|
||||
};
|
||||
lock_read(&self.state)
|
||||
lock_read(&self.shared.state)
|
||||
.statuses
|
||||
.get(&key)
|
||||
.cloned()
|
||||
.ok_or_else(|| AppError::NotFound(format!("peer {name}: unknown peer")))
|
||||
}
|
||||
|
||||
fn status_internal(&self, name: &PeerName) -> Option<PeerStatus> {
|
||||
lock_read(&self.shared.state).statuses.get(name).cloned()
|
||||
}
|
||||
|
||||
pub fn contains(&self, name: &str) -> bool {
|
||||
let Ok(key) = PeerName::parse(name) else {
|
||||
return false;
|
||||
};
|
||||
lock_read(&self.state).statuses.contains_key(&key)
|
||||
lock_read(&self.shared.state).statuses.contains_key(&key)
|
||||
}
|
||||
|
||||
pub fn config(&self) -> Arc<PeersConfig> {
|
||||
lock_read(&self.state).config.clone()
|
||||
lock_read(&self.shared.state).config.clone()
|
||||
}
|
||||
|
||||
/// Single proxy entry point for REST and MCP.
|
||||
@@ -119,7 +247,7 @@ impl PeerRegistry {
|
||||
) -> Result<PeerResponse, AppError> {
|
||||
let key = PeerName::parse(name).ok();
|
||||
let (peer, timeout) = {
|
||||
let st = lock_read(&self.state);
|
||||
let st = lock_read(&self.shared.state);
|
||||
let peer = key.and_then(|key| st.config.peer.iter().find(|p| p.name == key).cloned());
|
||||
(peer, Duration::from_secs(st.config.request_timeout_secs))
|
||||
};
|
||||
@@ -128,30 +256,114 @@ impl PeerRegistry {
|
||||
}
|
||||
|
||||
/// Replace the peer set. Statuses of peers that survive a reload are kept
|
||||
/// so `since` survives a config change. Validation arrives with the
|
||||
/// config and secrets tasks; this body only swaps state.
|
||||
/// so `since` survives a config change. Loops for new or changed peers are
|
||||
/// started, loops for removed or disabled peers are cancelled.
|
||||
///
|
||||
/// A bad `[peers]` block never reaches this method: the handler logs the
|
||||
/// config error and keeps the old registry, as TLS reload does.
|
||||
pub async fn reload(&self, new: PeersConfig) -> Result<(), PeerConfigError> {
|
||||
let mut st = lock_write(&self.state);
|
||||
let old = std::mem::take(&mut st.statuses);
|
||||
st.statuses = new
|
||||
.peer
|
||||
.iter()
|
||||
.map(|peer| {
|
||||
let mut status = old
|
||||
.get(&peer.name)
|
||||
.cloned()
|
||||
.unwrap_or_else(|| initial_status(peer));
|
||||
status.enabled = peer.enabled;
|
||||
status.url = peer.display_url();
|
||||
(peer.name.clone(), status)
|
||||
})
|
||||
.collect();
|
||||
st.config = Arc::new(new);
|
||||
let params = LoopConfig::from_config(&new);
|
||||
{
|
||||
let mut st = lock_write(&self.shared.state);
|
||||
let old = std::mem::take(&mut st.statuses);
|
||||
st.statuses = new
|
||||
.peer
|
||||
.iter()
|
||||
.map(|peer| {
|
||||
let mut status = old
|
||||
.get(&peer.name)
|
||||
.cloned()
|
||||
.unwrap_or_else(|| initial_status(peer));
|
||||
status.enabled = peer.enabled;
|
||||
status.url = peer.display_url();
|
||||
(peer.name.clone(), status)
|
||||
})
|
||||
.collect();
|
||||
st.config = Arc::new(new.clone());
|
||||
}
|
||||
if self.loops_enabled {
|
||||
self.resync_loops(&new, params);
|
||||
}
|
||||
Ok(())
|
||||
}
|
||||
|
||||
/// Health loops stop on this; nothing is running yet.
|
||||
pub async fn shutdown(&self) {}
|
||||
fn resync_loops(&self, new: &PeersConfig, params: LoopConfig) {
|
||||
let mut old = std::mem::take(&mut *lock_mutex(&self.loops));
|
||||
let mut kept = BTreeMap::new();
|
||||
for (name, entry) in std::mem::take(&mut old) {
|
||||
let wanted = new
|
||||
.enabled
|
||||
.then(|| {
|
||||
new.peer
|
||||
.iter()
|
||||
.find(|peer| peer.name == name && peer.enabled)
|
||||
})
|
||||
.flatten();
|
||||
match wanted {
|
||||
Some(peer) if entry.peer == *peer && entry.params == params => {
|
||||
kept.insert(name, entry);
|
||||
}
|
||||
_ => self.stop_loop(entry),
|
||||
}
|
||||
}
|
||||
*lock_mutex(&self.loops) = kept;
|
||||
if new.enabled {
|
||||
for peer in new.peer.iter().filter(|peer| peer.enabled) {
|
||||
if !lock_mutex(&self.loops).contains_key(&peer.name) {
|
||||
self.spawn_peer(peer);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// Stop every loop and wait (bounded) for the tasks to exit.
|
||||
pub async fn shutdown(&self) {
|
||||
self.shutdown.cancel();
|
||||
let joins: Vec<JoinHandle<()>> = std::mem::take(&mut *lock_mutex(&self.loops))
|
||||
.into_values()
|
||||
.map(|entry| entry.join)
|
||||
.collect();
|
||||
let join_all = async {
|
||||
for join in joins {
|
||||
let _ = join.await;
|
||||
}
|
||||
};
|
||||
if tokio::time::timeout(Duration::from_secs(5), join_all)
|
||||
.await
|
||||
.is_err()
|
||||
{
|
||||
tracing::warn!("peer health loops did not stop in time");
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// Everything the SIGHUP handler needs to re-read the `[peers]` block.
|
||||
#[derive(Clone)]
|
||||
pub struct PeersReloader {
|
||||
registry: Arc<PeerRegistry>,
|
||||
path: Option<std::path::PathBuf>,
|
||||
overrides: CliOverrides,
|
||||
}
|
||||
|
||||
impl PeersReloader {
|
||||
pub fn new(
|
||||
registry: Arc<PeerRegistry>,
|
||||
path: Option<std::path::PathBuf>,
|
||||
overrides: CliOverrides,
|
||||
) -> Self {
|
||||
Self {
|
||||
registry,
|
||||
path,
|
||||
overrides,
|
||||
}
|
||||
}
|
||||
|
||||
/// Re-read the whole config and apply only `[peers]`. A failure returns
|
||||
/// the config error; the old registry stays in place.
|
||||
pub async fn reload(&self) -> Result<(), PeerConfigError> {
|
||||
let cfg = Config::load(self.path.as_deref(), &self.overrides)?;
|
||||
self.registry.reload(cfg.peers).await
|
||||
}
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
@@ -159,6 +371,7 @@ mod tests {
|
||||
use super::*;
|
||||
use crate::peers::testing::{fake_peers_config, registry_with, static_registry, FakeTransport};
|
||||
use crate::peers::transport::PeerBody;
|
||||
use crate::peers::types::AuthState;
|
||||
|
||||
fn status(name: &str, up: bool) -> PeerStatus {
|
||||
PeerStatus {
|
||||
@@ -184,7 +397,7 @@ mod tests {
|
||||
assert_eq!(names, ["beta", "gamma"]);
|
||||
let b = reg.status("beta").unwrap();
|
||||
assert!(b.server.up);
|
||||
assert_eq!(b.api.auth, crate::peers::types::AuthState::Unknown);
|
||||
assert_eq!(b.api.auth, AuthState::Unknown);
|
||||
}
|
||||
|
||||
#[test]
|
||||
@@ -247,7 +460,7 @@ mod tests {
|
||||
async fn reload_keeps_surviving_status_and_prunes_removed() {
|
||||
let reg = registry_with(FakeTransport::new(), fake_peers_config(&["beta", "gamma"]));
|
||||
{
|
||||
let mut st = lock_write(®.state);
|
||||
let mut st = lock_write(®.shared.state);
|
||||
let beta = st
|
||||
.statuses
|
||||
.get_mut(&PeerName::parse("beta").unwrap())
|
||||
@@ -256,8 +469,7 @@ mod tests {
|
||||
}
|
||||
let mut new = fake_peers_config(&["beta", "delta"]);
|
||||
new.peer[0].enabled = false;
|
||||
let reloaded: Result<(), PeerConfigError> = reg.reload(new).await;
|
||||
reloaded.unwrap();
|
||||
reg.reload(new).await.unwrap();
|
||||
|
||||
let snap = reg.snapshot();
|
||||
assert_eq!(snap.len(), 2);
|
||||
@@ -267,4 +479,149 @@ mod tests {
|
||||
assert_eq!(snap[1].name.as_str(), "delta");
|
||||
assert!(!snap[1].server.up);
|
||||
}
|
||||
|
||||
async fn wait_for<F: Fn(&PeerRegistry) -> bool>(reg: &PeerRegistry, f: F) {
|
||||
let deadline = tokio::time::Instant::now() + Duration::from_secs(5);
|
||||
loop {
|
||||
if f(reg) {
|
||||
return;
|
||||
}
|
||||
assert!(
|
||||
tokio::time::Instant::now() < deadline,
|
||||
"condition not met: {:?}",
|
||||
reg.snapshot()
|
||||
);
|
||||
tokio::time::sleep(Duration::from_millis(10)).await;
|
||||
}
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn loops_mark_up_and_honour_the_failure_threshold() {
|
||||
let fake = FakeTransport::new();
|
||||
let mut cfg = fake_peers_config(&["beta"]);
|
||||
cfg.fail_threshold = 2;
|
||||
let reg = PeerRegistry::start_with_transport(cfg, fake.clone(), CancellationToken::new());
|
||||
wait_for(®, |r| {
|
||||
let s = r.status("beta").unwrap();
|
||||
s.server.up && s.api.up && s.api.auth == AuthState::Ok
|
||||
})
|
||||
.await;
|
||||
let up = reg.status("beta").unwrap();
|
||||
assert!(up.server.since.is_some());
|
||||
assert_eq!(up.api.peer_version.as_deref(), Some("0.0.0-fake"));
|
||||
|
||||
fake.peer("beta").tcp(Err(crate::peers::PeerError::Timeout));
|
||||
fake.peer("beta")
|
||||
.respond_err("/api/v1/ping", crate::peers::PeerError::Timeout);
|
||||
// One failure each: still up with the error visible.
|
||||
wait_for(®, |r| {
|
||||
let s = r.status("beta").unwrap();
|
||||
s.server.last_error.as_deref() == Some("timeout")
|
||||
&& s.api.last_error.as_deref() == Some("timeout")
|
||||
})
|
||||
.await;
|
||||
assert!(reg.status("beta").unwrap().server.up);
|
||||
|
||||
reg.shutdown().await;
|
||||
let after = reg.status("beta").unwrap();
|
||||
assert!(after.server.up, "one failure below threshold = still up");
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn a_hung_peer_does_not_delay_another_peers_checks() {
|
||||
let fake = FakeTransport::new();
|
||||
fake.peer("beta").delay(Duration::from_secs(2));
|
||||
fake.peer("gamma")
|
||||
.respond("/api/v1/ping", 200, "application/json", "{}");
|
||||
let reg = PeerRegistry::start_with_transport(
|
||||
fake_peers_config(&["beta", "gamma"]),
|
||||
fake.clone(),
|
||||
CancellationToken::new(),
|
||||
);
|
||||
wait_for(®, |r| {
|
||||
r.status("gamma").unwrap().api.last_checked.is_some()
|
||||
})
|
||||
.await;
|
||||
let gamma_checked = reg.status("gamma").unwrap().api.last_checked;
|
||||
assert!(reg.status("beta").unwrap().api.last_checked.is_none());
|
||||
assert!(gamma_checked.is_some());
|
||||
reg.shutdown().await;
|
||||
}
|
||||
|
||||
#[test]
|
||||
#[allow(clippy::result_large_err)] // figment::Jail closures return figment::Error
|
||||
fn reloader_rereads_the_peers_block_and_keeps_the_old_set_on_error() {
|
||||
use figment::Jail;
|
||||
|
||||
let reg = registry_with(FakeTransport::new(), fake_peers_config(&["beta"]));
|
||||
Jail::expect_with(|jail| {
|
||||
jail.create_file(
|
||||
"config.toml",
|
||||
r#"
|
||||
[peers]
|
||||
enabled = true
|
||||
api_interval_secs = 2
|
||||
[[peers.peer]]
|
||||
name = "gamma"
|
||||
url = "http://gamma:1984"
|
||||
token_file = "/run/credentials/skyg-linux-tracker.service/peer-gamma"
|
||||
allow_plain_http = true
|
||||
"#,
|
||||
)?;
|
||||
let reloader = PeersReloader::new(
|
||||
reg.clone(),
|
||||
Some(std::path::PathBuf::from("config.toml")),
|
||||
CliOverrides::default(),
|
||||
);
|
||||
tokio::runtime::Builder::new_current_thread()
|
||||
.enable_all()
|
||||
.build()
|
||||
.unwrap()
|
||||
.block_on(reloader.reload())
|
||||
.unwrap();
|
||||
let snap = reg.snapshot();
|
||||
assert_eq!(snap.len(), 1);
|
||||
assert_eq!(snap[0].name.as_str(), "gamma");
|
||||
assert_eq!(reg.config().api_interval_secs, 2);
|
||||
|
||||
let broken = PeersReloader::new(
|
||||
reg.clone(),
|
||||
Some(std::path::PathBuf::from("missing.toml")),
|
||||
CliOverrides::default(),
|
||||
);
|
||||
assert!(tokio::runtime::Builder::new_current_thread()
|
||||
.enable_all()
|
||||
.build()
|
||||
.unwrap()
|
||||
.block_on(broken.reload())
|
||||
.is_err());
|
||||
assert!(reg.contains("gamma"), "a bad reload keeps the old peers");
|
||||
Ok(())
|
||||
});
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn reload_moves_loops_to_the_new_peer_set() {
|
||||
let fake = FakeTransport::new();
|
||||
let reg = PeerRegistry::start_with_transport(
|
||||
fake_peers_config(&["beta"]),
|
||||
fake.clone(),
|
||||
CancellationToken::new(),
|
||||
);
|
||||
wait_for(®, |r| r.status("beta").unwrap().server.up).await;
|
||||
|
||||
reg.reload(fake_peers_config(&["beta", "gamma"]))
|
||||
.await
|
||||
.unwrap();
|
||||
wait_for(®, |r| r.status("gamma").unwrap().server.up).await;
|
||||
assert!(
|
||||
reg.status("beta").unwrap().server.up,
|
||||
"kept peer keeps state"
|
||||
);
|
||||
|
||||
reg.reload(fake_peers_config(&["gamma"])).await.unwrap();
|
||||
assert!(reg.status("beta").is_err());
|
||||
assert!(reg.contains("gamma"));
|
||||
reg.shutdown().await;
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user