Skip to content

fix(stargate): back off discovery and report registration failures #1823

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
barrygreengus wants to merge 5 commits into from
Draft

fix(stargate): back off discovery and report registration failures #1823

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
764539b
fix(stargate): back off discovery and report registration failures
barrygreengus Sep 11, 2026
a0faace
fix(stargate): preserve safe registration token error causes
barrygreengus Sep 14, 2026
e4a0528
refactor(pylon): inline registration token resolution
barrygreengus Sep 15, 2026
0b61f9f
fix(pylon): log each non-TLS registration failure
barrygreengus Sep 16, 2026
0c7eb21
refactor(pylon): use tracing macros for registration logs
barrygreengus Sep 16, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion MODULE.bazel.lock
View file
Open in desktop

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

56 changes: 41 additions & 15 deletions src/libraries/rust/stargate/crates/pylon-lib/src/registration/discovery.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,7 +28,7 @@ use stargate_runtime::{OwnedTask, TASK_SHUTDOWN_TIMEOUT};
use tracing::warn;

use super::grpc_endpoint::{
StargateGrpcEndpoint, log_stargate_grpc_certificate_failure, log_stargate_grpc_connect_attempt,
StargateGrpcEndpoint, grpc_error_chain, log_stargate_grpc_certificate_failure,
};
use super::topology::{RegistrationRouterTopology, publish_registration_router_topology};

Expand Down Expand Up @@ -165,23 +165,32 @@ async fn watch_stargate_endpoint(
return;
}

log_stargate_grpc_connect_attempt(&target, "watch_stargates", "lazy");
tracing::debug!(
transport = "grpc",
operation = "watch_stargates",
endpoint = %target,
connect_mode = "lazy",
"attempting Stargate gRPC connection"
);
let stream = match target.channel_endpoint(grpc_tls_ca_cert_pem.as_deref()) {
Ok(endpoint) => {
let mut client = StargateControlPlaneClient::new(endpoint.connect_lazy());
tokio::select! {
response = client.watch_stargates(WatchStargatesRequest {}) => {
match response {
Ok(response) => {
last_certificate_failure = None;
Some(response.into_inner())
}
Err(error) => {
last_certificate_failure = log_stargate_grpc_certificate_failure(
&target,
"watch_stargates",
&error,
last_certificate_failure,
&target, "watch_stargates", &error, last_certificate_failure,
);
warn!(
transport = "grpc",
operation = "watch_stargates",
endpoint = %target,
error = %grpc_error_chain(&error),
"Stargate gRPC operation failed"
);
None
}
Expand All @@ -197,6 +206,13 @@ async fn watch_stargate_endpoint(
error.as_ref(),
last_certificate_failure,
);
warn!(
transport = "grpc",
operation = "watch_stargates",
endpoint = %target,
error = %grpc_error_chain(error.as_ref()),
"Stargate gRPC operation failed"
);
None
}
};
Expand All @@ -206,16 +222,26 @@ async fn watch_stargate_endpoint(
return;
};
let snapshot = match message {
Ok(response) => response.map(|response| {
Ok(Some(response)) => {
last_certificate_failure = None;
watch_endpoint_snapshot_from_response(&watch_url, response)
}),
Some(watch_endpoint_snapshot_from_response(&watch_url, response))
}
Ok(None) => {
warn!(
transport = "grpc",
operation = "watch_stargates_stream",
endpoint = %target,
"Stargate discovery response stream ended"
);
None
}
Err(error) => {
last_certificate_failure = log_stargate_grpc_certificate_failure(
&target,
"watch_stargates_stream",
&error,
last_certificate_failure,
warn!(
transport = "grpc",
operation = "watch_stargates_stream",
endpoint = %target,
error = %grpc_error_chain(&error),
"Stargate gRPC operation failed"
);
None
}
Expand Down
116 changes: 54 additions & 62 deletions src/libraries/rust/stargate/crates/pylon-lib/src/registration/grpc_endpoint.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -124,10 +124,19 @@ pub(super) fn grpc_origin_uri(

impl fmt::Display for StargateGrpcEndpoint {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
let safe_endpoint = |endpoint: &str| match stargate_grpc_debug_target(endpoint) {
Ok(target) => format!("{}://{}:{}", target.scheme, target.host, target.port),
Err(_) => "<invalid endpoint>".to_string(),
};
let authority = safe_endpoint(&self.authority_endpoint());
if self.uses_authority_override() {
write!(f, "{} via {}", self.authority_addr, self.dial_addr)
write!(
f,
"{authority} via {}",
safe_endpoint(&self.dial_endpoint())
)
} else {
write!(f, "{}", self.authority_addr)
f.write_str(&authority)
}
}
}
Expand Down Expand Up @@ -164,12 +173,23 @@ pub(super) async fn connect_stargate_grpc_channel(
grpc_tls_ca_cert_pem: Option<&[u8]>,
operation: &'static str,
) -> anyhow::Result<Channel> {
log_stargate_grpc_connect_attempt(router_endpoint, operation, "eager");
tracing::debug!(
transport = "grpc",
operation,
endpoint = %router_endpoint,
connect_mode = "eager",
"attempting Stargate gRPC connection"
);
let channel = router_endpoint
.channel_endpoint(grpc_tls_ca_cert_pem)?
.connect()
.await?;
log_stargate_grpc_channel_connected(router_endpoint, operation);
tracing::debug!(
transport = "grpc",
operation,
endpoint = %router_endpoint,
"Stargate gRPC channel connected"
);
Ok(channel)
}

Expand Down Expand Up @@ -287,65 +307,37 @@ pub(super) fn log_stargate_grpc_certificate_failure(
Some(failure)
}

macro_rules! log_stargate_grpc_target {
($target:expr, $operation:expr, [$($extra:tt)*], $message:literal, $error_message:literal) => {{
if !tracing::enabled!(tracing::Level::DEBUG) {
return;
}
let dial_endpoint = $target.dial_endpoint();
let authority_endpoint = $target.authority_endpoint();
let override_authority = dial_endpoint != authority_endpoint;
match (
stargate_grpc_debug_target(&dial_endpoint),
stargate_grpc_debug_target(&authority_endpoint),
) {
(Ok(dial), Ok(authority)) => tracing::debug!(
transport = "grpc",
operation = $operation,
http_version = "h2",
dial_scheme = %dial.scheme,
tls = dial.scheme == "https",
dial_host = %dial.host,
dial_port = dial.port,
authority_host = %authority.host,
authority_port = authority.port,
override_authority,
$($extra)*
$message
),
(Err(_), _) | (_, Err(_)) => tracing::debug!(
transport = "grpc",
operation = $operation,
override_authority,
$($extra)*
$error_message
),
pub(super) fn grpc_error_chain(error: &(dyn Error + 'static)) -> String {
let mut causes = Vec::new();
for error in anyhow::Chain::new(error) {
// Parser diagnostics can include input excerpts from token files.
let detail = if let Some(error) = error.downcast_ref::<sonic_rs::Error>() {
format!(
"invalid JSON at line {} column {}",
error.line(),
error.column()
)
} else if let Some(status) = error.downcast_ref::<tonic::Status>() {
// Metadata and binary details are not needed to diagnose the RPC.
format!("gRPC {:?}: {}", status.code(), status.message())
} else if let Some(error) = error.downcast_ref::<reqwest::Error>() {
// Token-issuer URLs can contain credentials or sensitive queries.
// Keep the failure category and its sources without displaying the URL.
if error.is_timeout() {
"HTTP request timed out".into()
} else if error.is_connect() {
"HTTP connection failed".into()
} else {
"HTTP request failed".into()
}
} else {
error.to_string()
};
if causes.last() != Some(&detail) {
causes.push(detail);
}
}};
}

pub(super) fn log_stargate_grpc_connect_attempt(
target: &StargateGrpcEndpoint,
operation: &'static str,
connect_mode: &'static str,
) {
log_stargate_grpc_target!(
target,
operation,
[connect_mode,],
"attempting Stargate gRPC connection",
"could not parse Stargate gRPC endpoint for connection debug logging"
);
}

fn log_stargate_grpc_channel_connected(target: &StargateGrpcEndpoint, operation: &'static str) {
log_stargate_grpc_target!(
target,
operation,
[],
"Stargate gRPC channel connected",
"Stargate gRPC channel connected but endpoint metadata could not be parsed"
);
}
causes.join(": ")
}

fn normalize_addr_with_default_scheme(addr: &str, default_scheme: &str) -> String {
Expand Down
56 changes: 47 additions & 9 deletions src/libraries/rust/stargate/crates/pylon-lib/src/registration/router_stream.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,7 +29,8 @@ use stargate_proto::pb::{InferenceServerAck, InferenceServerRegistration, Infere
use stargate_runtime::{OwnedTask, TASK_SHUTDOWN_TIMEOUT};

use super::grpc_endpoint::{
StargateGrpcEndpoint, connect_stargate_grpc_channel, log_stargate_grpc_certificate_failure,
StargateGrpcEndpoint, connect_stargate_grpc_channel, grpc_error_chain,
log_stargate_grpc_certificate_failure,
};
use super::reverse_tunnel::{
ReverseTunnelState, reverse_tunnel_endpoint_from_ack, run_reverse_tunnel_loop,
Expand Down Expand Up @@ -58,17 +59,22 @@ pub(super) async fn run_router_registration_stream(
) => connection,
};
let (mut ack_stream, update_tx) = match connection {
Ok(connection) => {
last_certificate_failure = None;
connection
}
Ok(connection) => connection,
Err(error) => {
last_certificate_failure = log_stargate_grpc_certificate_failure(
&router_endpoint,
"register_inference_server",
error.as_ref(),
last_certificate_failure,
);
tracing::warn!(
transport = "grpc",
operation = "register_inference_server",
endpoint = %router_endpoint,
cluster_id = %config.cluster_id,
error = %grpc_error_chain(error.as_ref()),
"Stargate gRPC operation failed"
);
if stop
.run_until_cancelled(tokio::time::sleep(Duration::from_secs(1)))
.await
Expand Down Expand Up @@ -160,8 +166,32 @@ pub(super) async fn run_router_registration_stream(
connected
}
maybe_ack = ack_stream.message() => {
let Ok(Some(ack)) = maybe_ack else {
break false;
let ack = match maybe_ack {
Ok(Some(ack)) => {
last_certificate_failure = None;
ack
}
Ok(None) => {
tracing::warn!(
transport = "grpc",
operation = "register_inference_server_stream",
endpoint = %router_endpoint,
cluster_id = %config.cluster_id,
"Stargate registration response stream ended"
);
break false;
}
Err(error) => {
tracing::warn!(
transport = "grpc",
operation = "register_inference_server_stream",
endpoint = %router_endpoint,
cluster_id = %config.cluster_id,
error = %grpc_error_chain(&error),
"Stargate gRPC operation failed"
);
break false;
}
};
if config.reverse_tunnel {
let endpoint = reverse_tunnel_endpoint_from_ack(&ack);
Expand All @@ -185,7 +215,12 @@ pub(super) async fn run_router_registration_stream(
if let Some(task) = reverse_task {
task.shutdown(TASK_SHUTDOWN_TIMEOUT).await;
}
if stopped {
if stopped
|| stop
.run_until_cancelled(tokio::time::sleep(Duration::from_secs(1)))
.await
.is_none()
{
return;
}
}
Expand Down Expand Up @@ -298,7 +333,10 @@ pub(super) async fn open_registration_stream(
min_update_interval.as_millis().to_string().parse()?,
);
if let Some(provider) = auth_token_provider {
let token = provider.resolve_token().await?;
let token = provider
.resolve_token()
.await
.context("failed to resolve registration token")?;
request.metadata_mut().insert(
"authorization",
format!("Bearer {token}")
Expand Down
Loading
Loading