feat(telemetry)!: make telemetry worker wasm-compatible for the TraceExporter

LICENSE-3rdparty.csv

@@ -498,6 +498,7 @@ wasm-bindgen-macro,https://github.com/rustwasm/wasm-bindgen/tree/master/crates/m
 wasm-bindgen-macro-support,https://github.com/rustwasm/wasm-bindgen/tree/master/crates/macro-support,MIT OR Apache-2.0,The wasm-bindgen Developers
 wasm-bindgen-shared,https://github.com/rustwasm/wasm-bindgen/tree/master/crates/shared,MIT OR Apache-2.0,The wasm-bindgen Developers
 web-sys,https://github.com/rustwasm/wasm-bindgen/tree/master/crates/web-sys,MIT OR Apache-2.0,The wasm-bindgen Developers
+web-time,https://github.com/daxpedda/web-time,MIT OR Apache-2.0,The web-time Authors
 webpki-root-certs,https://github.com/rustls/webpki-roots,CDLA-Permissive-2.0,The webpki-root-certs Authors
 webpki-roots,https://github.com/rustls/webpki-roots,CDLA-Permissive-2.0,The webpki-roots Authors
 widestring,https://github.com/VoidStarKat/widestring-rs,MIT OR Apache-2.0,The widestring Authors

bin_tests/src/lib.rs

@@ -136,11 +136,20 @@ fn cargo_build_artifact(c: &ArtifactsBuild) -> anyhow::Result<PathBuf> {
     };
 
     if c.panic_abort == Some(true) {
+        // `-C panic=abort` suppresses `.eh_frame` unwind tables, which the aarch64
+        // backtrace unwinder relies on; without them `RUST_BACKTRACE` can loop forever
+        // on certain binary layouts (rust-lang/rust#123733), hanging the test until the
+        // runner OOMs. Force the tables back on so backtraces terminate. This is
+        // layout-sensitive, so it can appear/disappear across unrelated changes; keep
+        // until #123733 (fix: rust-lang/rust#143613) ships in a stable toolchain.
         let existing_rustflags = std::env::var("RUSTFLAGS").unwrap_or_default();
         let new_rustflags = if existing_rustflags.is_empty() {
-            "-C panic=abort".to_string()
+            "-C panic=abort -C force-unwind-tables=yes".to_string()
         } else {
-            format!("{} -C panic=abort", existing_rustflags)
+            format!(
+                "{} -C panic=abort -C force-unwind-tables=yes",
+                existing_rustflags
+            )
         };
         build_cmd.env("RUSTFLAGS", new_rustflags);
     }

datadog-sidecar/src/self_telemetry.rs

@@ -4,12 +4,15 @@ use crate::config::Config;
 use crate::log;
 use crate::service::SidecarServer;
 use crate::watchdog::WatchdogHandle;
+use libdd_capabilities_impl::NativeCapabilities;
 use libdd_common::{tag, tag::Tag, MutexExt};
 use libdd_telemetry::data::metrics::{MetricNamespace, MetricType};
 use libdd_telemetry::metrics::ContextKey;
-use libdd_telemetry::worker::{
-    LifecycleAction, TelemetryActions, TelemetryWorkerBuilder, TelemetryWorkerHandle,
-};
+use libdd_telemetry::worker::{LifecycleAction, TelemetryActions, TelemetryWorkerBuilder};
+
+/// The sidecar runs the telemetry worker on native, so its handle is pinned to
+/// [`NativeCapabilities`].
+type TelemetryWorkerHandle = libdd_telemetry::worker::TelemetryWorkerHandle<NativeCapabilities>;
 use manual_future::ManualFuture;
 use std::sync::atomic::{AtomicU64, Ordering};
 use std::time::Duration;
@@ -205,7 +208,7 @@ impl SelfTelemetry {
             crate::sidecar_version!().to_string(),
         );
         builder.config = self.config.clone();
-        let (worker, join_handle) = builder.spawn();
+        let (worker, join_handle) = builder.spawn::<NativeCapabilities>();
 
         let metrics = MetricData {
             worker: &worker,

datadog-sidecar/src/service/stats_flusher.rs

@@ -20,7 +20,9 @@ use http::uri::PathAndQuery;
 use libdd_capabilities_impl::{HttpClientCapability, NativeCapabilities};
 use libdd_common::{Endpoint, MutexExt};
 use libdd_telemetry::config::Config;
-use libdd_telemetry::worker::TelemetryWorkerHandle;
+/// Sidecar's telemetry worker is native-only, so its handle is pinned to
+/// [`NativeCapabilities`].
+type TelemetryWorkerHandle = libdd_telemetry::worker::TelemetryWorkerHandle<NativeCapabilities>;
 use libdd_trace_stats::stats_exporter::{StatsExporter, StatsMetadata};
 use std::collections::HashMap;
 use std::ffi::CString;

datadog-sidecar/src/service/telemetry.rs

@@ -23,6 +23,7 @@ use std::time::{Duration, Instant};
 use tokio::task::JoinHandle;
 use zwohash::ZwoHasher;
 
+use libdd_capabilities_impl::NativeCapabilities;
 use libdd_common::tag::Tag;
 use libdd_telemetry::worker::TelemetryWorkerBuilder;
 use serde::{Deserialize, Serialize};
@@ -33,7 +34,11 @@ use std::time::SystemTime;
 use libdd_telemetry::config::Config;
 use libdd_telemetry::data::{self, Integration};
 use libdd_telemetry::metrics::{ContextKey, MetricContext};
-use libdd_telemetry::worker::{LifecycleAction, TelemetryActions, TelemetryWorkerHandle};
+use libdd_telemetry::worker::{LifecycleAction, TelemetryActions};
+
+/// Sidecar's telemetry worker is native-only, so its handle is pinned to
+/// [`NativeCapabilities`].
+type TelemetryWorkerHandle = libdd_telemetry::worker::TelemetryWorkerHandle<NativeCapabilities>;
 use manual_future::ManualFuture;
 use serde_with::{serde_as, VecSkipError};
 use tokio::time::{sleep, sleep_until, Instant as TokioInstant};

libdd-capabilities-impl/src/http.rs

@@ -4,6 +4,8 @@
 //! Native HTTP client implementation backed by hyper.
 
 mod native {
+    use std::fs::OpenOptions;
+    use std::io::Write;
     use std::sync::{Arc, OnceLock};
 
     use libdd_capabilities::http::{HttpClientCapability, HttpError};
@@ -26,6 +28,30 @@ mod native {
         }
     }
 
+    /// Write `body` as a newline-terminated record to the file referenced by `uri` (which must
+    /// have a `file://` scheme), then return a synthetic 202 response.
+    fn write_to_file_endpoint(
+        uri: &http::Uri,
+        body: bytes::Bytes,
+    ) -> Result<http::Response<bytes::Bytes>, HttpError> {
+        let path = libdd_common::decode_uri_path_in_authority(uri)
+            .map_err(|e| HttpError::Other(anyhow::anyhow!("invalid file:// URI: {e}")))?;
+        let mut file = OpenOptions::new()
+            .create(true)
+            .append(true)
+            .open(&path)
+            .map_err(|e| HttpError::Other(anyhow::anyhow!("opening {path:?}: {e}")))?;
+        let mut record = body.to_vec();
+        record.push(b'\n');
+        file.write_all(&record)
+            .map_err(|e| HttpError::Other(anyhow::anyhow!("writing {path:?}: {e}")))?;
+
+        http::Response::builder()
+            .status(http::StatusCode::ACCEPTED)
+            .body(bytes::Bytes::new())
+            .map_err(|e| HttpError::Other(e.into()))
+    }
+
     impl HttpClientCapability for NativeHttpClient {
         fn new_client() -> Self {
             Self {
@@ -39,8 +65,15 @@ mod native {
             req: http::Request<bytes::Bytes>,
         ) -> impl std::future::Future<Output = Result<http::Response<bytes::Bytes>, HttpError>> + MaybeSend
         {
-            let client = self.client.get_or_init(new_default_client).clone();
+            let client_lock = self.client.clone();
             async move {
+                // file:// URIs short-circuit to the on-disk recorder used by tests.
+                if req.uri().scheme_str() == Some("file") {
+                    let (parts, body) = req.into_parts();
+                    return write_to_file_endpoint(&parts.uri, body);
+                }
+
+                let client = client_lock.get_or_init(new_default_client).clone();
                 let hyper_req = req.map(Body::from_bytes);
 
                 let response = client

libdd-crashtracker/Cargo.toml

@@ -49,6 +49,8 @@ anyhow = "1.0"
 chrono = {version = "0.4", default-features = false, features = ["std", "clock", "serde"]}
 cxx = { version = "1.0", optional = true }
 errno = "0.3"
+libdd-capabilities = { version = "2.1.0", path = "../libdd-capabilities" }
+libdd-capabilities-impl = { version = "3.0.0", path = "../libdd-capabilities-impl" }
 libdd-common = { version = "5.1.0", path = "../libdd-common" }
 libdd-telemetry = { version = "6.0.0", path = "../libdd-telemetry" }
 http = "1.1"

libdd-crashtracker/src/crash_info/telemetry.rs

@@ -8,6 +8,8 @@ use crate::{ErrorKind, SigInfo};
 use super::{CrashInfo, Metadata, TARGET_TRIPLE};
 use anyhow::Context;
 use chrono::{DateTime, Utc};
+use libdd_capabilities::HttpClientCapability;
+use libdd_capabilities_impl::NativeCapabilities;
 use libdd_common::Endpoint;
 use libdd_telemetry::{
     build_host,
@@ -355,9 +357,9 @@ impl TelemetryCrashUploader {
         self.send_telemetry_payload(&payload).await
     }
 
-    /// Helper to perform actual HTTP (or file) submission via configured telemetry client
+    /// Helper to perform actual HTTP submission via the native HTTP capability.
     async fn send_telemetry_payload(&self, payload: &data::Telemetry<'_>) -> anyhow::Result<()> {
-        let client = libdd_telemetry::worker::http_client::from_config(&self.cfg);
+        let client = NativeCapabilities::new_client();
         let req = request_builder(&self.cfg)?
             .method(http::Method::POST)
             .header(
@@ -372,19 +374,20 @@ impl TelemetryCrashUploader {
                 libdd_telemetry::worker::http_client::header::REQUEST_TYPE,
                 "logs",
             )
-            .body(serde_json::to_string(&payload)?.into())?;
-
-        tokio::time::timeout(
-            core::time::Duration::from_millis({
-                if let Some(endp) = self.cfg.endpoint() {
-                    endp.timeout_ms
-                } else {
-                    Endpoint::DEFAULT_TIMEOUT
-                }
-            }),
-            client.request(req),
-        )
-        .await??;
+            .body(libdd_capabilities::Bytes::from(serde_json::to_vec(
+                &payload,
+            )?))?;
+
+        let timeout = core::time::Duration::from_millis({
+            if let Some(endp) = self.cfg.endpoint() {
+                endp.timeout_ms
+            } else {
+                Endpoint::DEFAULT_TIMEOUT
+            }
+        });
+        tokio::time::timeout(timeout, client.request(req))
+            .await
+            .map_err(|_| anyhow::anyhow!("Telemetry crash report timed out"))??;
 
         Ok(())
     }

libdd-data-pipeline/Cargo.toml

@@ -24,6 +24,7 @@ serde_json = "1.0.127"
 bytes = "1.11.1"
 sha2 = "0.10"
 either = "1.13.0"
+futures = { version = "0.3", default-features = false, features = ["alloc"] }
 tokio = { version = "1.23", features = [
     "rt",
     "sync",
@@ -46,6 +47,7 @@ libdd-tinybytes = { version = "1.1.1", path = "../libdd-tinybytes", features = [
     "bytes_string",
     "serialization",
 ] }
+web-time = "1"
 
 [target.'cfg(not(target_arch = "wasm32"))'.dependencies]
 tokio = { version = "1.23", features = ["time", "test-util"], default-features = false }

libdd-data-pipeline/src/otlp/metrics.rs

@@ -16,8 +16,9 @@ use libdd_trace_utils::otlp_encoder::mapper::status_code;
 use libdd_trace_utils::otlp_encoder::OtlpResourceInfo;
 use serde_json::{json, Value};
 use std::sync::{Arc, Mutex};
-use std::time::{Duration, SystemTime};
+use std::time::Duration;
 use tracing::error;
+use web_time::SystemTime;
 
 const METRIC_NAME: &str = "traces.span.sdk.metrics.duration";
 const NANOS_PER_SECOND: f64 = 1_000_000_000.0;

libdd-data-pipeline/src/telemetry/metrics.rs

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: Apache-2.0
 
 //! Provides an abstraction layer to hold metrics that comes from 'SendDataResult'.
+use libdd_capabilities::{HttpClientCapability, MaybeSend, SleepCapability};
 use libdd_common::tag;
 use libdd_telemetry::data::metrics::{MetricNamespace, MetricType};
 use libdd_telemetry::metrics::ContextKey;
@@ -146,7 +147,9 @@ impl Index<MetricKind> for Metrics {
 
 impl Metrics {
     /// Creates a new Metrics instance
-    pub fn new(worker: &TelemetryWorkerHandle) -> Self {
+    pub fn new<C: HttpClientCapability + SleepCapability + MaybeSend + Sync + 'static>(
+        worker: &TelemetryWorkerHandle<C>,
+    ) -> Self {
         let mut keys = Vec::new();
         for metric in METRICS {
             let key = worker.register_metric_context(
@@ -171,6 +174,7 @@ impl Metrics {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use libdd_capabilities_impl::NativeCapabilities;
     use libdd_telemetry::worker::TelemetryWorkerBuilder;
 
     #[cfg_attr(miri, ignore)]
@@ -182,7 +186,7 @@ mod tests {
             "0.1".to_string(),
             "1.0".to_string(),
         )
-        .spawn();
+        .spawn::<NativeCapabilities>();
 
         let metrics = Metrics::new(&worker);