Trace reconnect failure reasons

Cargo.lock

@@ -436,7 +436,7 @@ dependencies = [
 
 [[package]]
 name = "dosh"
-version = "1.0.0-rc38"
+version = "1.0.0-rc39"
 dependencies = [
  "anyhow",
  "base64",

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "dosh"
-version = "1.0.0-rc38"
+version = "1.0.0-rc39"
 edition = "2024"
 license = "MIT"
 

src/bin/dosh-client.rs

@@ -828,6 +828,9 @@ struct TraceReport {
     sent_input_events: usize,
     pty_write_events: usize,
     reconnect_events: usize,
+    reconnect_none_events: usize,
+    reconnect_live_fail_events: usize,
+    reconnect_ticket_fail_events: usize,
     unknown_resume_events: usize,
     ticket_attach_start_events: usize,
     ticket_attach_ok_events: usize,
@@ -1345,6 +1348,15 @@ fn summarize_trace_file_with_mode(
         if event.contains("reconnect") || event.contains("resume") {
             report.reconnect_events += 1;
         }
+        if event == "client.reconnect_none" {
+            report.reconnect_none_events += 1;
+        }
+        if event == "client.reconnect_live_fail" {
+            report.reconnect_live_fail_events += 1;
+        }
+        if event == "client.reconnect_ticket_fail" {
+            report.reconnect_ticket_fail_events += 1;
+        }
         if event == "server.resume_unknown_client" {
             report.unknown_resume_events += 1;
         }
@@ -1413,8 +1425,11 @@ fn print_trace_report(label: &str, report: &TraceReport) {
         report.hex_events
     );
     println!(
-        "  reconnect: events={} unknown_resume={} ticket_attach={}/{}/{} roam={}",
+        "  reconnect: events={} none={} live_fail={} ticket_fail={} unknown_resume={} ticket_attach={}/{}/{} roam={}",
         report.reconnect_events,
+        report.reconnect_none_events,
+        report.reconnect_live_fail_events,
+        report.reconnect_ticket_fail_events,
         report.unknown_resume_events,
         report.ticket_attach_start_events,
         report.ticket_attach_ok_events,
@@ -1452,6 +1467,12 @@ fn trace_report_warnings(report: &TraceReport) -> Vec<String> {
             report.ticket_attach_reject_events
         ));
     }
+    if report.reconnect_none_events > 0 {
+        warnings.push(format!(
+            "{} reconnect attempt(s) exhausted live resume and ticket attach",
+            report.reconnect_none_events
+        ));
+    }
     if report.server_mouseish_pty_write_events > 0 {
         warnings.push(format!(
             "{} mouse-like input event(s) reached the server PTY",
@@ -7106,7 +7127,8 @@ async fn reconnect(
             ("last_rendered_seq", cred.last_rendered_seq.to_string()),
         ],
     );
-    if let Ok((frame, next)) = try_live_resume(socket, cred, send_seq, size.0, size.1).await {
+    match try_live_resume(socket, cred, send_seq, size.0, size.1).await {
+        Ok((frame, next)) => {
             *cred = next;
             frame_buffer.clear();
             predictor.reset();
@@ -7127,11 +7149,24 @@ async fn reconnect(
             );
             return Ok(Some(frame));
         }
+        Err(err) => {
+            dosh::trace::event(
+                "client.reconnect_live_fail",
+                &[("reason", format!("{err:#}"))],
+            );
+        }
+    }
 
-    let Ok((frame, next)) = try_ticket_attach(socket, cred, size.0, size.1, Vec::new()).await
-    else {
+    let (frame, next) = match try_ticket_attach(socket, cred, size.0, size.1, Vec::new()).await {
+        Ok(value) => value,
+        Err(err) => {
+            dosh::trace::event(
+                "client.reconnect_ticket_fail",
+                &[("reason", format!("{err:#}"))],
+            );
             dosh::trace::event("client.reconnect_none", &[]);
             return Ok(None);
+        }
     };
     *cred = next;
     *send_seq = 2;
@@ -10832,6 +10867,32 @@ mod tests {
         );
     }
 
+    #[test]
+    fn trace_report_warns_when_reconnect_exhausts_all_paths() {
+        let dir = tempfile::tempdir().unwrap();
+        let path = dir.path().join("client.log");
+        fs::write(
+            &path,
+            concat!(
+                "ts_ms=10 pid=1 event=client.start version=1.0.0\n",
+                "ts_ms=11 pid=1 event=client.reconnect_start session=term mode=read-write last_rendered_seq=9\n",
+                "ts_ms=12 pid=1 event=client.reconnect_live_fail reason='live resume timed out'\n",
+                "ts_ms=13 pid=1 event=client.reconnect_ticket_fail reason='ticket attach timed out'\n",
+                "ts_ms=14 pid=1 event=client.reconnect_none\n",
+            ),
+        )
+        .unwrap();
+
+        let report = summarize_trace_file(&path, 10).unwrap();
+        let warnings = trace_report_warnings(&report);
+        assert_eq!(report.reconnect_none_events, 1);
+        assert_eq!(report.reconnect_live_fail_events, 1);
+        assert_eq!(report.reconnect_ticket_fail_events, 1);
+        assert!(warnings.iter().any(|warning| {
+            warning == "1 reconnect attempt(s) exhausted live resume and ticket attach"
+        }));
+    }
+
     #[test]
     fn update_installer_uses_platform_native_script() {
         assert_eq!(