From 5a0021c6a3ca698fa6ffb9cb5e9c8c3371444f4e Mon Sep 17 00:00:00 2001
From: Valentin Haudiquet <valentin.haudiquet@canonical.com>
Date: Thu, 21 May 2026 20:46:40 +0200
Subject: [PATCH] fix: p attach on dead/unknown sessions

---
 p/src/commands/attach.rs | 36 ++++++++++++++++++++++++++++++------
 1 file changed, 30 insertions(+), 6 deletions(-)

diff --git a/p/src/commands/attach.rs b/p/src/commands/attach.rs
index 30e7c0f..58510ff 100644
--- a/p/src/commands/attach.rs
+++ b/p/src/commands/attach.rs
@@ -11,20 +11,23 @@ pub fn execute(job_id: &str) -> Result<()> {
     let cfg = config::load()?;
     let job = db::find(job_id)?.with_context(|| format!("job '{}' not found", job_id))?;
 
-    if job.status != JobStatus::Running {
-        anyhow::bail!(
+    // Allow attach on Running (normal) and Unknown (worker was temporarily
+    // unreachable — the job may still be live).
+    match job.status {
+        JobStatus::Running | JobStatus::Unknown => {}
+        _ => anyhow::bail!(
             "job {} is not running (status: {})\n\
              use 'p logs {}' to view its output",
             job.short_id(),
             job.status.as_str(),
             job.short_id(),
-        );
+        ),
     }
 
     let worker = cfg.resolve_worker(Some(&job.worker))?.clone();
     let session = format!("p-{}", job.short_id());
-
     let sid = job.short_id().to_string();
+
     {
         let sid = sid.clone();
         ctrlc::set_handler(move || {
@@ -36,9 +39,30 @@ pub fn execute(job_id: &str) -> Result<()> {
         .ok();
     }
 
-    ssh::run_interactive(&worker, &format!("tmux attach -t '{}'", session))?;
+    let attach_status = ssh::run_interactive(&worker, &format!("tmux attach -t '{}'", session))?;
+
+    // If tmux exited non-zero the session doesn't exist on the worker.
+    if !attach_status.success() {
+        let exit_code = ssh::read_job_exitcode(&worker, &job.id);
+        if exit_code.is_none() {
+            // No exit code either — the job was lost without completing
+            // (e.g. worker restarted, tmux server killed).
+            eprintln!(
+                "error: tmux session for job {} no longer exists on '{}'.",
+                sid, worker.name
+            );
+            eprintln!("The job was likely interrupted (worker restart or tmux server exit).");
+            eprintln!("Use 'p logs {}' to see whatever output was captured.", sid);
+            db::save(&Job {
+                status: JobStatus::Unknown,
+                ..job
+            })?;
+            return Ok(());
+        }
+        // Exit code exists — job finished cleanly but the session was already
+        // cleaned up. Fall through to normal reconciliation.
+    }
 
-    // Same reconciliation logic as the run command.
     let exit_code = ssh::read_job_exitcode(&worker, &job.id);
     let now = Utc::now().timestamp();