fix(agent): prevent zombie processes on agent cleanup

Root cause:
- AgentDriverClient::kill() is called during Drop to clean up child processes
- Old implementation called child.wait() and ignored errors with let _
- When wait() fails (e.g., process already killed by signal), child is not reaped and becomes zombie

Fix:
- Use try_wait() non-blocking polling, up to 100ms timeout
- Avoid long blocking in Drop during async cleanup
- Still perform final blocking wait() as fallback after timeout
- Log all wait errors for debugging

Impact:
- All JDBC agents (Oracle/DB2/DM8/XuguDB etc.)
- Prevents zombie process accumulation during long-running sessions
This commit is contained in:
t8y2 committed 2026-06-22 14:59:05 +08:00
1 parent 7adee1cacf
commit 791d55661e
1 file changed
+18 -2
+18 -2
View File
@@ -722,8 +722,24 @@ impl AgentDriverClient {
if let Err(e) = self.child.kill() {
log::warn!("Failed to kill agent process: {e}");
}
// Reap the child to avoid zombie processes
let _ = self.child.wait();
// Reap the child to avoid zombie processes.
// Use try_wait() with a timeout instead of blocking wait() to avoid
// hanging in Drop during async cleanup. Poll up to 100ms for the
// process to exit after kill().
for _ in 0..10 {
match self.child.try_wait() {
Ok(Some(_status)) => return,
Ok(None) => std::thread::sleep(Duration::from_millis(10)),
Err(e) => {
log::warn!("Failed to wait for agent process: {e}");
return;
}
}
}
// Final blocking wait as a last resort
if let Err(e) = self.child.wait() {
log::warn!("Final wait failed for agent process: {e}");
}
}
pub fn pid(&self) -> u32 {