[MINOR] kill_process_tree SIGKILL verify timeout too short #32

New issue

Open

opened 2026-05-11 10:52:02 +00:00 by thabeta · 1 comment

thabeta commented

2026-05-11 10:52:02 +00:00

Owner

Problem

After sending SIGKILL in kill_process_tree, the code sleeps 100ms then verifies processes are dead. On a heavily loaded system, 100ms might not be enough for the kernel to reap all processes, especially if they're in uninterruptible sleep (D state).

Impact

kill_process_tree may report partial failures when processes are actually dying but haven't been reaped yet. Or it may miss dead processes that are in D state.

Files

crates/my_init_server/src/process.rs -- kill_process_tree

Suggested Fix

Use a retry loop with exponential backoff (e.g., 10ms, 20ms, 50ms, 100ms, 200ms) instead of a single 100ms sleep.

## Problem After sending SIGKILL in `kill_process_tree`, the code sleeps 100ms then verifies processes are dead. On a heavily loaded system, 100ms might not be enough for the kernel to reap all processes, especially if they're in uninterruptible sleep (D state). ## Impact `kill_process_tree` may report partial failures when processes are actually dying but haven't been reaped yet. Or it may miss dead processes that are in D state. ## Files - `crates/my_init_server/src/process.rs` -- `kill_process_tree` ## Suggested Fix Use a retry loop with exponential backoff (e.g., 10ms, 20ms, 50ms, 100ms, 200ms) instead of a single 100ms sleep.

thabeta added the

bug

label

2026-05-11 10:53:00 +00:00

salmaelsoly commented

2026-05-11 13:35:24 +00:00

Member

Classification: valid-bug — kill_process_tree uses single 100ms sleep after SIGKILL; on loaded systems, kernel may not have reaped processes yet.

Fixed 100ms sleep is insufficient under load. Use retry loop with exponential backoff.

> Classification: valid-bug — kill_process_tree uses single 100ms sleep after SIGKILL; on loaded systems, kernel may not have reaped processes yet. Fixed 100ms sleep is insufficient under load. Use retry loop with exponential backoff.