In environments running MongoDB on RHEL-based systems, the ftdc thread responsible for collecting diagnostic metrics may enter an uninterruptible sleep (D state). This occurs due to a known kernel-level bug in autofs, where a closed file descriptor is mishandled during automount routines, causing the thread to hang indefinitely in autofs_mount_wait. This renders FTDC inoperative and impacts observability.

Technical Analysis:{}

Thread Details:{}

• Affected Thread: ftdc (ID: 923934)

• State: D (Blocked I/O)

• Wait Channel: autofs_mount_wait

• Stack Trace:

$ cat /proc/923911/task/923934/stack
[<0>] autofs_wait+0x319/0x79d
[<0>] autofs_mount_wait+0x49/0xf0
[<0>] autofs_d_automount+0xe6/0x1f0
[<0>] follow_managed+0x17f/0x2e0
[<0>] lookup_fast+0x135/0x2a0
[<0>] walk_component+0x48/0x300
[<0>] path_lookupat.isra.43+0x79/0x220
[<0>] filename_lookup.part.58+0xa0/0x170
[<0>] user_statfs+0x43/0xa0
[<0>] __do_sys_statfs+0x20/0x60
[<0>] do_syscall_64+0x5b/0x1b0
[<0>] entry_SYSCALL_64_after_hwframe+0x61/0xc6 

Root Cause:{}

• A known bug in autofs (RHEL Bugzilla #2023740, Red Hat Case ID available)

• Improper error handling during delayed automount logic.

• Results in EBADF or EINVAL, blocking the kernel thread waiting on autofs_mount_wait.

Impact:{}

• FTDC is unable to collect metrics, affecting monitoring tools.

• Can result in performance degradation or hangs if other threads also block.

• Troubleshooting is hindered by lack of observability.

Proposed Solutions:{}

Solution A: Add FTDC Watchdog and Self-Restart Mechanism{}

Description:

Introduce a lightweight watchdog thread inside the ftdc subsystem to monitor the health/status of FTDC’s main thread. If a D state is detected beyond a threshold (e.g., 5 minutes), gracefully stop and restart FTDC collection.

Solution B: Implement SERVER-103431