PSMDB 8.0.17 aarch64 SIGSEGV in WiredTiger checkpoint on Linux kernel 6.19 (Fedora CoreOS 43)

MongoDB Percona Server for MongoDB
, ,
1

Summary

Percona Server for MongoDB 8.0.17 (aarch64) crashes with SIGSEGV approximately 60 seconds after startup. The crash occurs in a background thread during
what appears to be the first WiredTiger checkpoint. No errors in mongod logs — the process dies silently.

Environment

  • Image: percona/percona-server-mongodb:8.0.17 (aarch64)
  • Image digest: sha256:f1170f8bf68d051816cd4d956ca1f6ee9885c6cf0e1e5db5dc00a137af3603ee
  • Host: macOS 26.x on Apple Silicon (M-series)
  • Container runtime: Podman Machine (tested both applehv and libkrun providers)
  • Broken guest kernel: 6.19.7-200.fc43.aarch64 (Fedora CoreOS 43)
  • Working guest kernel: 6.12.13-200.fc41.aarch64 (Fedora CoreOS 41)

Steps to reproduce

  1. Create a Podman Machine with Fedora CoreOS 43 (kernel 6.19.x) on macOS arm64
  2. Run the container:
    podman run -d --name mongo -p 27017:27017
    percona/percona-server-mongodb:8.0.17
    /usr/bin/mongod --bind_ip_all --replSet rs0
  3. Wait ~60 seconds
  4. Container exits with code 139 (SIGSEGV)

Coredump stack trace

Signal: 11/SEGV
Process 4289 (mongod) of user 1001 dumped core.

Stack trace of thread 13:
#0 0x0000aaaad95277fc n/a (/usr/bin/mongod + 0x75a77fc)
#1 0x0000aaaad8fe9e88 n/a (/usr/bin/mongod + 0x7069e88)
#2 0x0000aaaad8f16b04 n/a (/usr/bin/mongod + 0x6f96b04)
#3 0x0000aaaad8f10398 n/a (/usr/bin/mongod + 0x6f90398)
#4 0x0000aaaad8f0b8a0 n/a (/usr/bin/mongod + 0x6f8b8a0)
#5 0x0000aaaad8f0beec n/a (/usr/bin/mongod + 0x6f8beec)
#6 0x0000aaaad8f3696c n/a (/usr/bin/mongod + 0x6fb696c)
#7 0x0000aaaad8f36b5c n/a (/usr/bin/mongod + 0x6fb6b5c)
#8 0x0000ffff9abf6930 n/a (linux-vdso.so.1 + 0x930)
#9 0x0000aaaad621b770 n/a (/usr/bin/mongod + 0x429b770)
#10 0x0000aaaad621b770 n/a (/usr/bin/mongod + 0x429b770)
#11 0x0000aaaad621a2d4 n/a (/usr/bin/mongod + 0x429a2d4)
#12 0x0000aaaad6181584 n/a (/usr/bin/mongod + 0x4201584)
#13 0x0000aaaad6181c94 n/a (/usr/bin/mongod + 0x4201c94)
#14 0x0000ffff9a234e00 n/a (/usr/lib64/libc.so.6 + 0x80e00)
#15 0x0000ffff9a29f49c n/a (/usr/lib64/libc.so.6 + 0xeb49c)
ELF object binary architecture: AARCH64

What works

  • Same image digest on Fedora CoreOS 41 (kernel 6.12.13) — no crash
  • percona/percona-server-mongodb:7.0.15-multi on Fedora CoreOS 43 — no crash
  • Reproduced by 3 developers independently on different Macs

Notes

  • The crash timing (~60s) matches the default WiredTiger checkpoint_wait=60 interval
  • Frames #0–#7 appear to be mongod’s own crash signal handler (recursing after the original SEGV)
  • The original fault is in frames #9–#13 in a background thread (created via pthread, per frames #14–#15)
  • Both rootful and rootless Podman Machine configurations are affected
  • This may be related to aarch64 LSE atomics or memory barrier handling changes in kernel 6.19
2

Hi, thank you for bringing this to our attention. Would you mind opening a bug at Jira the dev team will take a look

3

image
image1118×744 44 KB

Your link leads to unknown destination. Also when I hit “create” I get unauthorized error. Before creating a forum topic I tried to use Jira.

4

My apologies, the correct link is Jira

let me know if you have any trouble during register