SIGSEGV crash in T1_TbltTblUp0 thread during stress test on macOS

[cms-cms]

Environment

• Version/Build: seekdb (macOS/arm64) deployed from package: https://mirrors.aliyun.com/oceanbase/community/stable/darwin/15/arm64/seekdb--100000092026051510-macos15-arm64.tar.gz
• Deployment Time: 2026-07-07 17:10:27
• OS: macOS 26.2 (25C56), arch=arm64
• Work Directory: /Users/Shared/seekdb_jobs/duomo_merged_4h
• Port: 4545
• Binary UUID: 4886FD04-2EB4-36E4-8E7A-4A9A722044BD
• Is this an RD-provided temporary version? No

Description

During a stress test (duomo_merged workload), the seekdb process crashes with a segmentation fault (SIGSEGV) after approximately 5 minutes of runtime. The crash occurs in the T1_TbltTblUp0 thread, which appears to be related to tenant 1 tablet/table metadata updates. The crash is associated with a fork operation on macOS.

Steps to Reproduce

1. Run the stress test job stress_duomo_lite on a macOS system.
   • Jenkins job link: http://:9090/jenkins/job/stress_duomo_lite/
2. Select the macOS version for the run.
3. The crash occurs consistently after about 5 minutes.

Logs & Crash Information

• Database Log: /Users/Shared/seekdb_jobs/duomo_merged_4h/base/log/seekdb.log
• Crash Report (macOS .ips file): /Users/admin/Library/Logs/DiagnosticReports/seekdb-2026-07-07-021657.ips
  • Also available at: /Users/Shared/seekdb_jobs/duomo_merged_4h/base/log/duomo_merged_stack/20260707_172116/diagnostic_reports/seekdb-2026-07-07-021657.ips

Key Crash Details from .ips file:

app_name: seekdb
timestamp: 2026-07-07 02:16:57 -0700
incident_id: 79799A03-F549-4ABD-94DF-0B4F716D8359
os_version: macOS 26.2 (25C56)
pid: 48810
procLaunch: 2026-07-07 02:15:43 -0700
captureTime: 2026-07-07 02:16:56 -0700
seekdb uuid: 4886FD04-2EB4-36E4-8E7A-4A9A722044BD

exception: EXC_BAD_ACCESS
signal: SIGSEGV
subtype: KERN_INVALID_ADDRESS at 0x0000000000000000
termination: Segmentation fault: 11
faultingThread: 57
thread name: T1_TbltTblUp0

asi: libsystem_c.dylib: crashed on child side of fork pre-exec

Crash Thread Stack Summary (offsets):

0   __pthread_kill
1   pthread_kill
2   raise
3   seekdb offset 0xec0c8a8
4   _sigtramp
5   seekdb offset 0x3f7c308
6   seekdb offset 0x3f8dee4
7   seekdb offset 0x3f87480
8   seekdb offset 0xec0d524
9   seekdb offset 0xec1c650

Partial Symbolization Attempt (requires dSYM):

atos -> 0x000000010ec0c8a8 (in seekdb) + 1652
atos -> 0x0000000103f7c308 (in seekdb) + 7604
atos -> 0x0000000103f8dee4 (in seekdb) + 120

Analysis & Proposed Next Steps

1. Root Cause: A null pointer dereference (KERN_INVALID_ADDRESS at 0x0) in the T1_TbltTblUp0 thread.
2. Context: The crash occurs "on child side of fork pre-exec", indicating it's related to a fork() system call on macOS, likely during high-concurrency DDL, vector, or tablet metadata operations.
3. Reproducibility: The issue is consistently reproducible on macOS by running the specified stress test.
4. Required for Debugging: The provided binary is stripped. To map the stack offsets to specific functions and line numbers, the debug symbols (dSYM) for the binary with UUID 4886FD04-2EB4-36E4-8E7A-4A9A722044BD are needed.

Impact

A crash during stress testing indicates a stability issue in the macOS version of seekdb under concurrent load, specifically in code paths involving metadata updates and process forking.

Additional Context

• Test Changes: Standard stress test scenario.
• Other Environmental Issues: None reported.
• Is the site preserved? The crash report and logs are available at the paths specified above.