Repository navigation
pwsh 7.6.6 on Linux segfaults in libcoreclr.so under parallel load (exit 139) #28119
Description
Environment: PowerShell 7.6.6 (portable tar.gz install), .NET 10.0.12, Ubuntu with kernel 7.0.0-34-generic, x64, 28 cores.
What happens: while a test suite runs many pwsh processes in parallel (ForEach-Object -Parallel -ThrottleLimit 8, each test starting further pwsh processes; load average 11 to 33), a pwsh process now and then dies with SIGSEGV (exit code 139). Four times on 2026-09-28, always at the same place:
pwsh[...]: segfault at 28 ip 00007b3737ddc9da sp 00007b2bedbfe940 error 4 in libcoreclr.so[3db9da,7b3737bc9000+4b1000]
The offset in libcoreclr.so is 0x3db9da every time, the fault address 0x28 (a read through a null pointer plus 0x28).
Repro: no reliable one. 400 short pwsh -NoProfile -Command runs, 24 at a time, did not crash. It shows up about once in three to eight suite runs of about 50 s, in processes that run a script for several seconds.
Expected: no crash. A core dump can be collected if that helps.
Activity
kartalbas commentedon Oct 4, 2026
Six more crashes since 2026-10-03, on a second machine with the same versions: PowerShell 7.6.6, .NET 10.0.12, kernel 7.0.0-34-generic, x64, 32 logical CPUs. Swap was in heavy use (about 25 of 31 GB), load average 14 to 46.
Five of six are at the same offset 0x3db9da with fault address 0x28, as in the first report. One is at a second offset, 0x44246c, with fault address 0x7.
The workload is the same as before: a test suite with
ForEach-Object -Parallel -ThrottleLimit 8, where each test starts further short-livedpwsh -NoProfile -Fileprocesses as stand-ins for other commands. Three runs of that suite at once, at load 46, did not crash, so it still has no reliable reproduction.Name a crash of pwsh when the suite goes red (#20)