Assert failure(PID 1640 [0x00000668], Thread: 19388 [0x4bbc]): x > base

[cshung]

This fails occassionally on the plug test case with DATAS turned on.

Assert failure(PID 1640 [0x00000668], Thread: 19388 [0x4bbc]): x > base

CORECLR! SVR::gc_heap::calculate_new_heap_count + 0xB70 (0x00007ffe`27d00590)
CORECLR! SVR::gc_heap::update_end_gc_time_per_heap + 0x298 (0x00007ffe`27d364d8)
CORECLR! SVR::gc_heap::gc1 + 0x155C (0x00007ffe`27d12aac)
CORECLR! SVR::gc_heap::garbage_collect + 0x99E (0x00007ffe`27d114fe)
CORECLR! SVR::gc_heap::gc_thread_function + 0x580 (0x00007ffe`27d13170)
CORECLR! SVR::gc_heap::gc_thread_stub + 0x7E (0x00007ffe`27d1358e)
CORECLR! <lambda_c106ffbf81d302a6ee5c9510e81a6997>::<lambda_invoker_cdecl> + 0x74 (0x00007ffe`279dd0a4)
KERNEL32! BaseThreadInitThunk + 0x1D (0x00007ffe`b52d257d)
NTDLL! RtlUserThreadStart + 0x28 (0x00007ffe`b664aa48)
    File: C:\dev\runtime\src\coreclr\gc\gc.cpp:25235
    Image: C:\dev\runtime\artifacts\tests\coreclr\windows.x64.Checked\Tests\Core_Root\corerun.exe

Configuration: DevBox, Windows, x64, chk, datas

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/gc
See info in area-owners.md if you want to be subscribed.

[mangod9]

Is this a recent failure or has this been seen before?

[cshung]

Is this a recent failure or has this been seen before?

I just noticed it.

[mangod9]

Ok, could be related to recent change from @Maoni0

[cshung]

Ok, could be related to recent change from @Maoni0

Probably not, it was ran on a commit before her changes.

It looks like the problem has to do with this math.

log_with_base ((smoothed_median_throughput_cost_percent - target_tcp), log_base)

The implementation of log_with_base asserts that x must be greater than base.

It looks like x is derived from measurement, so it could be really close to target_tcp to the point that it is less than log_base, and so it will fail.

I don't know where the new logic went, it might well be gone after @Maoni0's change, as far as I can tell, in the latest code, it looks like log_with_base is now a dead function that nobody calls.

[mangod9]

Ok then guessing it might not be relevant again. We can leave it open but if we don't hit it again in the next week or so can get it closed.

[Maoni0]

somehow I never got notified.. this code is no longer in use so we can ignore this. thanks!