GC pause times grow exponentially when DOTNET_EnableEventLog=1

[a74nh]

Description

Running GCPerfsim with DOTNET_EnableEventLog=1 the pause times for GC2 get increasingly longer.
This does not happen when DOTNET_EnableEventLog is not set.

Found on Arm64 Linux. Not sure if on X64 (suspect so) and/or Windows (suspect not).
TODO: Test on these

Reproduction Steps

dotnet-trace collect --output "$OUTPUT_FILE" --show-child-io --clrevents gc+gchandle -- corerun GCPerfSim.dll -tc 2 -tagb 200 -tlgb 2 -lohpi 0 -sohsi 50 -ramb 20 -rlmb 0.2 -sohpi 0

open the logs in perfview

Expected behavior

Gen2 pause times are small.

Results without DOTNET_EnableEventLog set:

Actual behavior

Gen2 pause times increase exponentially. See pause MSec column:

Regression?

Unsure

Known Workarounds

Don't set DOTNET_EnableEventLog

Configuration

Head: aec77dd

Linux Ubuntu 24.04.02. Cobalt 100.

Other information

No response

[dotnet-policy-service]

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

[Maoni0]

my guess is this is not a GC issue at all.

I think this could be due to your also enabling the verbose level events as we observed. and since there are more and more survival with each gen2 GC, it meant you are probably spending more and more time in walking the heap to report for heap walk events. if you look at the the samples in methods like walk_heap_* they are probably taking more and more samples during the BGC pauses since that's when we walk to heap to reporting data for those events.