See  Issue 610629  and https://codereview.chromium.org/1956323002/ for more context.
TL;DR I think requires just a few lines patch to s/EnabledStateObserver/AsyncEnabledStateObserver/ + a weakptr factory in v8 sampling profiler.

This is the problem: we recently found out that EnabledStateObserver is not safe to use if the observer is not indefinitely lived (i.e. if it gets destroyed at some point). The reason is that the EnabledStateObserver callbacks happen synchronously from the tracing side on the FILE thread. Nothing can prevent the observer to get destroyed on another thread at the same time.

This indefinite observer lifetime is the case for most observers, but not all.
The exceptions I found here are:
1. BlameContext, which has been fixed in https://codereview.chromium.org/1956333002/
2. V8SamplingProfiler, which this bug is about

The suggested fix is to switch V8SamplingProfiler to the new AsyncEnabledStateObserver.
This new class guarantees that the OnTraceLog{Dis,En}abled are posted to the same thread where AddEnabledStateObserver was called. This prevents races on dtor.
If I read the code correctly V8SamplingProfiler is destroyed together with the RenderThread (question: when does that happen? Is it only on renderer shutdown?).

Also I do suspect that this is causing the DrMemory issues that have been suppressed in tools/valgrind/drmemory/suppressions_full.txt