[Core] dlmalloc allocate bottom-most memory chunk failed #21439
Conversation
|
@Catch-Bull The issue you linked is a Ray Tune issue, is it the correct one? |
| # keep reference for fourth object, avoid released by plasma GC. | ||
| __ = ray.put(data) # noqa | ||
|
|
||
| # Check fourth object whether allocate to disk |
There was a problem hiding this comment.
Can you use check_spilled_mb to see if it was disk allocated instead?
| @@ -56,6 +56,9 @@ int fake_munmap(void *, int64_t); | |||
| #define HAVE_MORECORE 0 | |||
| #define DEFAULT_MMAP_THRESHOLD MAX_SIZE_T | |||
| #define DEFAULT_GRANULARITY ((size_t)128U * 1024U) | |||
| // Copied from plasma_allocator.cc variable kAllocationAlignment, | |||
| // make sure to keep in sync | |||
There was a problem hiding this comment.
Can you add a link to GH issue for context?
oh... it's my mistake, I've made it correct. |
| __ = ray.put(data) # noqa | ||
|
|
||
| # Check fourth object allocate in memory. | ||
| _check_spilled_mb(address, spilled=180) |
There was a problem hiding this comment.
you can also write a unit test in C++ demonstrate the issue. Here are some examples you can follow:
https://github.com/ray-project/ray/blob/master/src/ray/object_manager/plasma/test/fallback_allocator_test.cc#L35
|
Thanks for addressing the comments. Looks the lint issues is unrelated but let me retry them before merging. |
Why are these changes needed?
fix dlmalloc allocate bug, details in here #21310
Related issue number
Checks
scripts/format.shto lint the changes in this PR.