I want plasma store to work with huge pages
I tried creating a a file system and activate huge pages with. Like the example in plasmastore
sudo mkdir -p /mnt/hugepages
gid=id -g
uid=id -u
sudo mount -t hugetlbfs -o size=170G -o uid=$uid -o gid=$gid none /mnt/hugepages
sudo bash -c “echo $gid > /proc/sys/vm/hugetlb_shm_group”
sudo bash -c “echo 20000 > /proc/sys/vm/nr_hugepages”
and I set in ray.init _plasma_directory="/mnt/hugepages"
This is the output of raylet.out
2021-12-16 10:50:07,727 I 83753 83753] io_service_pool.cc:35: IOServicePool is running with 1 io_service.
[2021-12-16 10:50:07,728 I 83753 83753] store_runner.cc:30: Allowing the Plasma store to use up to 161.631GB of memory.
[2021-12-16 10:50:07,728 I 83753 83753] store_runner.cc:46: Starting object store with directory /mnt/hugepages, fallback /tmp/ray, and huge page support disabled
[2021-12-16 10:50:07,728 I 83753 83782] dlmalloc.cc:146: create_and_mmap_buffer(161630912520, /mnt/hugepages/plasmaXXXXXX)
[2021-12-16 10:50:07,751 C 83753 83782] dlmalloc.cc:164: failed to ftruncate file /mnt/hugepages/plasmaUeo901, errorInvalid argument
*** StackTrace Information ***
_ZN3ray12GetCallTraceB5cxx11Ev
ray::SpdLogMessage::Flush()
ray::RayLog::~RayLog()
plasma::create_and_mmap_buffer()
plasma::fake_mmap()
dlmalloc
plasma::internal_memalign()
plasma::PlasmaAllocator::Allocate()
plasma::PlasmaAllocator::PlasmaAllocator()
plasma::PlasmaStoreRunner::Start()
std::_State_impl<>::_M_run()
start_thread
As a secondery issue I saw that it prints in the raylet.out
“and huge page support disabled”
I see in raylet main.cc the following argument
DEFINE_bool(huge_pages, false, “Whether enable huge pages”);
and it is used in the code regarding the plasma store
However I don’t see anywhere were it is set to true
So how can I enable it?