Ceph » RADOS

Hello everyone,
We test ceph 13.2.2 in our server using rbd service, we find the osd process restart in runtime When we test 4k sequential reading.
We have three servers as storage pool and use another server to test rbd service.We create 8 image in test server and use fio to read and write these images.

Configuration of each server is as follow: 
    OS:CentOs 7.5
    drivers:35 4T HDD 
    nvme: 1.6T nvme ssd as wal log and db
    memory:256GB

We check the log of the crashing osd process, it seems like the process can't get required memory that leads to crash. Attachment contains the detailed stack info.

We use the google-perftools to check the heap usage of ecah osd and find the heap freelist comsumed alomost 5~6GB per osd. We try to use command "ceph osd tell osd.* heap release" but it doesn't work, the memory still occupied by osd process.
    The os on our server is centos7.5 and which pagesize is 64kb on default. We guess the huge memory consumption is related to the hugepage, so we recompiled the kernel image by setting the pagesize to 4kb, the total memory comsumption is 23GB when we use the new kernel to test 4kb sequential reading.The osd process will not restart again.
    I am wondering why the osd consume that huge memory when pagesize is 64kb, is it a tcmalloc bug or something else? Is there a config in ceph or tcmalloc that can limit the memory usage of each osd ?