[LU-19208] lmm_stripe_count: 0 in DoM component triggers divide-by-zero in MPIIO

• Type: Bug
• Resolution: Unresolved
• Priority: Medium
• Affects Version/s: Lustre 2.15.7
• Labels: llnl topllnl
• Environment: lustre-2.15.7_1.llnl-2.t4.x86_64
• Severity: 3
• Assignee: Peter Jones
• Reporter: Olaf Faaland

Description

When PFL is used to create a file with a DoM component, the DoM component has lmm_stripe_count == 0,

# lfs getstripe RVDC_03_03022_10_2023.h5 | head -n 15
RVDC_03_03022_10_2023.h5
  lcm_layout_gen:    12
  lcm_mirror_count:  1
  lcm_entry_count:   7
    lcme_id:             1
    lcme_mirror_id:      0
    lcme_flags:          init
    lcme_extent.e_start: 0
    lcme_extent.e_end:   65536
      lmm_stripe_count:  0
      lmm_stripe_size:   65536
      lmm_pattern:       mdt
      lmm_layout_gen:    0
      lmm_stripe_offset: 0 

This triggers a divide-by-zero in MPIIO in in ADIOI_LUSTRE_Get_striping_info(), here:

 63         avail_cb_nodes =
 64                 stripe_count * ADIOI_MIN(nprocs_for_coll/stripe_count, CO);

Where the value in stripe_count originally comes from ADIOI_LUSTRE_Open() via:

115             err = ioctl(fd->fd_sys, LL_IOC_LOV_SETSTRIPE, lum); 

and

137         fd->hints->striping_factor = lum->lmm_stripe_count;

 
Is lmm_stripe_count 0 for DoM components because it has to be, or does Lustre ignore it internally, and so the value hasn't mattered?

I'm wondering if Lustre should change, or MPIIO, or both.

Issue Links

• is related to: LU-14337 return valid stripe_count and stripe_size instead of 0 for DoM files
• is related to: LU-10026 Client-side data compression