[lustre-discuss] Serious interoperability problem lustre 2.8 server side with lustre 2.9 clients

[Fernando Perez]

Dear lustre experts.

I detect a serious interoperability problem between lustre 2.8 server 
side and lustre 2.9 clients:

We have a lustre 2.8 on mgs/mds and oss's and lustre 2.8 clients. 
Yesterday I update a couple of clients to lustre 2.9 and I detect that 
this clients cannot read a lot of files of our lustre filesystem. The 
outputs when we try to read several files is the following:

[root at turing lustre-2.9.0_el7_client]# lfs getstripe 
/mnt/lustre/users/cgabarro/L3_cadena_interna/comparacio_SSSL2_ARGO/filter_case_3/TOT_results_SMOS_argo_filter_v3_20110221_20110302.dat
error opening 
/mnt/lustre/users/cgabarro/L3_cadena_interna/comparacio_SSSL2_ARGO/filter_case_3/TOT_results_SMOS_argo_filter_v3_20110221_20110302.dat: 
No such file or directory (2)
llapi_semantic_traverse: Failed to open 
'/mnt/lustre/users/cgabarro/L3_cadena_interna/comparacio_SSSL2_ARGO/filter_case_3/TOT_results_SMOS_argo_filter_v3_20110221_20110302.dat': 
No such file or directory (2)
error: getstripe failed for 
/mnt/lustre/users/cgabarro/L3_cadena_interna/comparacio_SSSL2_ARGO/filter_case_3/TOT_results_SMOS_argo_filter_v3_20110221_20110302.dat

As you can see in other clients with lustre 2.8 clients there is no 
problem to read this file:

[root at gaia ~]# lfs getstripe 
/mnt/lustre/users/cgabarro/L3_cadena_interna/comparacio_SSSL2_ARGO/filter_case_3/TOT_results_SMOS_argo_filter_v3_20110221_20110302.dat
/mnt/lustre/users/cgabarro/L3_cadena_interna/comparacio_SSSL2_ARGO/filter_case_3/TOT_results_SMOS_argo_filter_v3_20110221_20110302.dat
lmm_stripe_count:   1
lmm_stripe_size:    1048576
lmm_pattern:        1
lmm_layout_gen:     2
lmm_stripe_offset:  3
     obdidx         objid         objid         group
          3           5235570         0x4fe372                 0

This problem only happens with lustre 2.9 clients and there is no errors 
in the oss and mds/mgs servers. I see that this problems only happens 
when I read some files, not all files, from some ldiskfs ost's.

After downgrade to 2.8 lustre the clients we cannot reproduce this problem.

Regards.


-- 
=============================================
Fernando Pérez
Institut de Ciències del Mar (CMIMA-CSIC)
Departament Oceanografía Física i Tecnològica
Passeig Marítim de la Barceloneta,37-49
08003 Barcelona
Phone:  (+34) 93 230 96 35
=============================================