[LMR-6] Why lemur needs two lustre mount for agent and plugin

Thu, 31 Oct 2019 05:04:07 +0000

Hello Team,

I was trying hsm operations with lemur set up (Followed steps given in https://wiki.hpdd.intel.com/display/PUB/HPDD+HSM+Agent+and+Data+Movers+%28Lemur%29+Getting+Started+Guide) and observed that lhsmd creates 2 separate agent and plugin mounts to do hsm operations. Can anyone explain why such configuration (2mounts) required ?

The reason behind this query is that I observed few difference in IO pattern followed by lhsm_posix tool and lemur.

ex: lfs hsm_restore file

In lhsm_posix case, I see 2 sync calls going to ost once after the file is copied back(restored)
1. ll_writepages() leads to ofd_commitrw() on ost
2. ll_fsync() leads to ofd_sync_hdl()
This call flow is true in all restore cases with lhsm_posix tool.

But in case of lemur, I observed
1. ofd_sync_hdl() is going 1st then
2. ofd_commitrw()
This happens during every restore tried with lemur

I have seen few issues while trying lemur tool (mainly block size update not happening correctly when restore is triggered through cat or md5sum on released file)

Thanks,

Whamcloud Community JIRA

[LMR-6] Why lemur needs two lustre mount for agent and plugin