|
Actually it's working now because we don't set the DEAD_OBJ on volatile files yet. I think we should have set it to make volatile file look more like unlinked file otherwise we still can access volatile from .lustre/fid path. If they strongly objected to, we can simply not fix it.
|
|
In upcoming kernel versions, they will also gain the concept of a volatile file and also the ability to link it to the namespace using flink_at() or similar, so it doesn't seem so unlikely that these files will be seen in normal usage.
|
|
So the actual situation is that opening_by_fid() a volatile file on a different client works just fine currently. Jinshan just threatened to remove this capability.
After further discussions, we all agreed that this functionality is required for parallel restore and we will just leave it in.
|
|
I manually verified that this works on today's master (lustre-2.4.93-2.6.32_358.18.1.el6_lustre.gdadfda6.x86_64_ga217228.x86_64) but it would be good to add a sanity test so that we don't regress on this.
|
Generated at Sat Feb 10 01:38:57 UTC 2024 using Jira 9.4.14#940014-sha1:734e6822bbf0d45eff9af51f82432957f73aa32c.