Details
-
Improvement
-
Resolution: Unresolved
-
Medium
-
None
-
None
-
None
-
3
-
9223372036854775807
Description
currently we use OST_DESTROY and OST_SETATTR for each separate object and track it's committness to cancel corresponding llog record after.
there are few problems with this approach when used at scale (say,
1) network bandwidth (tiny payload)
2) sizeof(struct ptlrpc_request)=1048 * max_rpcs_in_progress – limit us to have enough sync RPCs and this can lead to llog overflow