Metadata writeback cache support (LU-10938)

[LU-13046] WBC2: lockless IO Created: 04/Dec/19  Updated: 04/Dec/19

Status: Open
Project: Lustre
Component/s: None
Affects Version/s: None
Fix Version/s: None

Type: Technical task Priority: Minor
Reporter: Qian Yingjin Assignee: Qian Yingjin
Resolution: Unresolved Votes: 0
Labels: None

Rank (Obsolete): 9223372036854775807

 Description   

If metadata update does not need to revoke the root WBC EX lock (which usually need to get the WBC EX lock back on the children files or directories), these metadata operations on MDT can not even execute in a lockless mode, which means it does not need to acquire any locks (LDLM lock).

Also I/O operations to Lustre OSTs can be also in the lockless mode (does not need any extent lock for IO) if the file is under the protection of a root WBC EX lock.

This is possible for WBC because in WBC_FLUSH_AGE_LOCK_HOLD flush mode, a regular file may be with data flushed into Lustre OSTs, but the root WBC EX lock protected it is not revoked.

but when the WBC EX lock of the regular file is revoking, during the lock blocking callback, the client must acquire all required extent locks on the file which has flushed data into Lustre OSTs, and then can it release the WBC EX lock.

With this lockless IO support for WBC,  the performance should be improved.

 


Generated at Sat Feb 10 02:57:54 UTC 2024 using Jira 9.4.14#940014-sha1:734e6822bbf0d45eff9af51f82432957f73aa32c.