[LU-6350] LFSCK 4: lock the MDT-object based on prediction for preparing bad linkEA

Fri, 1 May 2015 03:56:07 +0000

Currently, when the namespace LFSCK verifies the object's linkEA, it will check the linkEA without ldlm/dt lock firstly, if finds inconsistency, it will acquire the ldlm/dt lock on the object, then re-check the linkEA, if still invalid, it will repair the bad linkEA. For the system containing many inconsistent objects, such double-check mechanism is inefficient.

Be as some improvement, the LFSCK can make some prediction, for example: if former object contains bad linkEA, then when verifies current object, the namespace LFSCK can acquire the ldlm/dt lock on the object in advance; if the prediction is wrong, it will NOT take ldlm/dt lock in advance for next object.

Whamcloud Community JIRA

[LU-6350] LFSCK 4: lock the MDT-object based on prediction for preparing bad linkEA