[Gluster-devel] Regarding afr making sure flush, fsync complete delayed post op
Pranith Kumar Karampuri
pkarampu at redhat.com
Fri Dec 7 12:32:22 UTC 2012
Here is one possible solution for achieving this: http://review.gluster.org/#change,4280.
The solution posted achieves it by pausing flush/fsync fop until the delayed post-op is completed.
Writevs are unwound only after its post-op is marked for delay. This prevents the race of flush not
witnessing delayed post-op in fd-ctx. The problem is that the solution is ugly.
I am wondering if you guys have other solutions to achieve the same without making flush/fsync a transaction.
More information about the Gluster-devel