Red Hat Bugzilla – Bug 763126
Sparse files not in sync
Last modified: 2015-12-01 11:45:32 EST
Before use glusterfs in production I made some tests:
Scenario: 2 replicated nodes
1. Create a sparse file of 50g in a folder of a mountpoint created by glusterfs.
2. Connect to a loop device (losetup /dev/loop0 ...)
-> both nodes in sync, show 50g, "du -sh" the real small size ...
3. Create a PV on /dev/loop0 (pvcreate /dev/loop0)
4. Create VG ...
5. Copy LV´s of a local VG into this VG (dd for each LV)
-> both sparse files grow in sync
Problem: If one node disapears for a time during the copy and joins again during that (test: kill glusterfsd and restart on one node) the result is that the sparse files have a different size and don´t get in sync again.
I ran the tests that you have specified.
On the backend, the ls and md5sum match after selfheal completes.
The du -sh size differs.
This is valid behavior of gluster handling spare files.