Description of problem: When a brick becomes full and a file is modified that causes the storage to report no space on the brick, the gluster volume will report a no space left on device. Version-Release number of selected component (if applicable): How reproducible: Fill a brick on a volume up, pick a file in that brick and add data to it so it will not longer fit in the full brick. Steps to Reproduce: 1. fill brick 2. edit a file to cause it to no longer fit on the full brick 3. No space left on device error reported Actual results: No space error. Expected results: I would expect gluster to find space for the file on an alternate brick and move the data to that brick as part of the write process. Additional info:
With Sharding this is not required. If we get into such scenario, we recommend sharding feature in general. For now, other than sharding, we wouldn't be focusing on fixing this problem. Time to add more machine and call it a rebalance, to fix the problem in real world.