[SERVER-26743] Have a pre-allocated "crash zone" file to more gracefully handle out of disk scenarios Created: 24/Oct/16 Updated: 24/Oct/16 Resolved: 24/Oct/16 |
|
| Status: | Closed |
| Project: | Core Server |
| Component/s: | Storage |
| Affects Version/s: | None |
| Fix Version/s: | None |
| Type: | New Feature | Priority: | Minor - P4 |
| Reporter: | Rod Adams | Assignee: | Unassigned |
| Resolution: | Duplicate | Votes: | 0 |
| Labels: | None | ||
| Remaining Estimate: | Not Specified | ||
| Time Spent: | Not Specified | ||
| Original Estimate: | Not Specified | ||
| Issue Links: |
|
||||||||
| Participants: | |||||||||
| Description |
|
Of late, we've had a rash of customer run out of disk space, and then have to perform resyncs or other actions to make their instance consistent again after freeing up some empty space. What if we made an option to pre-allocate a 2-3G "crash pad" file, which when we experience an out of disk scenario, we can steer records of the writes we'd like to have done into that file, and then crash, with an intention of being able to restart later on, and replay the missed writes, returning us to a consistent state. This is very similar to regular journalling, but the key difference would be that the space is pre-allocated, so would be able to work around running out of space. I've left out ~1000 implementation details I haven't worked out, mainly just noting the general thought. |
| Comments |
| Comment by Rod Adams [ 24/Oct/16 ] |
|
That it does. |
| Comment by Ramon Fernandez Marina [ 24/Oct/16 ] |
|
This looks like a duplicate of |
| Comment by Rod Adams [ 24/Oct/16 ] |
|
Excellent question. I know we've had a lot of cases like this of late (these things come in waves), but I'll have to track down if it's all MMAP, and/or some other piece. I'll get back to you. |
| Comment by Eric Milkie [ 24/Oct/16 ] |
|
Do you need to resync WiredTiger nodes when they run out of disk space, or does this only affect mmap? |