CEPH Journal is created unsafely
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Fuel for OpenStack |
Fix Released
|
Critical
|
Ryan Moe |
Bug Description
We prepare a raw partition for each CEPH journal device specified. Then the NUM_OSDS / JOURNAL_DEVS to assign a number of OSD's to each journal device. This is NOT OK, the OSD's can not share raw devices
<xarses> sagewk: hmm, multiple osd's appear to work with the same raw journal device
<xarses> (dumpling)
<sagewk> oh hmm, yeah it's not actually locking it.
<sagewk> that should be fixed
<sagewk> they may appear to work that way but will fail spectacularly on replay
pmanager needs to be updated so that the each journal devices is sub-divided into enough devices to provide a 1-1 mapping of devices.
pmanager should allocate for each partition the between 12-2Gb (less if less than 10% of the drive will remain free) or statically 2Gb for now if complicated to implement.
Case 1:
Journal: sdc
OSD: sdd sde sdf sdg
result pmanager should create sdc1 sdc2 sdc3 sdc4
Case 2:
Journal sdb sdc
OSD: sdd sde sdf sdg
result pmanager should create sdb1 sdb2 sdc1 sdc2
---
the fact should already be capable of pairing the OSD's and journals correctly with out modification
description: | updated |
description: | updated |
Changed in fuel: | |
assignee: | nobody → Ryan Moe (rmoe) |
Changed in fuel: | |
milestone: | 4.0 → 4.1 |
Changed in fuel: | |
milestone: | 4.1 → 4.0 |
Changed in fuel: | |
status: | Fix Committed → Fix Released |
Can we just use file-based targets and one partition for every journal?