ceph installer should support journal configuration while creating the OSDs: 1. journal co-residing on the OSD 2. journal resides on a dedicated disk 3. user provides the size of the journal, the installer create it on the dedicated disk or on the same disk based the configuration
(In reply to Nishanth Thomas from comment #0) > ceph installer should support journal configuration while creating the OSDs: > > 1. journal co-residing on the OSD > 2. journal resides on a dedicated disk > 3. user provides the size of the journal, the installer create it on the > dedicated disk or on the same disk based the configuration We are not going to support collocation of an OSD with a Journal as that creates a big performance impact and is merely a convenient thing to do. The only OSD deployment scenario that the installer is going to expose is #2 #3 Should not be needed because the format required is: {"/path/to/osd/device": "/path/to/journal/device"} There is no need to provide the size of the journal.
Collocation of an OSD and a Journal, as well as "OSD in a directory" are not ideal OSD deployments and will go against "best practices". See: https://bugzilla.redhat.com/show_bug.cgi?id=1213084
So let me summarize: 1. journal co-residing on the OSD - not supported 2. journal resides on a dedicated disk - supported. But there will not be any option to specify the journal size. That means the whole disk will be consumed as the OSD journal. There was a discussion in sds-mgmt around the journal configuration which is different than this. I will reply to that mail with this Bz so that everyone in the loop
(In reply to Nishanth Thomas from comment #4) > So let me summarize: > > 1. journal co-residing on the OSD - not supported > 2. journal resides on a dedicated disk - supported. But there will not be > any option to specify the journal size. That means the whole disk will be > consumed as the OSD journal. > > There was a discussion in sds-mgmt around the journal configuration which is > different than this. I will reply to that mail with this Bz so that everyone > in the loop I can't edit my previous comment. Apologies. It is possible to specify the size of the journal and that can be set when configuring OSDs: http://docs.ceph.com/ceph-installer/docs/#post--api-osd-configure- It is a required field.
So that means if I give disk of size 10 GB and specify journal size as 2 GB, the ceph-installer will create a lv out of 2 GB and allocated to the requesting OSD. Another OSDs can also use the same disk as journal if space is available. Does my understanding is right?
(In reply to Nishanth Thomas from comment #6) > So that means if I give disk of size 10 GB and specify journal size as 2 GB, > the ceph-installer will create a lv out of 2 GB and allocated to the > requesting OSD. Another OSDs can also use the same disk as journal if space > is available. Does my understanding is right? Yes that is correct. 1 journal device can be used for more than 1 OSD. The request allows this kind of mapping: {"/dev/sdb": "/dev/sdx", "/dev/sdc": "/dev/sdx"} ^ OSD ^ Journal ^ OSD ^ Same Journal
We should allow multiple journals to reside on SSD devices up to a max of 6 (for this release - it's simple let's start here - there are more complicated scenarios later). *We should not allow multiple journals on spinning media. In the SSD scenario I would imagine we simply add default journal partition sizes to the disk after zapping it. are we utilizing the salt parted module? https://docs.saltstack.com/en/latest/ref/modules/all/salt.modules.parted.html
Hi Gregory, can you please set the "dev_ack" flag to "+"?
This bug was accidentally moved from POST to MODIFIED via an error in automation, please see mmccune with any questions
I am closing this since we are allowing the configuration for different journals per OSD. Feel free to re-open if the documented configure options are not adequate: http://docs.ceph.com/ceph-installer/docs/#post--api-osd-configure-