question about pg mapping

question about pg mapping

[Henry C Chang]

[-- Attachment #1: Type: text/plain, Size: 573 bytes --]

Hi all,

In my ceph cluster, I saw a pg membership change as below:

At osdmap epch 160, pg 1.184 was mapped to [4, 6].

        osdmaptool: osdmap file '160'
         parsed '1.184' -> 1.184
        1.184 raw [4,6] up [4,6] acting [4,6]

At osdmap epoch 161, however, pg 1.184 was mapped to [0, 5].

        osdmaptool: osdmap file '161'
         parsed '1.184' -> 1.184
        1.184 raw [0,5] up [0,5] acting [0,5]

The osdmap are attached.
Is it normal that the pg membership changes like this? I thought at
least one member will stay in the pg in this case?

--
Henry

[-- Attachment #2: 160 --]
[-- Type: application/octet-stream, Size: 5559 bytes --]

[-- Attachment #3: 161 --]
[-- Type: application/octet-stream, Size: 5559 bytes --]

Re: question about pg mapping

[Gregory Farnum]

On Fri, May 20, 2011 at 3:05 AM, Henry C Chang <henry.cy.chang@gmail.com> wrote:
> Hi all,
>
> In my ceph cluster, I saw a pg membership change as below:
>
> At osdmap epch 160, pg 1.184 was mapped to [4, 6].
>
>        osdmaptool: osdmap file '160'
>         parsed '1.184' -> 1.184
>        1.184 raw [4,6] up [4,6] acting [4,6]
>
> At osdmap epoch 161, however, pg 1.184 was mapped to [0, 5].
>
>        osdmaptool: osdmap file '161'
>         parsed '1.184' -> 1.184
>        1.184 raw [0,5] up [0,5] acting [0,5]
>
> The osdmap are attached.
> Is it normal that the pg membership changes like this? I thought at
> least one member will stay in the pg in this case?
What caused the osd map to change?
And did the mapping stay that way, or did it get overridden in one of
the subsequent map updates?

In general we'd expect the mappings to maintain membership, but I
don't think it's a guarantee. What you will see, though, are manual
overrides of the CRUSH mapping to a different acting set in subsequent
updates, as the new OSDs in the PG start peering and realize that an
old OSD would be better-suited for the job (since it has data and none
of them do).
-Greg
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html