On 18 Apr 2015, at 17:36, Doug Ledford dledford@redhat.com wrote:

On Sat, 2015-04-18 at 15:09 +0100, Justin Clift wrote:

...

On 18 Apr 2015, at 10:39, Bart Van Assche bart.vanassche@sandisk.com wrote:

...

On 03/11/15 18:01, Justin Clift wrote:

...

On 10 Mar 2015, at 11:09, Bart Van Assche bart.vanassche@sandisk.com wrote:

...

Hello,

Sorry but I will not be able to attend this meeting because at that time I will be on a plane. But I have been thinking further about how to provide SCST RPMs for CentOS. How about using the approach documented on http://wiki.centos.org/HowTos/BuildingKernelModules (weak updates) and taking the risk that the IB drivers may break if an update introduces an RDMA ABI change ?

The "take a risk X might break" bit doesn't really sound suitable for the CentOS audience.

That being said... how often does the RDMA ABI change? If it's once every X years, then it might be live-able (with sufficient catches/warning so users aren't affected).

(back from traveling - sorry for the delay in replying)

Hello Justin,

So far I have only seen the RDMA ABI change between RHEL releases (e.g. from 7.0 to 7.1) but not yet due to a kernel update. Which of course is no guarantee that a change of the RDMA API will never happen in the future between kernel releases ...

Hmmmm... in general RHEL has an attitude of "don't change ABI in updates",

It's not an attitude, it's a hard requirement that requires managerial approval for an exception to break it.

...

but RDMA may or may not have different assumptions. I have no idea. ;)

It does. We exempt the kernel portion of the RDMA stack from any ABI claims entirely. For user space, we make a best effort to preserve backward binary compatibility, but not forward. Meaning if you compile a user space app against 7.0, our 7.1 and later updates will all be backward compatible to your compiled program. However, we add extensions, so if you compile against 7.2 let's say, and use one of the new extensions, then you program will not run on 7.0. However, keep in mind that this is a best effort. On occasion, with managerial approval, we break this too. The RDMA simply moves too fast to keep it static through the life of a RHEL product.

Thanks Doug. :)

+ Justin

-- GlusterFS - http://www.gluster.org

An open source, distributed file system scaling to several petabytes, and handling thousands of clients.

My personal twitter: twitter.com/realjustinclift