Hi,
On Mon, Feb 01, 2021 at 03:46:16PM +0000, Kevin Traynor wrote: > On 01/02/2021 15:41, Kevin Traynor wrote: > > On 21/12/2020 13:10, Kevin Traynor wrote: > >> v6: > >> - Split patches to add log to existing interval param as a seperate fix > >> first > >> - Fix reference in vswitchd.xml > >> - Kept Acks as minor changes > >> > >> v5: > >> - Make naming more consistent in vswitchd.xml/NEWS/commit msgs > >> - Added Co-Authored-By > >> > >> v4: > >> - Changed naming of params > >> - Updated default #defines names > >> - Used atomic for param accessed in the pmd thread > >> - Revert to default if param range is exceeded > >> - Only set params if they are changed > >> - Add logging for params when they are changed > >> - Add log for auto load balance enabled/disabled when param changed > >> > > > > A question has been raised about whether this series should be > > backported to 2.13/2/14. > > > >> Christophe Fontaine (1): > >> dpif-netdev: Add parameters to configure PMD auto load balance. > >> > > > > ^ This is adding new params for auto load balance and it is not a direct > > fix for existing functionality. > > > > OTOH, I have heard that as the hardcoded load threshold is 95%, this is > > too high and params are needed to make the feature usable, so in that > > sense it could be considered a fix for the feature. This feature is still tagged as 'experimental'. Although I am not finding a formal project definition of 'experimental', I would assume it means 'not ready for production' and 'not supported' in terms of backports. The experimental features users are supposed to use latest version anyways otherwise improvements or other bug fixes might not be available. Perhaps we want to allow more users and then backporting a fix would help that, but then we might open a can of worms with other 'experimental' features and possibly instabilities to stable branches. It's blurring the lines. Changing a bit the angle, it makes little sense to me to have a branch-2.13 release containing code changes in experimental code, because from an administrator perspective, updating production systems has a risk and a cost, including evaluating the changes, then adding experimental changes increases both for no good reason to them. In summary: - I think we don't recommend users to use experimental features in production. - Stable branches are for production systems only. - Experimental feature users should always use latest code. Thoughts? Perhaps I missed another angle to this. fbl > >> Kevin Traynor (3): > >> dpif-netdev: Add log for PMD auto load balance interval parameter. > > > > ^ This is a clearer fix to 2.13/2.14 so I think it is a good candidate > > to be backported. > > > >> dpif-netdev: Add PMD auto load balance status log. > > > > ^ This is a small stats improvement, if the new params are backported I > > would take this too, if not, then it's not worth to backport as it would > > have conflicts. > > > > s/stats/logs/ > > > What do others think? > > > > Aside from NEWS, it is a clean backport, I can test and send backports > > if needed. > > > >> AUTHORS: Add Christophe Fontaine. > >> > >> AUTHORS.rst | 1 + > >> NEWS | 1 + > >> lib/dpif-netdev.c | 69 +++++++++++++++++++++++++++++++++++--------- > >> vswitchd/vswitch.xml | 29 +++++++++++++++++-- > >> 4 files changed, 85 insertions(+), 15 deletions(-) > >> > > > > _______________________________________________ > > dev mailing list > > [email protected] > > https://mail.openvswitch.org/mailman/listinfo/ovs-dev > > > > _______________________________________________ > dev mailing list > [email protected] > https://mail.openvswitch.org/mailman/listinfo/ovs-dev -- fbl _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
