On 8/11/12 1:18 PM, "Wido den Hollander" <w...@widodh.nl> wrote:
> > >On 08/10/2012 10:19 PM, Alena Prokharchyk wrote: >> As I'm not a git expert, I would appreciate any help from the community >>in >> figuring out what is the best way to merge vpc branch into master. >> >> >> Couple of ways I can think of: >> >> 1) git checkout master; git merge -s recursive -Xignore-space-change vpc >> >> Puts all commits from the vpc that are differnt from master, to the >>master >> branch. >> >> Pros: >> >> * no tree rebase happens >> * easier to revert individual commits if they cause any issues >> * Have 0 conflicts after the merge is done (just tested on my local >>branch) >> * Very close to vpc branch, so reduces the chances for regression bugs >> (vpc branch was tested) >> >> >> Cons: >> >> * harder to track down all VPC commits as they've been happening for a >> couple of months, and appear in the git log on the date they were >> originally submitted. >> >> >> >> 2) Do interactive rebase on master from vpc and squash all the commits >> into one >> > >Have you thought about rebasing the vpc branch against master? > >$ git checkout vpc >$ git rebase master >$ git checkout master >$ git merge vpc > >It will then come in as one clean commit if you squash it all during >your vpc rebase. > >I did that with RBD to make it one nice commit which could be reverted >if any issues would occur. > >Wido Yes, Wido, I've tried to do that as a part of 2) and ran into tons of conflicts. The conflicts are explained by the fact the changes were coming in to vpc branch for more than 2 months in a form of cherry picks along with daily merges with master branch, and rebase was never done on the vpc to master. I agree that for the future merges it might be the best way to go - interactive rebase and squash (may be into multiple commit groups per component basis). For this time, I'll probably go with 1) as the branch in this state was already tested. And of course, I'll run another round of feature testing + regression before pushing everything into repo. -Alena. > >> git checkout master; git rebase -i asf/vpc >> >> Pros: >> >> * Easier to see all VPC changes as they come as a part of a single >>commit >> >> Cons: >> >> * Requires rebase which leads to lots of conflicts. >> * Not sure if there is an option to revert an individual commit happened >> as a part of squashed commit. >> >> >> I personally prefer way 1) as it doesn't lead to conflicts + leaves less >> chances for regression bugs as the asf/master structure becomes very >> similar to the structure of asf/vpc branch, and this branch was tested. >> But let me know if anybody sees any downsides of doing the merge this >>way >> and can suggest a better way. >> >> >> Thank you, >> -Alena. >> >> >> >> On 8/10/12 10:33 AM, "Alena Prokharchyk" <alena.prokharc...@citrix.com> >> wrote: >> >>> >>> Considering that there is a code freeze happening soon, and as the main >>> part for VPC feature development is complete + the VPC code >>> has been well tested by cloudStack QA engineers, it's time to discuss >>> merging the asf/vpc branch into asf/master. At this point its too hard >>>to >>> break it to several components-to-merge as there were too many single >>> commits for the last couple of months from 7 developers involved in the >>> process + we were doing the daily merges with master branch in order to >>> keep asf/vpc branch up to date. It makes it hard to componentize at >>>this >>> point without running into lots of rebase conflicts >>> >>> So most likely it will be done as a single merge if >>> the merge gets approved by the community. We would highly >>> appreciate any help involving code/documentation review as well as any >>> other suggestions on how to make this particular merge safer and >>>smoother. >>> And the suggestions made towards the merge process in general, will >>>all be >>> addressed in the future. >>> >>> We've tried to keep vpc branch in sync with asf/master by performing >>>daily >>> merges from master branch to vpc, so there shouldn't be too many >>> conflicts. The VPC code went through several rounds of QA including the >>> regression testing for existing network components. Most of the VPC >>>code >>> resides in the new classes added as a part of the feature; but some >>> existing Network related code got some changes and can be considered >>>for >>> the review (NetworkManagerImpl/VirtualNetworkApplianceManager in >>> particular as well as some Agent APIs and cloudStack UI). The VPC FS >>> contains the reference to new classes (including client/agent APIs) >>> created for the VPC feature (see the link below). >>> >>> >>> Functional specification: >>> >>> >>>http://wiki.cloudstack.org/display/RelOps/Inter-VLAN+Routing+functional+ >>>sp >>> e >>> c >>> >>> >>> Engineer participated in VPC development: >>> >>> Sheng Yang - S2SVPN implementation >>> Alena Prokharchyk - Client/Agent APIs, >>> VpcManager/VpcVirtualNetworkApplianceManager/NetworkACLManager >>> Anthony Xu - All VirtualRuter script changes + feature implementation >>>for >>> Xen resource >>> Kelven Yang - Implementation for VmWare resource >>> Sonny Chhan/Jessica Wang/Brian Federle - UI component >>> >>> >>> QA test cases execution matrix: >>> >>> http://wiki.cloudstack.org/display/QA/Inter-VLAN+Test+Execution >>> >>> >>> I'm well aware the 4.0 will be branched on Monday, so once the merge is >>> approved, I can do the merge over the weekend including doing basic >>> regression for the networking key components on the merged branch. >>> >>> >>> -Alena. >>> >>> >> >> >
