Infrastructure goals
Build / tree
- remove all the dead code that have not been updated in ages
- consider switching to autotools
- swith the tree to build with -MMD to detect header changes
- (DONE)
- automatic daily build
test suites
- we have tests laying around in several places but they are not used automatically
- evaluate if the option to revive those tests
- make sure that all new code has test cases associated?
release schedule
- we need to define and discuss a possible release schedule and possible deadlines in agreement with all the different parts we are working with.
- Red Hat would like to have cluster3 in Fedora 10.
Draft: http://sources.redhat.com/cluster/wiki/cluster3schedule
more coordination with packagers
- get in touch with all packagers of different distros
- make sure they are up to date with our new goals
- make them part of the process for producing test packages for our alpha milestones up to release
- give them an easy path to feed us back info/patches.
Feature goals
init/boot system
- we need to review the whole init system and interaction.
- too many complains and bits missing here and there.
- this is clearly a distro specific issue but we can help packages understand where it is best to start the stack etc.
init scripts
- it would be nice for the cman init script to allow more control of what it does via the command line
- e.g. "cman join" to do everthing up to cman_tool join, "cman join -e 1" for cman_tool join -e 1
- we need to define a specific, limited set of options that people really want to use, the join option above has already been requested
- this needs an owner who really has interest in using the init script
cluster config system
- make ccs one (of many) config plugin loaders.
- fabio and chrissie working on this
look at new "augeas" project: https://www.redhat.com/archives/et-mgmt-tools/2008-April/msg00133.html
common logging system
- started a long time ago, now openais has liblogsys that provides a well designed (lon, chrissie, sdake) logging system.
- cman has been converted already a long time ago.
- other subsystems should follow.
- all developers should switch as they start touching their code again.
removing groupd/libgroup
- fenced, dlm_controld, gfs_controld to use libcpg directly
gfs_controld and ocfs2_controld to do some similar things [ocfs2_controld uses libcpg already and has a different mount<->daemon protocol, but expects to share the plock code -- Joel]
- dave working on this
corosync split from openais
- steve dake working on this
- some new AIS services (MSG, LCK, AMF ?) may be completed
upgrading from cluster2 to cluster3
- by default we'll start up cluster2 infrastructure in F10 so existing clusters don't have to be taken off line
- when a cluster2 cluster is taken offline, something can be set in cluster.conf such that the next time the cluster starts up cluster3 infrastructure will come up
- dave working through different ideas on how to implement this
fencing
- add libfence with simple fence_node("nodename") api that will be used by fence_node, fenced, hopefully AMF, pacemaker?
rgmanager
- bugfixes only; efforts to integrate pacemaker with the other components should take priority
ocfs2
- we're working to have ocfs2 work as naturally with cluster3, and fs/dlm, as gfs
- ocfs2 kernel and tools branch stack-user works with cluster2+fs/dlm, wider exposure soon.
- ocfs2 needs recovery ordering with dlm_controld
- dave, joel working on this
gfs2
- will split plock handling out of lock_dlm and gfs_controld so that the same cluster plock code can be used by gfs, gfs2, ocfs2
- dave working on this