Change Details

Consider an "A then B" ordering arising in three ways: a symmetric ordering constraint, an asymmetric ordering constraint, and implicit group ordering. Now, consider that A is stopped (and possibly disabled with target-role=Stopped) and B is unmanaged (both the case where it's started and the case where it's stopped). In the symmetric and group cases, if B is started, A should be prevented from starting, since B should be stopped (but can't because it's unmanaged) before A is started. Currently, A is not prevented from starting. In all cases, if B is stopped, A should be allowed to start. This currently works.

Consider an "A then B" ordering arising in three ways: a symmetric ordering constraint, an asymmetric ordering constraint, and implicit group ordering. Now, consider that A is stopped (and possibly disabled with target-role=Stopped) and B is unmanaged (both the case where it's started and the case where it's stopped). In the symmetric and group cases, if B is started, A should be prevented from starting, since B should be stopped (but can't because it's unmanaged) before A is started. Currently, A is not prevented from starting. In all cases, if B is stopped, A should be allowed to start. This currently works. Very preliminary work was started on this: * [[https://github.com/kgaillot/pacemaker/tree/T956]] The commit adding a regression test is the most useful. The WIP commits help remove redundant orderings, but the graphs of blocked start actions will not show proper relationships. The effect on the unfence-definition and unfence-parameters tests may not be correct (probe vs stop ordering is still correct only because stop -> unfence -> probe).

Consider an "A then B" ordering arising in three ways: a symmetric ordering constraint, an asymmetric ordering constraint, and implicit group ordering. Now, consider that A is stopped (and possibly disabled with target-role=Stopped) and B is unmanaged (both the case where it's started and the case where it's stopped). In the symmetric and group cases, if B is started, A should be prevented from starting, since B should be stopped (but can't because it's unmanaged) before A is started. Currently, A is not prevented from starting. In all cases, if B is stopped, A should be allowed to start. This currently works. Very preliminary work was started on this: * [[https://github.com/kgaillot/pacemaker/tree/T956]] The commit adding a regression test is the most useful. The WIP commits help remove redundant orderings, but the graphs of blocked start actions will not show proper relationships. The effect on the unfence-definition and unfence-parameters tests may not be correct (probe vs stop ordering is still correct only because stop -> unfence -> probe).