Support online-maintenance in hroller

Make hroller take into account the nodes (redundant) instances
will be migrated to. This be behavior can be overridden by the
--offline-maintenance option which will make hroller plan under
the assumption that all instances will be shutdown before starting
with the rolling reboots.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Thomas Thrainer <thomasth@google.com>

man/hroller.rst

@@ -51,11 +51,13 @@ ALGORITHM FOR CALCULATING OFFLINE REBOOT GROUPS
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 hroller will view the nodes as vertices of an undirected graph,
-connecting by instances which have both a primary and a secondary node.
-It will then color the graph using a few different heuristics, and
-return the minimum-size color set found. Node with the same color don't
-share an edge, and as such don't have an instance with both primary and
-secondary node on them, so they are safe to be rebooted concurrently.
+with two kind of edges. Firstly, there are edges from the primary
+to the secondary node of every instance. Secondly, two nodes are connected
+by an edge if they are the primary nodes of two instances that have the
+same secondary node. It will then color the graph using a few different
+heuristics, and return the minimum-size color set found. Node with the same
+color can then simultaneously migrate all instance off to their respective
+secondary nodes, and it is safe to reboot them simultaneously.
 
 OPTIONS
 -------
@@ -69,6 +71,11 @@ For a description of the standard options check **htools**\(7) and
 \--one-step-only
   Restrict to the first reboot group. Output the group one node per line.
 
+\--offline-maintenance
+  Pretend that all instances are shutdown before the reboots are carried
+  out. I.e., only edges from the primary to the secondary node of an instance
+  are considered.
+
 \--force
   Do not fail, even if the master node cannot be determined.
 
@@ -90,8 +97,8 @@ should support them both with and without secondary node replacement.
 EXAMPLE
 -------
 
-Offline Rolling node reboot output
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+Rolling node reboot output
+~~~~~~~~~~~~~~~~~~~~~~~~~~
 
 With the default options, the program shows one reboot group per line as
 a comma-separated list.

src/Ganeti/HTools/CLI.hs

@@ -70,6 +70,7 @@ module Ganeti.HTools.CLI
   , oNoSimulation
   , oNodeSim
   , oNodeTags
+  , oOfflineMaintenance
   , oOfflineNode
   , oOneStepOnly
   , oOutputDir
@@ -139,6 +140,7 @@ data Options = Options
   , optNodeSim     :: [String]       -- ^ Cluster simulation mode
   , optNodeTags    :: Maybe [String] -- ^ List of node tags to restrict to 
   , optOffline     :: [String]       -- ^ Names of offline nodes
+  , optOfflineMaintenance :: Bool    -- ^ Pretend all instances are offline
   , optOneStepOnly :: Bool           -- ^ Only do the first step
   , optOutPath     :: FilePath       -- ^ Path to the output directory
   , optSaveCluster :: Maybe FilePath -- ^ Save cluster state to this file
@@ -188,6 +190,7 @@ defaultOptions  = Options
   , optNodeSim     = []
   , optNodeTags    = Nothing
   , optOffline     = []
+  , optOfflineMaintenance = False
   , optOneStepOnly = False
   , optOutPath     = "."
   , optSaveCluster = Nothing
@@ -465,6 +468,14 @@ oNodeTags =
     "TAG,...") "Restrict to nodes with the given tags",
    OptComplString)
      
+oOfflineMaintenance :: OptType
+oOfflineMaintenance =
+  (Option "" ["offline-maintenance"]
+   (NoArg (\ opts -> Ok opts {optOfflineMaintenance = True}))
+   "Schedule offline maintenance, i.e., pretend that all instance are\
+   \ offline.",
+   OptComplNone)
+
 oOfflineNode :: OptType
 oOfflineNode =
   (Option "O" ["offline"]

src/Ganeti/HTools/Program/Hroller.hs

@@ -58,6 +58,7 @@ options = do
     , oDataFile
     , oIAllocSrc
     , oOfflineNode
+    , oOfflineMaintenance
     , oVerbose
     , oQuiet
     , oNoHeaders
@@ -143,10 +144,13 @@ main opts args = do
                              , hasTag $ optNodeTags opts
                              , hasGroup wantedGroup ])
               nlf
+      mkGraph = if optOfflineMaintenance opts
+                   then Node.mkNodeGraph
+                   else Node.mkRebootNodeGraph nlf
 
   -- TODO: fail if instances are running (with option to warn only)
 
-  nodeGraph <- case Node.mkNodeGraph nodes ilf of
+  nodeGraph <- case mkGraph nodes ilf of
                      Nothing -> exitErr "Cannot create node graph"
                      Just g -> return g