Add a 'detectBroken' function that determines whether an instance is in an
unhealthy state, and what's needed to repair it. The repair is specified as
an AutoRepairType constant, and a list of opcodes. The opcodes will only be
executed (in following patches) if the repair type is allowed by the policy.
Signed-off-by: Dato Simó <d...@google.com>
---
src/Ganeti/HTools/Program/Harep.hs | 90 ++++++++++++++++++++++++++++++++++++--
1 file changed, 87 insertions(+), 3 deletions(-)
diff --git a/src/Ganeti/HTools/Program/Harep.hs
b/src/Ganeti/HTools/Program/Harep.hs
index c3cfadf..82ae63a 100644
--- a/src/Ganeti/HTools/Program/Harep.hs
+++ b/src/Ganeti/HTools/Program/Harep.hs
@@ -41,6 +41,7 @@ import Ganeti.Common
import Ganeti.Errors
import Ganeti.Jobs
import Ganeti.OpCodes
+import Ganeti.OpParams
import Ganeti.Types
import Ganeti.Utils
import qualified Ganeti.Constants as C
@@ -52,6 +53,7 @@ import Ganeti.HTools.ExtLoader
import Ganeti.HTools.Types
import qualified Ganeti.HTools.Container as Container
import qualified Ganeti.HTools.Instance as Instance
+import qualified Ganeti.HTools.Node as Node
-- | Options list and functions.
options :: IO [OptType]
@@ -265,6 +267,85 @@ commitChange client instData = do
return instData { tagsToRemove = [] }
+-- | Detect brokeness with an instance and suggest repair type and jobs to run.
+detectBroken :: Node.List -> Instance.Instance
+ -> Maybe (AutoRepairType, [OpCode])
+detectBroken nl inst =
+ let disk = Instance.diskTemplate inst
+ iname = Instance.name inst
+ offPri = Node.offline $ Container.find (Instance.pNode inst) nl
+ offSec = Node.offline $ Container.find (Instance.sNode inst) nl
+ in
+ case disk of
+ DTDrbd8
+ | offPri && offSec ->
+ Just (
+ ArReinstall,
+ [ OpInstanceRecreateDisks { opInstanceName = iname
+ , opRecreateDisksInfo = RecreateDisksAll
+ , opNodes = []
+ -- FIXME: there should be a better way
to
+ -- specify opcode paramteres than
abusing
+ -- mkNonEmpty in this way (using the
fact
+ -- that Maybe is used both for optional
+ -- fields, and to express failure).
+ , opIallocator = mkNonEmpty "hail"
+ }
+ , OpInstanceReinstall { opInstanceName = iname
+ , opOsType = Nothing
+ , opTempOsParams = Nothing
+ , opForceVariant = False
+ }
+ ])
+ | offPri ->
+ Just (
+ ArFailover,
+ [ OpInstanceFailover { opInstanceName = iname
+ -- FIXME: ditto, see above.
+ , opShutdownTimeout = fromJust $ mkNonNegative
+ C.defaultShutdownTimeout
+ , opIgnoreConsistency = False
+ , opTargetNode = Nothing
+ , opIgnoreIpolicy = False
+ , opIallocator = Nothing
+ }
+ ])
+ | offSec ->
+ Just (
+ ArFixStorage,
+ [ OpInstanceReplaceDisks { opInstanceName = iname
+ , opReplaceDisksMode = ReplaceNewSecondary
+ , opReplaceDisksList = []
+ , opRemoteNode = Nothing
+ -- FIXME: ditto, see above.
+ , opIallocator = mkNonEmpty "hail"
+ , opEarlyRelease = False
+ , opIgnoreIpolicy = False
+ }
+ ])
+ | otherwise -> Nothing
+
+ DTPlain
+ | offPri ->
+ Just (
+ ArReinstall,
+ [ OpInstanceRecreateDisks { opInstanceName = iname
+ , opRecreateDisksInfo = RecreateDisksAll
+ , opNodes = []
+ -- FIXME: ditto, see above.
+ , opIallocator = mkNonEmpty "hail"
+ }
+ , OpInstanceReinstall { opInstanceName = iname
+ , opOsType = Nothing
+ , opTempOsParams = Nothing
+ , opForceVariant = False
+ }
+ ])
+ | otherwise -> Nothing
+
+ _ -> Nothing -- Other cases are unimplemented for now: DTDiskless,
+ -- DTFile, DTSharedFile, DTBlock, DTRbd, DTExt.
+
-- | Main function.
main :: Options -> [String] -> IO ()
main opts args = do
@@ -276,13 +357,16 @@ main opts args = do
when (isNothing lsock) $
exitErr "Luxi socket (-L) required to execute jobs."
- (ClusterData _ _ il _ _) <- loadExternalData opts
+ (ClusterData _ nl il _ _) <- loadExternalData opts
let iniData = map setInitialState $ Container.elems il
master = fromJust lsock
-- First step: check all pending repairs, see if they are completed.
- _unused_iniData' <- bracket (L.getClient master) L.closeClient $
- forM iniData . processPending
+ iniData' <- bracket (L.getClient master) L.closeClient $
+ forM iniData . processPending
+
+ -- Second step: detect any problems.
+ let _unused_repairs = map (detectBroken nl . arInstance) iniData'
return ()
--
1.8.0.2-x20-1
