Status: New
Owner: ----
New issue 742 by [email protected]: gnt-cluster verify reports
ERROR:ENODEHOOKS when a node is offline
http://code.google.com/p/ganeti/issues/detail?id=742
What software version are you running? Please provide the output of "gnt-
cluster --version", "gnt-cluster version", and "hspace --version".
# gnt-cluster --version; gnt-cluster version; hspace --version
gnt-cluster (ganeti v2.10.0) 2.10.0
Software version: 2.10.0
Internode protocol: 2100000
Configuration format: 2100000
OS api version: 20
Export interface: 0
VCS version: (ganeti) version v2.10.0
hspace (ganeti) version v2.10.0
compiled with ghc 6.12
running on linux x86_64
What distribution are you using?
Debian squeeze.
What steps will reproduce the problem?
1. gnt-node modify -O yes $node
2. Ensure node is actually unreachable (ssh $node service ganeti stop
ganeti-noded is enough)
3. gnt-cluster verify --error-codes.
What is the expected output? What do you see instead?
Expected output: gnt-cluster verify passes with no errors related to the
node being offline.
Actual output:
Mon Mar 3 11:45:28 2014 * Other Notes
Mon Mar 3 11:45:28 2014 - NOTICE: 1 offline node(s) found.
Mon Mar 3 11:45:32 2014 - WARNING: Communication failure to node
$OFFLINENODE: Error 7: Failed connect to $OFFLINENODE:1811; Resource
temporarily unavailable
Mon Mar 3 11:45:32 2014 * Hooks Results
Mon Mar 3 11:45:32 2014 -
ERROR:ENODEHOOKS:node:$OFFLINENODE:Communication failure in hooks
execution: Error 7: Failed connect to $OFFLINENODE:1811; Resource
temporarily unavailable
Please provide any additional information below.
I tried to track down the problem in Ganeti to make sure it wasn't a
problem with our local setup. The cause seems to be that _CheckConfigNode
in lib/rpc.py does not do offline detection for nodes addressed by their
hostname instead of UUID, and it appears that all job hooks address nodes
by their hostnames (through the to_name lambda in HooksMaster.BuildFromLu
in lib/hooksmaster.py).
I've attached a diff containing a dirty workaround for the problem, which
fixes the immediate issue. I'm not sure if the proper fix should be to
switch from names to UUIDs for job hooks or changing the config API to be
able to look up by name again (or something else entirely) so I'll leave
that to you :).
Attachments:
gnt-cluster_verify_hacky_workaround.patch 847 bytes
--
You received this message because this project is configured to send all
issue notifications to this address.
You may adjust your notification preferences at:
https://code.google.com/hosting/settings
