hail - Ganeti IAllocator plugin


hail [ -t file | --simulate spec ] [options…] input-file

hail --version


hail is a Ganeti IAllocator plugin that implements the instance placement and movement using the same algorithm as hbal(1).

The program takes input via a JSON-file containing current cluster state and the request details, and output (on stdout) a JSON-formatted response. In case of critical failures, the error message is printed on stderr and the exit code is changed to show failure.

If the input file name is - (a single minus sign), then the request data will be read from stdin.

Apart from input data, hail collects data over the network from all MonDs with the –mond option. Currently it uses only data produced by the CPUload collector.


On regular node groups, the program uses a simplified version of the hbal algorithm; for allocation on node groups with exclusive storage see below.

For single-node allocations (non-mirrored instances), again we select the node which, when chosen as the primary node, gives the best score.

For dual-node allocations (mirrored instances), we chose the best pair; this is the only choice where the algorithm is non-trivial with regard to cluster size.

For relocations, we try to change the secondary node of the instance to all the valid other nodes; the node which results in the best cluster score is chosen.

For node changes (change-node mode), we currently support DRBD instances only, and all three modes (primary changes, secondary changes and all node changes).

For group moves (change-group mode), again only DRBD is supported, and we compute the correct sequence that will result in a group change; job failure mid-way will result in a split instance. The choice of node(s) on the target group is based on the group score, and the choice of group is based on the same algorithm as allocations (group with lowest score after placement).

The deprecated multi-evacuate modes is no longer supported.

In all cases, the cluster (or group) scoring is identical to the hbal algorithm.

For allocation on node groups with exclusive storage, the lost-allocations metrics is used instead to determine which node to allocate an instance on. For a node the allocation vector is the vector of, for each instance policy interval in decreasing order, the number of instances minimally compliant with that interval that still can be placed on that node. The lost-allocations vector for an instance on a node is the difference of the allocation vectors for that node before and after placing the instance on that node. The lost-allocations metrics is the lost allocation vector followed by the remaining disk space on the chosen node, all compared lexicographically.


The options that can be passed to the program are as follows:

-p, --print-nodes

Prints the before and after node status, in a format designed to allow the user to understand the node’s most important parameters. See the man page htools(1) for more details about this option.

-t datafile, --text-data=*datafile*

The name of the file holding cluster information, to override the data in the JSON request itself. This is mostly used for debugging. The format of the file is described in the man page htools(1).


If given the program will query all MonDs to fetch data from the supported data collectors over the network.

--mond-data datafile

The name of the file holding the data provided by MonD, to override querying MonDs over the network. This is mostly used for debugging. The file must be in JSON format and present an array of JSON objects , one for every node, with two members. The first member named node is the name of the node and the second member named reports is an array of report objects. The report objects must be in the same format as produced by the monitoring agent.


If given, all dynamic utilisation information will be ignored by assuming it to be 0. This option will take precedence over any data passed by the MonDs with the --mond and the --mond-data option.


If given, all checks for soft errors will be omitted when searching for possible allocations. In this way a useful decision can be made even in overloaded clusters.


Normally, hail will only consider those allocations where all instances of a node can immediately restarted should that node fail. With this option given, hail will check only N+1 redundancy for DRBD instances.


Only consider alloctions on the specified nodes. This overrides any restrictions given in the allocation request.

--simulate description

Backend specification: similar to the -t option, this allows overriding the cluster data with a simulated cluster. For details about the description, see the man page htools(1).

-S filename, --save-cluster=*filename*

If given, the state of the cluster before and the iallocator run is saved to a file named filename.pre-ialloc, respectively filename.post-ialloc. This allows re-feeding the cluster state to any of the htools utilities via the -t option.


This option increases verbosity and can be used for debugging in order to understand how the IAllocator request is parsed; it can be passed multiple times for successively more information.


For the tag-exclusion configuration (see the manpage of hbal for more details), the list of which instance tags to consider as exclusion tags will be read from the cluster tags, configured as follows:

  • get all cluster tags starting with htools:iextags:

  • use their suffix as the prefix for exclusion tags

For example, given a cluster tag like htools:iextags:service, all instance tags of the form service:X will be considered as exclusion tags, meaning that (e.g.) two instances which both have a tag service:foo will not be placed on the same primary node.


The options that can be passed to the program are as follows:


The exist status of the command will be zero, unless for some reason the algorithm fatally failed (e.g. wrong node or instance data).


Networks (as configured by gnt-network(8)) are not taken into account in Ganeti 2.7. The only way to guarantee that they work correctly is having your networks connected to all nodegroups. This will be fixed in a future version.