Ganeti Bulk Create¶
Current state and shortcomings¶
Creation of instances happens a lot. A fair load is done by just creating instances and due to bad allocation shifting them around later again. Additionally, if you turn up a new cluster you already know a bunch of instances, which need to exists on the cluster. Doing this one-by-one is not only cumbersome but might also fail, due to lack of resources or lead to badly balanced clusters.
Since the early Ganeti 2.0 alpha version there is a
batch-create command to allocate a bunch of instances based on a json
file. This feature, however, doesn’t take any advantages of iallocator
and submits jobs in a serialized manner.
To overcome this shortcoming we would extend the current iallocator interface to allow bulk requests. On the Ganeti side, a new opcode is introduced to handle the bulk creation and returning the resulting placement from the IAllocator.
Due to the design of chained jobs, we can guarantee, that with the state
at which the
multi-alloc opcode is run, all of the instances will
fit (or all won’t). But we can’t guarantee that once the instance
creation requests were submit, no other jobs have sneaked in between.
This might still lead to failing jobs because the resources have changed
in the meantime.
A new additional
type will be added called
distinguish between normal and bulk operation. For the bulk operation
request will be a finite list of request dicts.
multi-allocate is declared,
request must exist and is a list
request dicts as described in Operation specific input. The
result then is a list of instance name and node
placements in the order of the
In addition, the old
allocate request type will be deprecated and at
latest in Ganeti 2.8 incorporated into this new request. Current code
will need slight adaption to work with the new request. This needs
We add a new opcode
OpInstanceBulkAdd. It receives a list of
OpInstanceCreate on the
instances field. This is done to make
sure, that these two loosely coupled opcodes do not get out of sync. On
the RAPI side, however, this just is a list of instance create
definitions. And the client is adapted accordingly.
The opcode itself does some sanity checks on the instance creation opcodes which includes:
modeis not set
snodesis not set
iallocatoris not set
Any of the above error will be aborted with
OpPrereqError. Once the
list has been verified it is handed to the
iallocator as described
in IAllocator. Upon success we then return the result of the
At this point the current instance allocation would work with the
resources available on the cluster as perceived upon
OpInstanceBulkAdd invocation. However, there might be corner cases
where this is not true as described in Problems.