The interpreter module includes example code of how to use the module. The output of this command can be seen by invoking the module as below:
$ python -m crushha.interpreter <Bucket(node1, node, )> <Bucket(node4, node, )> <Bucket(node2, node, )> <Bucket(node3, node, )> <Bucket(node5, node, )> <Bucket(node6, node, )> Could not satisfy constraints of rule, erroring on rule 3 - use(bucket=<Bucket(sydney, region, ['dc1', 'dc2'])>) - select(type='datacenter', count=2) => take(count=1) - emit() - use(bucket=<Bucket(default, root, ['sydney', 'melbourne'])>) - take(count=1) - emit()
As you can see in the above, the code prints out 6 nodes and then indicates an issue with the rules. In this case there are only 6 nodes in the Buckets in the CRUSH map and hence no more are available to take when the 'take' rule is encountered.
A full CRUSH map is made of 2 parts, of which one is partially visible in the above example (The CRUSH hierarchy of failure domains) and the other is printed out in full (The ruleset that operates on this hierarchy).
The failure domain hierarchy is a tree structures that can have multiple roots.
That defines the 'failure domains' and the relationship between them.
does not attempt to enforce specific domains however examples of classic
failure domains are 'region', 'datacenter', 'rack', 'pdu', 'chassis' and host.
By defining the relationship rules can be created to prevent particular types
of scenarios. The ruleset in the example above is intended to select 2 machines
in Sydney and another offsite datacenter to prevent against failure of a
datacenter and subsequent data loss.
Diving into the Ruleset above further, the ruleset operate at the level of datecenters as defined by the select and use rules. as no further selection of nodes lower in the hierarchy are used, the take command is free to select from any machine below those points in the CRUSH failure domain hierarchy
Select statements are generally intended to be 'wide' or provide left/right transversal of a tree. as failure domains are intended to be ordered, select statements provide this horizontal navigation inside a type of failure domain.
Use can be used to restrict the scope of this left/right movement of select and is useful for ensuring a specific property of your intended availability is enforced, eg selecting a close by data center. This is best used with multiple roots as tree with one 'default' root may become constraining in expressing these other desired properties.
Take is a 'vertical' operation intended to select destination nodes. It will place a node on the stack as many times as requested for each previously selected item. The combination of select and take has a n*m relationship where n is the amount of selected nodes and m is the amount of nodes take is requested to return
Finally Emit returns these nodes to the caller and resets the state of the VM. There is an implicit Emit at the end of all rulesets that gets added by the interpreter and as such the last Emit statement may be omitted. Emit is best used as a 'barrier' due to its 'reset the world' nature, it can be used to separate explicit requirements of your HA plan. In the above example it is used as a barrier to select 2 nodes locally, then once the state is reset, select a remaining offsite DC. 2 separate and isolated concerns that are best dealt with separately. This also has the effect of ordering meaning that nodes at the local DC are emitted first, then once a critical point is reached (3 nodes in the above example) off site redundancy is achieved.