Hi guys, I know this topic as already been spoken many times, and I read a
lot of these discussions.
Yet, I have not been able to find a good way to do what I want.
We are receiving messages from our app that is a complex, dynamic, nested
JSON (can be a few or thousands of attributes). JSON is variable and can
contain nested arrays or sub-JSONs.
Please, consider this example:
JSON
{
"struct-id": 141241321,
"nested-1-1": {
"value-1-1-1": "36d1f74d-1663-418d-8b1b-665bbb2d9ecb",
"value-1-1-2": 5,
"value-1-1-3": 0.5,
"value-1-1-4": ["foo", "bar", "foobar"],
"nested-2-1": {
"test-2-1-1": "whatever",
"test-2-1-2": 42
}
},
"nested-1-2": {
"value-1-2-1": [{
"id": 1,
"deeply-nested": {
"data-1": "test",
"data-2": 4023
}
},
{
"id": 2,
"data-3": "that's enough data"
}]
}
}
We would like to store those messages to Cassandra and then run SPARK jobs
over it. Basically, storing it as a text (full JSON in one column) would
work but wouldn't be optimised since I might want to count how many times
"value-1-1-3" is bigger or equal to 1, I would have to read all the JSON
before answering this. I read a lot of things about people using composite
columns and dynamic composite columns, but no precise example. I am also
aware of collections support, yet nested collections are not supported
currently.
I would like to have:
- 1 column per attribute
- typed values
- something that would be able to parse and store any valid JSON (with
nested arrays of JSON or whatever).
- The most efficient model to use alongside with spark to query anything
inside.
What would be the possible CQL schemas to create such a data structure ?
What are the defaults of the following schema ?
Cassandra
CREATE TABLE test-schema (
struct-id int,
nested-1-1#value-1-1-1 string,
nested-1-1#value-1-1-2 int,
nested-1-1#value-1-1-3 float,
nested-1-1#value-1-1-4#array0 string,
nested-1-1#value-1-1-4#array1 string,
nested-1-1#value-1-1-4#array2 string,
nested-1-1#nested-2-1#test-2-1-1 string,
nested-1-1#nested-2-1#test-2-1-2 int,
nested-1-2#value-1-2-1#array0#id int,
nested-1-2#value-1-2-1#array0#deeply-nested#data-1 string,
nested-1-2#value-1-2-1#array0#deeply-nested#data-2 int,
nested-1-2#id int,
nested-1-2#data-3 string,
PRIMARY KEY (struct-id)
)
I could use:
nested-1-1#value-1-1-4 list<string>,
instead of:
nested-1-1#value-1-1-4#array0 string,
nested-1-1#value-1-1-4#array1 string,
nested-1-1#value-1-1-4#array2 string,
yet it wouldn't work here:
nested-1-2#value-1-2-1#array0#deeply-nested#data-1 string,
nested-1-2#value-1-2-1#array0#deeply-nested#data-2 int,
nested-1-2#value-1-2-1#array1#id int,
nested-1-2#value-1-2-1#array1#data-3 string,
since this is a nested structure inside the list.
To create this schema, could we imagine that the app logging this try to
write to the corresponding column, for each JSON attribute, and if the
column is missing, catch the error, create the column and reprocess write ?
This exception would happen for each new field, only once and would modify
the schema.
Any thought that would help us (and probably more people) ?
Alain
