theses

[snuyanzin]

1. DataSource could be used both as source and sink
2. MetaData should be in code (version control)
3. DataBlock should be immutable
4. DataBlock could be idenfined by DataBlockURL (and version ?)
5. Dimension is a type of limitation which could be used for partitioning (may be refrase?)
6. Invariant is a type of filtration before final results
7. DataBlockURL contains information
   • jdbc string in case of jdbc
   • file path in case of file system
   • query to perform on DataSource
   • ???
8. DataBlock could be reused if there is one more query for data with overlapping datetime intervals e.g. in case of success search by DatablockURL
9. If no DataBlock found in local cache then it should be built and put in the cache after successful build
10. In case of concurrent demands to the same DataBlock which is not currently available in local cache the only one request (thread, process, whatever) should build the Datablock while others should wait
11. Set of rules to convert various types between DataSource's presentations, rules to convert various types of logical relational expressions to corresponding relational expresseions of datasource's convention.
12. Query might finished with success only if they ask for data in the past
    If there is a query for datetime range in future then it should be failed or scheduled for the time when it comes

possible improvements:

[snuyanzin]