Pile.db Unistack object database engine

[itadapter]

• Store data documents (like Mongo)
• No rigid schema
• documents stored in named collections/spaces
• Indexes per space
• Indexes use B+tree
• Use pile as storage engine: in-mem or MMF
• Database is backed by pile. Can have many spaces+indexes
• System segment, stores collection and index metadata
• When collection dropped - record from sys is erased (instant action)
• Data and indexes are stored as pile records
• Each record has 2-4 byte metatag - a number in system table collection/index name
• Vacuum process goes through all records in pile and deallocates memory for metatags which do not exist
• Indexes: single/compound fields, conditional (only index if expression is met)
• Documents stored using BIX/Binary
• Object allocated with X% padding for in-place changing (table option)
• Query: API only. Use LINQ. Use index hash-joins for multiple indexes
• Query: handlers (stored procs) written in C#, deployed as assemblies on server (UNISTACK)
• Query: RPC-based proc invocation with NAME(params...): object pattern ala MVC

Example:

  using var db = PileDb.Mount("/home/sky/data/test-db");
//.....
  using var richest = db["customer"]
        ["idx_dob"]
        .GetRange(DateTime(1,1,1980), DateTime(1,1,1990))
        .Where(one => one.Married)
        .Max(one => one.Income);
 Console.WriteLine("The richest married customer is ${richest.FullName} making ${richest.Income}");                                

Despite what many would expect, the use of LINQ/extensions will work much faster on the server as that is what SQL does internally, only SQL does not know how to process data using primitives like indexes and open table scans, so it has to rely on statistics, whereas a custom written query in C# will outperform any "automatic" query as it is constructed around the specific database primitives (such as specific indexes) known to developer.