---

CQL query plan and execution

---

• Query Plan
  • A query in CQL is always converted into a query plan

    ---

  • Query plan composition:

    A query plan is composed of: operators (which perform the actual processing) queues (that buffer the tuples) synopses (which store the operator state)

    ---

  • Query plan operation:

    Each query plan reads from one or more input queues, processes the input based on the operators and writes any output to an output queue

    ---

  • An operator may create sypnoses of the input data for its need...

    Example: when a join is performed on 2 input streams, the operator may create 2 sypnosis (indices) from the tuples in the window

  ---

  ---

• Elements
  • Recall that there are two fundamental data types in CQL: stream and relation

  • The two fundamental data types are unified as sequences of time stamped tuples

  • In addition, each tuple is flagged as either:

    insertion (+), or deleteion (-)

    ---

  • An element in CQL is a triplet tuple-timestamp-flag

    ---

  • Streams only contain insertion (+) elements

  • Relations can contain insertion (+) and/or deletion (-) elements

  ---

  ---

• Operators
  • Here's a list of the operators of STREAM taken from the paper:

  ---

  ---

• Queues
  • A queue is a buffer space that connects a producer operator to a consumer operator:

    ---

  • An important requirement in CQL for queue is that

    elements inserted in queues always have non-decreasing timestamps

    (Mechanism to generate timestamps are discussed in another paper and will not be discussed here.)

  ---

  ---

• Synopis
  • A synopsis is a summary data structure on an input stream

  • They are constructed by some operators and are transparent to the user...

  ---

  ---

• Query Plan construction
  • A query plan is a tree:

    The nodes in the tree are the operators The edges of the query plan tree are queues (Synopsis are "attached" to operators as needed)

    ---

  • Example query plan construction:

    SELECT * FROM S1 [Rows 1000], S2 [Range 2 Minutes] WHERE S1.A = S2.A AND S1.A > 10

    ---

  • Resulting query plan:

    ---

  • Notes:

    For optimization purposes, the SELECT S.A > 10 can be pushed down... The [Rows 1000] and Rows 2 Minute] are operators These "sequence window" operators creates synopsis that contains the content of the current window. This information is necessary to generate deletion (-) tuples (when needed)

  ---

  ---

• Summary
  • STREAM takes a language oriented approach

  • Aurora takes a GUI oriented approach

  • Both system processes stream queries in a similar manner (pipelined processing)