[
https://issues.apache.org/jira/browse/DRILL-31?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Timothy Chen updated DRILL-31:
------------------------------
Description:
For each record of an incoming stream (let’s call this the target record), the
window operator will create a segment containing the records in a sliding
window surrounding the target record. The sliding window will include a
specified number of records before and/or after the target record. The window
frame operator can operate across an entire input stream or can be told to only
allow operations within each incoming segment.
A simple example: a window range of 2 before and 0 after is applied to a 5
record single segment input [0,1,2,3,4]. This would result in an output of 12
rows (brackets describe each output segment) [0] [0,1] [0,1,2] [1,2,3]
[2,3,4]. Each output record would have contain two additional fields:
ref.segment which holds the current window segment and ref.position which will
the negative, positive or zero position within the window. For the example
data, the values to each of these would be: ref.segment:
[0,1,1,2,2,2,3,3,3,4,4,4], ref.position: [0,-1,0,-1,0,1,-1,0,1,-1,0,1]
A window operator can also optionally take a segment key as input. This key is
used to restrict the windows so that they do not cross segment boundaries. The
window operator takes as inputs a ‘start’ and an ‘end’ value. These values
define the range of the window. The window range is defined based on thinking
of the target record as zero, and each record after incrementing positively and
each record incrementing negatively. In the case that start or end is not
defined, that portion of the range will be unbounded. In all cases, start must
be greater than or equal to end. At least one of start or end must be defined.
{ @id†: <opref>, op: “windowframe”,
input†: <input>,
within*: <name>,
start*: <number>,
end*: <number>
ref: {
segment: <name>,
position: <name>
},
}
> Implement WindowsFrame Reference Operator
> -----------------------------------------
>
> Key: DRILL-31
> URL: https://issues.apache.org/jira/browse/DRILL-31
> Project: Apache Drill
> Issue Type: New Feature
> Reporter: Timothy Chen
> Assignee: Timothy Chen
> Labels: logical
>
> For each record of an incoming stream (let’s call this the target record),
> the window operator will create a segment containing the records in a sliding
> window surrounding the target record. The sliding window will include a
> specified number of records before and/or after the target record. The
> window frame operator can operate across an entire input stream or can be
> told to only allow operations within each incoming segment.
> A simple example: a window range of 2 before and 0 after is applied to a 5
> record single segment input [0,1,2,3,4]. This would result in an output of
> 12 rows (brackets describe each output segment) [0] [0,1] [0,1,2] [1,2,3]
> [2,3,4]. Each output record would have contain two additional fields:
> ref.segment which holds the current window segment and ref.position which
> will the negative, positive or zero position within the window. For the
> example data, the values to each of these would be: ref.segment:
> [0,1,1,2,2,2,3,3,3,4,4,4], ref.position: [0,-1,0,-1,0,1,-1,0,1,-1,0,1]
> A window operator can also optionally take a segment key as input. This key
> is used to restrict the windows so that they do not cross segment boundaries.
> The window operator takes as inputs a ‘start’ and an ‘end’ value. These
> values define the range of the window. The window range is defined based on
> thinking of the target record as zero, and each record after incrementing
> positively and each record incrementing negatively. In the case that start
> or end is not defined, that portion of the range will be unbounded. In all
> cases, start must be greater than or equal to end. At least one of start or
> end must be defined.
> { @id†: <opref>, op: “windowframe”,
> input†: <input>,
> within*: <name>,
> start*: <number>,
> end*: <number>
> ref: {
> segment: <name>,
> position: <name>
> },
> }
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
