alpass163 opened a new pull request, #19227:
URL: https://github.com/apache/iotdb/pull/19227
Adds three built-in relational functions.
### SAMPLE(DATA => table, N => n)
Table function that performs reservoir sampling over whole rows, with n
rows
per partition. The first n rows fill the reservoir; each following row
replaces
a uniformly chosen slot with probability n / (rows seen so far). All input
columns are kept, partition columns first, and the sampled rows are
emitted in
input order.
### DERIVATIVE(DATA => table PARTITION BY ... ORDER BY time, [unit =>
duration])
Table function that emits one output row per pair of adjacent points in
ORDER
BY time within each partition. The output time is the time of the second
point,
and each calculation column c outputs (c2 - c1) / (t2 - t1) * unit. `unit`
defaults to 1s in the current timestamp precision; a null on either side
only
makes that column null for the pair.
### INTEGRAL(value [, time_unit [, time]])
Aggregate function that computes the trapezoidal integral of `value` over
`time`, divided by `time_unit`. `time_unit` defaults to 1s and `time`
defaults
to the time column of the table. Ordered and naive accumulators are
provided for
both grouped and ungrouped aggregation; the naive ones support intermediate
states so partial results can be merged in distributed aggregation.
## Bug fixes
Fixes an existing planning bug for table functions that must run once per
partition (`mergeChildrenForSingleProcessor` in
`TableDistributedPlanGenerator`):
- When the processor had several children, they were merged with the
ordering
common to all children, or by plain concatenation when there was none.
The
merged stream was therefore not ordered by the GROUP clause (partition
keys
followed by the in-partition ordering), and rows of one partition could
arrive
non-consecutively when a partition spans several children, e.g.
PARTITION BY
on non-device columns or a device spanning regions. Such a partition was
then
treated as several partitions. The merged input now uses the ordering of
the
GroupNode and inserts an explicit SortNode when the children do not
already
provide it.
- `canSplitTableFunctionProcessor` only excluded FFT from per-child
splitting.
M4, FORECAST, PATTERN_MATCH, CLASSIFY, LOWPASS, HIGHPASS, XCORR, SESSION,
VARIATION and CAPACITY could be cloned for every child, so one logical
partition could be handled by several processors with separate state.
They are
now listed in `NON_SPLITTABLE_TABLE_FUNCTIONS`.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]