We reorganized the functions in flux to have the structure:
/functions
/inputs
/transformations
/outputs
this PR catches up platform to work with the new package layout.
As a separate refactoring issue, we should discuss:
from(bucket: ) should migrate from flux --> platform
to_http and to_kafka should migrate from platform --> flux
It creates a simple client that follows the QueryService interface and
uses the `influxql.Compiler` type to determine where the query should be
routed to and to return the query as a `flux.ResultIterator`.
This will be useful for replaying transpiler queries against influxdb
1.x servers to verify correctness.
The transpiler will normalize the `_time` column by dropping any
existing time column and then duplicating `_start` when the query is an
aggregate type.
This works for the selectors because they did not normalize their
`_time` column at all and, while the aggregates did normalize their
`_time` column, we have made the decision to remove that functionality
and have aggregates not set a `_time` column at all.
Instead of generating multiple cursors, a pivot is used to join fields
within the same series.
This should be easier than generating a new cursor for everything.
WIP: saving state reword
feat(query/influxql): implement query.ResultIterator directly on response
review(influxql): add ResponseIterator that implement query.ResultIterator
review(influxql): update code is response to review
This commit provides a new join implementation that can join
tables with different group keys. It also implements a strategy
for evicting stale data from the join operation's cache when
it can be determined that a table will no longer be joined.
Currently early table eviction from the buffers of the join occurs
when the left-most group key column of both input streams are equal
and that column is part of the join key.
Moves idpe.QueryService into platform/query.ProxyQueryService
Splits the Request into ProxyRequest and Request.
Changes query.QueryService and query.AsyncQueryService to use a Request
type. This means that the Compiler interface is consumed by the service
to abstract out transpilation vs Flux compilation vs raw spec.
The transpiler handler is removed.
There are separate http handlers and service implementations for each of
the three query services.
Query logging types are moved into platform.
The ResultIterator now expects Cancel to always be called.
The fluxd binary exposes the query endpoint specified in the swagger
file.
The now time is stamped by the influxql transpiler and used inside of
the actual query. It will result in more accuracy if we take the
timestamp we have created and send it as part of the spec to queryd
rather than force ourselves to ensure absolute times exist everywhere.
previous versions only supported the first parameter to time() that set the window size. This version supports the second parameter, which shifts the offset a fixed amount from the epoch
* feat(query/influxql): add regex support to transpiler
also added test case to querytest
* Add raw_with_condition test to transpiler unit tests
* add unit tests for regex conditions on raw query
The package contains all of the transpiler specs and allows them to be
put into different files instead of keeping all of the tests in the same
file. They are all Go code so they are type checked rather than being
loaded as JSON from disk.
Additionally, to make it easier for a developer, the tests will report
the exact file and line where the test was created. So rather than
hunting for which file a test is located in, you will get something nice
like the following:
--- FAIL: TestTranspiler/SELECT_count(value)_FROM_db0..cpu_WHERE_host_=_'server01' (0.00s)
testing.go:51: aggregates_with_condition.go:16: unexpected error: unimplemented function: "count"
As can be seen, the test that failed can be found in the
`aggregates_with_condition.go` file at line 16 which is where the test
was created by the `AggregateTest` function and the relevant spec can be
found in that same file.
The transpiler compilation tests will now not allow skip to be
specified. Instead, it must return an error message that starts with
`unimplemented` and then the reason will be used as the skip message.
This way, it will be easier to identify the failing tests in the
transpiler. In the previous method, it was possible for a test to be
marked as skip, but for the transpiler to return the wrong error message
because the test did not differentiate between an unimplemented error
message and an incorrect error message.
The transpiler now supports basic windowing. The window offsets are not
supported yet at all.
For windowing, we use the window function to split the points, perform
the aggregate/selector operation, and then we put them back into the
same window so they are within the same table as they originally were
located in. This is now reflected in the spec and the code.
The compiler tests from the github.com/influxdata/influxdb/query have
been moved over to the influxql transpiler in platform. The framework
has updated to include a skip option so that all of the tests can be
there, but not all of them have to succeed at the moment. If a test
starts succeeding but is marked as one that should be skipped, it will
also cause an error to prevent us from doing work on the transpiler
without marking the test as something not to skip anymore (so progress
is always made).
This extra flexibility makes it easier for the transpiler to generate a
specification since the map step can be focused on only generating the
columns related to fields. In particular, it makes it easier to
implement wildcards for tags because the tags will get passed along with
the partition key.
The spec says to use the `_time` column for the time in the output, but
we were mapping `r._time` to `time` and using the `time` variable. This
modifies the encoder to use the `_time` column and rename it to `time`
for the column name.
The transpiler should use a bucket for the `from()` call instead of the
database parameter which will likely be deprecated. The bucket that it
will read data from is `db/rp` and, if the retention policy isn't
specified, `autogen` will be used as the default.
There are a few changes to how the transpiler works. The first is that
the streams are now abstracted behind a `cursor` interface. The
interface keeps track of which AST nodes (like variables or function
calls) are represented by the data inside of the stream and the method
of how to access the underlying data. This makes it easier to make a
generic interface for things like the join and map operations. This also
makes it easier to, in the future, use the same code from the map
operation for a filter so we can implement conditions.
This also follows the transpiler readme's methods and takes advantage of
the updates to the ifql language. This means it will group the relevant
cursors into a cursor group, perform any necessary joins, and allow us
to continue building on this as we flesh out more parts of the
transpiler and the language.
The cursor interface makes it so we no longer have to keep a symbol
table mapping the generated names to the locations because that is all
kept within the incoming cursor rather than as a separate data
structure.
It also splits the transpiler into more files so it is easier to find
the relevant code for each stage of the transpiler.
Update the transpiler docs to account for a change in the query
language. A slightly different method is now used with some
clarifications previously existing problems.
The readme is also indexed with a table of contents for easier reading.
This fixes the encoder so that it will encode the response correctly to
a JSON blob using the outputs of the transpiler. The transpiler has also
been modified to pass through the correct values so the map function is
correctly constructed and the aggregate function is also correctly
constructed.
This removes the group function temporarily because it does not seem to
be working.
This modifies the `MultiResultDecoder` interface to accept an
`io.ReadCloser` so the `ResultIterator` can close the `io.Reader`
instead of doing it through a defer call. It then makes it so the
`Cancel()` method will close the reader or the reader will be
automatically closed when `More()` returns false.
In order to facilitate this change the query.PartitionKey
interface was changed to use the values.Value interface.
Additionally map was previously broken when it needed to repartition.
Tests have been added.
The query service now handles some errors (not all) from the ifqld
service. This way, the transpiler, which uses the query service, can
report when it encounters an error from ifqld.
The transpiler will now correctly return errors in the JSON response.
The transpiler will now yield each statement using the statement id so
the result encoder can properly order the results and encode the
statement id. This behavior is now in the transpiler spec.
Adam
Jonathan A. Sternberg
jlapacik
Nathaniel Cook
Mark Rushakoff
Chris Goller
Leonardo Di Donato
Stuart Carnie
Michael Desa
Adam Perlin
Adam