sparklyr (development version)
Sparklyr 1.7.5
Misc
Addresses deprecation of
rlang::is_env()
function. (@lionel- #3217)Updates
pivot_wider()
to support new version oftidyr
(@DavisVaughan #3215)
Sparklyr 1.7.4
Misc
- Edgar Ruiz (https://github.com/edgararuiz) will be the new maintainer of {sparklyr} moving forward.
Sparklyr 1.7.3
Data
Implemented support for the
.groups
parameter fordplyr::summarize()
operations on Spark dataframesFixed the incorrect handling of the
remove = TRUE
option forseparate.tbl_spark()
Optimized away an extra count query when collecting Spark dataframes from Spark to R.
Misc
By default, use links from the https://dlcdn.apache.org site for downloading Apache Spark when possible.
Attempt to continue
spark_install()
process even if the Spark version specified is not present ininst/extdata/versions*.json
files (in which casesparklyr
will guess the URL of the tar ball based on the existing and well-known naming convention used by https://archive.apache.org, i.e., https://archive.apache.org/dist/spark/spark-\({spark version}/spark-\){spark version}-bin-hadoop${hadoop version}.tgz)Revised
inst/extdata/versions*.json
files to reflect recent releases of Apache Spark.Implemented
sparklyr_get_backend_port()
for querying the port number used by thesparklyr
backend.
Sparklyr 1.7.2
Connections
Added support for notebook-scoped libraries on Databricks connections. R library tree paths (i.e., those returned from
.libPaths()
) are now shared between driver and worker in sparklyr for Databricks connection use cases.Java version validation function of
sparklyr
was revised to be able to parsejava -version
outputs containing only major version or outputs containing data values.Spark configuration logic was revised to ensure “sparklyr.cores.local” takes precedence over “sparklyr.connect.cores.local”, as the latter is deprecated.
Renamed “sparklyr.backend.threads” (an undocumented, non-user-facing,
sparklyr
internal-only configuration) to “spark.sparklyr-backend.threads” so that it has the required “spark.” prefix and is configurable throughsparklyr::spark_config()
.For Spark 2.0 or above, if
org.apache.spark.SparkEnv.get()
returns a non- null env object, thensparklyr
will use that env object to configure “spark.sparklyr-backend.threads”.Support for running custom callbacks before the
sparklyr
backend starts processing JVM method calls was added for Databricks-related use cases, which will be useful for implementing ADL credential pass-through.
Data
Revised
spark_write_delta()
to usedelta.io
library version 1.0 when working with Apache Spark 3.1 or above.Fixed a problem with
dbplyr::remote_name()
returningNULL
on Spark dataframes returned from adplyr::arrange()
operation followed bydplyr::compute()
(e.g.,<a spark_dataframe> %>% arrange(<some column>) %>% compute()
).Implemented
tidyr::replace_na()
interface for Spark dataframes.The
n_distinct()
summarizer for Spark dataframes was revised substantially to properly supportna.rm = TRUE
orna.rm = FALSE
use cases when performingdplyr::summarize(<colname> = n_distinct(...))
types of operations on Spark dataframes.Spark data interface functions that create Spark dataframes will no longer check whether any Spark dataframe with identical name exists when the dataframe being created has a randomly generated name (as randomly generated table name will contain a UUID and any chance of name collision is vanishingly small).
Documentation
- Create usage example for
ml_prefixspan()
.
Sparklyr 1.7.1
Connections
- Fixed an issue with connecting to Apache Spark 3.1 or above.
Sparklyr 1.7.0
Data
Revised
tidyr::fill()
implementation to respect any ‘ORDER BY’ clause from the input while ensuring the same ‘ORDER BY’ operation is never duplicated twice in the generated Spark SQL queryHelper functions such as
sdf_rbeta()
,sdf_rbinom()
, etc were implemented for generating Spark dataframes containing i.i.d. samples from commonly used probability distributions.Fixed a bug with
compute.tbl_spark()
’s handling of positional args.Fixed a bug that previously affected
dplyr::tbl()
when the source table is specified usingdbplyr::in_schema()
.Internal calls to
sdf_schema.tbl_spark()
andspark_dataframe.tbl_spark()
are memoized to reduce performance overhead from repeatedspark_invoke()
s.spark_read_image()
was implemented to support image files as data sources.spark_read_binary()
was implemented to support binary data sources.A specialized version of
tbl_ptype()
was implemented so that no data will be collected from Spark to R whendplyr
callstbl_ptype()
on a Spark dataframe.Added support for
database
parameter tosrc_tbls.spark_connection()
(e.g.,src_tbls(sc, database = "default")
wheresc
is a Spark connection).Fixed a null pointer issue with
spark_read_jdbc()
andspark_write_jdbc()
.
Distributed R
spark_apply()
was improved to supporttibble
inputs containing list columns.Spark dataframes created by
spark_apply()
will be cached by default to avoid re-computations.spark_apply()
anddo_spark()
now supportqs
and custom serializations.The experimental
auto_deps = TRUE
mode was implemented forspark_apply()
to infer required R packages for the closure, and to only copy required R packages to Spark worker nodes when executing the closure.
Extensions
Sparklyr extensions can now customize dbplyr SQL translator env used by
sparklyr
by supplying their own dbplyr SQL variant when callingspark_dependency()
(see https://github.com/r-spark/sparklyr.sedona/blob/1455d3dea51ad16114a8112f2990ec542458aee2/R/dependencies.R#L38 for an example).jarray()
was implemented to convert a R vector into anArray[T]
reference. A reference returned byjarray()
can be passed toinvoke*
family of functions requiring anArray[T]
as a parameter where T is some type that is more specific thanjava.lang.Object
.jfloat()
function was implemented to cast any numeric type in R tojava.lang.Float
.jfloat_array()
was implemented to instantiateArray[java.lang.Float]
from numeric values in R.
Serialization
Added null checks that were previously missing when collecting array columns from Spark dataframe to R.
array<byte>
andarray<boolean>
columns in a Spark dataframe will be collected asraw()
andlogical()
vectors, respectively, in R rather than integer arrays.Fixed a bug that previously caused invoke params containing
NaN
s to be serialized incorrectly.
Spark ML
ml_compute_silhouette_measure()
was implemented to evaluate the Silhouette measure of k-mean clustering results.spark_read_libsvm()
now supports specifications of additional options via theoptions
parameter. Additional libsvm data source options currently supported by Spark includenumFeatures
andvectorType
(see https://spark.apache.org/docs/latest/api/java/org/apache/spark/ml/source/libsvm/LibSVMDataSource.html).ml_linear_svc()
will emit a warning ifweight_col
is specified while working with Spark 3.0 or above, as it is no longer supported in recent versions of Spark.Fixed an issue with
ft_one_hot_encoder.ml_pipeline()
not working as expected.
Sparklyr 1.6.3
Data
Reduced the number of
invoke()
calls needed forsdf_schema()
to avoid performance issues when processing Spark dataframes with non-trivial number of columnsImplement memoization for
spark_dataframe.tbl_spark()
andsdf_schema.tbl_spark()
to reduce performance overhead for somedplyr
use cases involving Spark dataframes with non-trivial number of columns
Sparklyr 1.6.2
Data
- A previous bug fix related to
dplyr::compute()
caching a Spark view needed to be further revised to take effect with dbplyr backend API edition 2
Sparklyr 1.6.1
Data
sdf_distinct()
is implemented to be an R interface fordistinct()
operation on Spark dataframes (NOTE: this is different from thedplyr::distinct()
operation, asdplyr::distinct()
operation on a Spark dataframe now supports.keep_all = TRUE
and has more complex ordering requirements)Fixed a problem of some expressions being evaluated twice in
transmute.tbl_spark()
(see tidyverse/dbplyr#605)dbExistsTable()
now performs case insensitive comparison with table names to be consistent with how table names are handled by Spark catalog APIFixed a bug with
sql_query_save()
not overwriting a temp table with identical nameRevised
sparklyr:::process_tbl_name()
to correctly handle inputs that are not table namesBug fix:
db_save_query.spark_connection()
should also cache the view it created in Spark
Sparklyr 1.6.0
Data
Made
sparklyr
compatible with both dbplyr edition 1 and edition 2 APIsRevised
sparklyr
’s integration withdbplyr
API so thatdplyr::select()
,dplyr::mutate()
, anddplyr::summarize()
verbs on Spark dataframes involvingwhere()
predicates can be correctly translated to Spark SQL (e.g., one can havesdf %>% select(where(is.numeric))
andsdf %>% summarize(across(starts_with("Petal"), mean))
, etc)Implemented
dplyr::if_all()
anddplyr::if_any()
support for Spark dataframesAdded support for
partition_by
option instream_write_*
methodsFixed a bug with URI handling affecting all
spark_read_*
methodsAvoided repeated creations of SimpleDataFormat objects and setTimeZone calls while collecting Data columns from a Spark dataframe
Schema specification for struct columns in
spark_read_*()
methods are now supported (e.g.,spark_read_json(sc, path, columns = list(s = list(a = "integer, b = "double")))
says expect a struct column nameds
with each element containing a field nameda
and a field namedb
)sdf_quantile()
andft_quantile_discretizer()
now support approximation of weighted quantiles using a modified version of the Greenwald-Khanna algorithm that takes relative weight of each data point into consideration.Fixed a problem of some expressions being evaluated twice in
transmute.tbl_spark()
(see tidyverse/dbplyr#605)Made
dplyr::distinct()
behavior for Spark dataframes configurable: settingoptions(sparklyr.dplyr_distinct.impl = "tbl_lazy)
will switchdplyr::distinct()
implementation to a basic one that only adds ‘DISTINCT’ clause to the current Spark SQL query, does not support the.keep_all = TRUE
option, and (3) does not have any ordering guarantee for the output.
Serialization
spark_write_rds()
was implemented to support exporting all partitions of a Spark dataframe in parallel into RDS (version 2) files. Such RDS files will be written to the default file system of the Spark instance (i.e., local file if the Spark instance is running locally, or a distributed file system such as HDFS if the Spark instance is deployed over a cluster). The resulting RDS files, once downloaded onto the local file system, should be deserialized into R dataframes usingcollect_from_rds()
(which callsreadRDS()
internally and also performs some important post-processing steps to support timestamp columns, date columns, and struct columns properly in R).copy_to()
can now import list columns of temporal values within a R dataframe as arrays of Spark SQL date/timestamp types when working with Spark 3.0 or aboveFixed a bug with
copy_to()
’s handling of NA values in list columns of a R dataframeSpark map type will be collected as list instead of environment in R in order to support empty string as key
Fixed a configuration-related bug in
sparklyr:::arrow_enabled()
Implemented spark-apply-specific configuration option for Arrow max records per batch, which can be different from the
spark.sql.execution.arrow.maxRecordsPerBatch
value from Spark session config
Connections
Created convenience functions for working with Spark runtime configurations
Fixed buggy exit code from the
spark-submit
process launched by sparklyr
Spark ML
Implemented R interface for Power Iteration Clustering
The
handle_invalid
option is added toft_vector_indexer()
(supported by Spark 2.3 or above)
Misc
- Fixed a bug with
~
within some path components not being normalized insparklyr::livy_install()
Sparklyr 1.5.2
Connections
Fixed
op_vars()
specification indplyr::distinct()
verb for Spark dataframesspark_disconnect()
now closes the Spark monitoring connection correctly
Data
Implement support for stratified sampling in
ft_dplyr_transformer()
Added support for
na.rm
in dplyrrowSums()
function for Spark dataframes
Sparklyr 1.5.1
Connections
A bug in how multiple
--conf
values were handled in some scenarios within the spark-submit shell args which was introduced in sparklyr 1.4 has been fixed now.A bug with
livy.jars
configuration was fixed (#2843)
Data
-
tbl()
methods were revised to be compatible withdbplyr
2.0 when handling inputs of the form"<schema name>.<table name>"
Sparklyr 1.5.0
Connections
spark_web()
has been revised to work correctly in environments such as RStudio Server or RStudio Cloud where the Spark web UI URLs such as “http://localhost:4040/jobs/” needs to be translated withrstudioapi::translateLocalUrl()
to be accessible.The problem with bundle file name collisions when
session_id
is not provided has been fixed inspark_apply_bundle()
.Support for
sparklyr.livy.sources
is removed completely as it is no longer needed as a workaround when Spark version is specified.
Data
stream_lag()
is implemented to provide the equivalent functionality ofdplyr::lag()
for streaming Spark dataframes while also supporting additional filtering of “outdated” records based on timestamp threshold.A specialized version of
dplyr::distinct()
is implemented for Spark dataframes that supports.keep_all = TRUE
and correctly satisfies the “rows are a subset of the input but appear in the same order” requirement stated in thedplyr
documentation.The default value for the
repartition
parameter ofsdf_seq()
has been corrected.Some implementation detail was revised to make
sparklyr
1.5 fully compatible withdbplyr
2.0.sdf_expand_grid()
was implemented to support roughly the equivalent ofexpand.grid()
for Spark dataframes while also offering additional Spark- specific options such as broadcast hash joins, repartitioning, and caching of the resulting Spark dataframe in memory.sdf_quantile()
now supports calculation for multiple columns.Both
lead()
andlag()
methods for dplyr interface ofsparklyr
are fixed to correctly accept theorder_by
parameter.The
cumprod()
window aggregation function for dplyr was reimplemented to correctly handle null values in Spark dataframes.Support for
missing
parameter is implemented for theifelse()
/if_else()
function for dplyr.A
weighted.mean()
summarizer was implemented for dplyr interface ofsparklyr
.A workaround was created to ensure
NA_real_
is handled correctly within the contexts ofdplyr::mutate()
anddplyr::transmute()
methods (e.g.,sdf %>% dplyr::mutate(z = NA_real_)
should result in a column named “z” with double-precision SQL type)Support for R-like subsetting operator (
[
) was implemented for selecting a subset of columns from a Spark dataframe.The
rowSums()
function was implemented for dplyr interface ofsparklyr
.The
sdf_partition_sizes()
function was created to enable efficient query of partition sizes within a Spark dataframe.Stratified sampling for Spark dataframes has been implemented and can be expressed using dplyr grammar as
<spark dataframe> %>% dplyr::group_by(<columns>) %>% dplyr::sample_n(...)
or<spark dataframe> %>% dplyr::group_by(<columns>) %>% dplyr::sample_frac(...)
where<columns>
is a list of grouping column(s) defining the strata (i.e., the sampling specified bydplyr::sample_n()
ordplyr::sample_frac()
will be applied to each group defined bydplyr::group_by(<columns>)
)The implementations of
dplyr::sample_n()
anddplyr::sample_frac()
have been revised to first perform aggregations on individual partitions before merging aggregated results from all partitions, which is more efficient thanmapPartitions()
followed byreduce()
.sdf_unnest_longer()
andsdf_unnest_wider()
were implemented and offer the equivalents oftidyr::unnest_longer()
andtidyr::unnest_wider()
for for Spark dataframes.
Serialization
copy_to()
now serializes R dataframes into RDS format instead of CSV format ifarrow
is unavailable. RDS serialization is approximately 48% faster than CSV and allows multiple correctness issues related to CSV serialization to be fixed easily insparklyr
.copy_to()
andcollect()
now correctly preserveNA_real_
(NA_real_
from a R dataframe, once translated asnull
in a Spark dataframe, used to be incorrectly collected asNaN
in previous versions ofsparklyr
).copy_to()
can now distinguish"NA"
fromNA
as expected.copy_to()
now supports importing binary columns from R dataframes to Spark.Reduced serialization overhead in Spark-based
foreach
parallel backend created withregisterDoSpark()
.
Sparklyr 1.4.0
Connections
RAPIDS GPU acceleration plugin can now be enabled with
spark_connect(..., package = "rapids")
and configured withspark_config
options prefixed with “spark.rapids.”Enabled support for http{,s} proxy plus additional CURL options for Livy connections
In sparklyr error message, suggest
options(sparklyr.log.console = TRUE)
as a trouble-shooting step whenever the “sparklyr gateway not responding” error occursAddressed an inter-op issue with Livy + Spark 2.4 (https://github.com/sparklyr/sparklyr/issues/2641)
Added configurable retries for Gateway ports query (https://github.com/sparklyr/sparklyr/pull/2654)
App name setting now takes effect as expected in YARN cluster mode (https://github.com/sparklyr/sparklyr/pull/2675)
Data
Support for newly introduced higher-order functions in Spark 3.0 (e.g.,
array_sort
,map_filter
,map_zip_with
, and many others)Implemented parallelizable weighted sampling methods for sampling from a Spark data frames with and without replacement using exponential variates
Replaced
dplyr::sample_*
implementations based onTABLESAMPLE
with alternative implementation that can return exactly the number of rows or fraction specified and also properly support sampling with-replacement, without-replacement, and repeatable sampling use casesAll higher-order functions and sampling methods are made directly accessible through
dplyr
verbsMade
grepl
part of thedplyr
interface for Spark data framesTidyr verbs such as
pivot_wider
,pivot_longer
,nest
,unnest
,separate
,unite
, andfill
now have specialized implementations insparklyr
for working with Spark data framesMade
dplyr::inner_join
,dplyr::left_join
,dplyr::right_join
, anddplyr::full_join
replace'.'
with'_'
insuffix
parameter when working with Spark data frames (https://github.com/sparklyr/sparklyr/issues/2648)
Distributed R
Fixed an issue with global variables in
registerDoSpark
(https://github.com/sparklyr/sparklyr/pull/2608)Revised
spark_read_compat_param
to avoid collision on names assigned to different Spark data frames
Misc
Fixed a rendering issue with HTML reference pages
Made test reporting in Github CI workflows more informative (https://github.com/sparklyr/sparklyr/pull/2672)
Spark ML
-
ft_robust_scaler
was created as the R interface for theRobustScaler
functionality in Spark 3 or above
Sparklyr 1.3.1
Distributed R
- Fixed a bug in ordering of parameters for a lamba expression when the lambda expression passed to a
hof_*
method is specified with a R formula and the lambda takes 2 parameters
Sparklyr 1.3.0
Spark ML
-
ml_evaluate()
methods are implemented for ML clustering and classification models
Distributed R
Created helper methods to integrate Spark SQL higher-order functions with
dplyr::mutate
Implemented option to pass partition index as a named parameter to
spark_apply()
transform functionEnabled transform function of
spark_apply()
to return nested listsAdded option to return R objects instead of Spark data frame rows from transform function of
spark_apply
sdf_collect()
now supports fetching Spark data frame row-by-row rather than column-by-column, and fetching rows using iterator instead of collecting all rows into memorySupport for
partition
when using barrier execution inspark_apply
(#2454)
Connections
Sparklyr can now connect with Spark 2.4 built with Scala 2.12 using
spark_connect(..., scala_version = "2.12")
Hive integration can now be disabled by configuration in
spark_connect()
(#2465)A JVM object reference counting bug affecting secondary Spark connections was fixed (#2515)
Revised JObj envs initialization for Databricks connections (#2533)
Serialization
Timezones, if present in data, are correctly represented now in Arrow serialization
Embedded nul bytes are removed from strings when reading strings from Spark to R (#2250)
Support to collect objectts of type
SeqWrapper
(#2441)
Data
Created helper methods to integrate Spark SQL higher-order functions with
dplyr::mutate
New
spark_read()
method to allow user-defined R functions to be run on Spark workers to import data into a Spark data framespark_write()
method is implemented allow user-defined functions to be run on Spark workers to export data from a Spark data frameAvro functionalities such as
spark_read_avro()
,spark_write_avro()
,sdf_from_avro()
, andsdf_to_avro()
are implemented and can be optionally enabled withspark_connect(..., package = "avro")
Extensions
- Fixed a bug where Spark package repositories specification was not honored by
spark_dependency()
. Therepositories
parameter ofspark_dependency()
now works as expected.
Misc
Fixed warnings for deprecated functions (#2431)
More test coverage for Databricks Connect and Databricks Notebook modes
Embedded R sources are now included as resources rather than as a Scala string literal in
sparklyr-*.jar
files, so that they can be updated without re-compilation of Scala source filesA mechanism is created to verify embedded sources in
sparklyr-*.jar
files are in-sync with current R source files and this verification is now part of the Github CI workflow forsparklyr
Sparklyr 1.2.0
Distributed R
Add support for using Spark as a foreach parallel backend
Fixed a bug with how
columns
parameter was interpreted inspark_apply
Data
Allow
sdf_query_plan
to also get analyzed planAdd support for serialization of R date values into corresponding Hive date values
Fixed the issue of date or timestamp values representing the UNIX epoch (1970-01-01) being deserialized incorrectly into NAs
Better support for querying and deserializing Spark SQL struct columns when working with Spark 2.4 or above
Add support in
copy_to()
for columns with nested lists (#2247).Significantly improve
collect()
performance for columns with nested lists (#2252).
Connection
Add support for Databricks Connect
Add support for
copy_to
in Databricks connectionEnsure spark apply bundle files created by multiple Spark sessions don’t overwrite each other
Fixed an interop issue with spark-submit when running with Spark 3 preview
Fixed an interop issue with Sparklyr gateway connection when running with Spark 3 preview
Fixed a race condition of JVM object with refcount 1 being removed from JVM object tracker before pending method invocation(s) on them could be initiated (NOTE: previously this would only happen when the R process was running under high memory pressure)
Allow a chain of JVM method invocations to be batched into 1
invoke
callRemoval of unneeded objects from JVM object tracker no longer blocks subsequent JVM method invocations
Add support for JDK11 for Spark 3 preview.
Misc
Support for installing Spark 3.0 Preview 2.
Emit more informative error message if network interface required for
spark_connect
is not upFixed a bug preventing more than 10 rows of a Spark table to be printed from R
Fixed a spelling error in
print
method forml_model_naive_bayes
objectsMade
sdf_drop_duplicates
an exported function (previously it was not exported by mistake)Fixed a bug in
summary()
ofml_linear_regression
Sparklyr 1.1.0
Distributed R
- Add support for barrier execution mode with
barrier = TRUE
inspark_apply()
(@samuelmacedo83, #2216).
Streaming
Add support for
stream_read_delta()
andstream_write_delta()
.Fixed typo in
stream_read_socket()
.
Data
Allow using Scala types in schema specifications. For example,
StringType
in thecolumns
parameter forspark_read_csv()
(@jozefhajnala, #2226)Add support for
DBI 1.1
to implement missingdbQuoteLiteral
signature (#2227).
Livy
Add support for Livy 0.6.0.
Deprecate uploading sources to Livy, a jar is now always used and the
version
parameter inspark_connect()
is always required.Add config
sparklyr.livy.branch
to specify the branch used for the sparklyr JAR.Add config
sparklyr.livy.jar
to configure path or URL to sparklyr JAR.
Data
- Add support for
partition_by
when usingspark_write_delta()
(#2228).
Sparklyr 1.0.5
Serialization
- R environments are now sent to Scala Maps rather than
java.util.Map[Object, Object]
(#1058).
Data
Allow
sdf_sql()
to accept glue strings (@yutannihilation, #2171).Support to read and write from Delta Lake using
spark_read_delta()
andspark_write_delta()
(#2148).
Connections
spark_connect()
supports newpackages
parameter to easily enablekafka
anddelta
(#2148).spark_disconnect()
returns invisibly (#2028).
Configuration
- Support to specify config file location using the
SPARKLYR_CONFIG_FILE
environment variable (@AgrawalAmey, #2153).
Compilation
- Support for Scala 12 (@lu-wang-dl, #2154).
YARN
- Fix
curl_fetch_memory
error when using YARN Cluster mode (#2157).
Sparklyr 1.0.4
Arrow
- Support for Apache Arrow 0.15 (@nealrichardson, #2132).
Sparklyr 1.0.3
Kuberenetes
- Support for port forwarding in Windows using RStudio terminal.
dplyr
- Fix support for
compute()
in Spark 1.6 (#2099)
Data
- The
spark_read_()
functions now support multiple parameters (@jozefhajnala, #2118).
Connections
- Fix for Qubole connections for single user and multiple sessions (@vipul1409, #2128).
Sparklyr 1.0.2
Connections
- Support for Qubole connections using
mode = "quobole"
(@vipul1409, #2039).
Extensions
- When
invoke()
fails due to mismatched parameters, warning with info is logged.
RStudio
- Spark UI path can now be accessed even when the R session and Spark are bussy.
Distributed
Configuration setting
sparklyr.apply.serializer
can be used to select serializer version inspark_apply()
.Fix for
spark_apply_log()
and useRClosure
as logging component.
ML
-
ml_corr()
retrieve atibble
for better formatting.
Misc
- Support for Spark 2.3.3 and 2.4.3.
Data
The
infer_schema
parameter now defaults tois.null(column)
.The
spark_read_()
functions support loading data with namedpath
but no explicitname
.
Sparklyr 1.0.1
ML
ml_lda()
: Allow passing of optional arguments via...
to regex tokenizer, stop words remover, and count vectorizer components in the formula API.Implemented
ml_evaluate()
for logistic regression, linear regression, and GLM models.Implemented
print()
method forml_summary
objects.Deprecated
compute_cost()
for KMeans in Spark 2.4 (#1772).Added missing internal constructor for clustering evaluator (#1936).
sdf_partition()
has been renamed tosdf_random_split()
.Added
ft_one_hot_encoder_estimator()
(#1337).
Misc
Added
sdf_crosstab()
to create contingency tables.Fix
tibble::as.tibble()
deprecation warning.
Connections
- Reduced default memory for local connections when Java x64 is not installed (#1931).
Batches
- Add support in
spark-submit
with R file to pass additional arguments to R file (#1942).
Distributed R
- Fix support for multiple library paths when using
spark.r.libpaths
(@mattpollock, #1956).
Extensions
Support for creating an Spark extension package using
spark_extension()
.Add support for repositories in
spark_dependency()
.
DataFrames
- Fix
sdf_bind_cols()
when usingdbplyr
1.4.0.
Kubernetes
- Fix regression in
spark_config_kubernetes()
configuration helper.
Sparklyr 1.0.0
Arrow
- Support for Apache Arrow using the
arrow
package.
ML
The
dataset
parameter for estimator feature transformers has been deprecated (#1891).ml_multilayer_perceptron_classifier()
gains probabilistic classifier parameters (#1798).Removed support for all undocumented/deprecated parameters. These are mostly dot case parameters from pre-0.7.
Remove support for deprecated
function(pipeline_stage, data)
signature insdf_predict/transform/fit
functions.Soft deprecate
sdf_predict/transform/fit
functions. Users are advised to useml_predict/transform/fit
functions instead.Utilize the ellipsis package to provide warnings when unsupported arguments are specified in ML functions.
Livy
Support for sparklyr extensions when using Livy.
Significant performance improvements by using
version
inspark_connect()
which enables using the sparklyr JAR rather than sources.Improved memory use in Livy by using string builders and avoid print backs.
Data
Fix for
DBI::sqlInterpolate()
and related methods to properly quote parameterized queries.copy_to()
names tablessparklyr_tmp_
instead ofsparklyr_
for consistency with other temp tables and to avoid rendering them under the connections pane.copy_to()
andcollect()
are not re-exported since they are commonly used even when usingDBI
or outside data analysis use cases.Support for reading
path
as the second parameter inspark_read_*()
when no name is specified (e.g.spark_read_csv(sc, "data.csv")
).Support for batches in
sdf_collect()
anddplyr::collect()
to retrieve data incrementally using a callback function provided through acallback
parameter. Useful when retrieving larger datasets.Support for batches in
sdf_copy_to()
anddplyr::copy_to()
by passing a list of callbacks that retrieve data frames. Useful when uploading larger datasets.spark_read_source()
now has apath
parameter for specifying file path.Support for
whole
parameter forspark_read_text()
to read an entire text file without splitting contents by line.
Broom
Connections
Local connection defaults now to 2GB.
Support to install and connect based on major Spark versions, for instance:
spark_connect(master = "local", version = "2.4")
.Support for installing and connecting to Spark 2.4.
Serialization
- Faster retrieval of string arrays.
YARN
New YARN action under RStudio connection pane extension to launch YARN UI. Configurable through the
sparklyr.web.yarn
configuration setting.Support for property expansion in
yarn-site.xml
(@lgongmsft, #1876).
Distributed R
- The
memory
parameter inspark_apply()
now defaults toFALSE
when thename
parameter is not specified.
Other
Removed dreprecated
sdf_mutate()
.Remove exported
ensure_
functions which were deprecated.Fixed missing Hive tables not rendering under some Spark distributions (#1823).
Remove dependency on broom.
Fixed re-entrancy job progress issues when running RStudio 1.2.
Tables with periods supported by setting
sparklyr.dplyr.period.splits
toFALSE
.sdf_len()
,sdf_along()
andsdf_seq()
default to 32 bit integers but allow support for 64 bits throughbits
parameter.Support for detecting Spark version using
spark-submit
.
Sparklyr 0.9.4
Improved multiple streaming documentation examples (#1801, #1805, #1806).
Fix issue while printing Spark data frames under
tibble
2.0.0 (#1829).Support for
stream_write_console()
to write to console log.Support for
stream_read_scoket()
to read socket streams.Fix to
spark_read_kafka()
to remove unusedpath
.
Sparklyr 0.9.3
Fix to make
spark_config_kubernetes()
work with variablejar
parameters.Support to install and use Spark 2.4.0.
Improvements and fixes to
spark_config_kubernetes()
parameters.Support for
sparklyr.connect.ondisconnect
config setting to allow cleanup of resources when using kubernetes.spark_apply()
andspark_apply_bundle()
properly dereference symlinks when creating package bundle (@awblocker, #1785)Fix
tableName
warning triggered while connecting.Deprecate
sdf_mutate()
(#1754).Fix requirement to specify
SPARK_HOME_VERSION
whenversion
parameter is set inspark_connect()
.Cloudera autodetect Spark version improvements.
Fixed default for
session
inreactiveSpark()
.Removed
stream_read_jdbc()
andstream_write_jdbc()
since they are not yet implemented in Spark.Support for collecting NA values from logical columns (#1729).
Proactevely clean JVM objects when R object is deallocated.
Sparklyr 0.9.2
Support for Spark 2.3.2.
Fix installation error with older versions of
rstudioapi
(#1716).Fix missing callstack and error case while logging in
spark_apply()
.Proactevely clean JVM objects when R object is deallocated.
Broom
- Implemented
tidy()
,augment()
, andglance()
forml_linear_svc()
andml_pca()
models (@samuelmacedo83)
Sparklyr 0.9.2
Support for Spark 2.3.2.
Fix installation error with older versions of
rstudioapi
(#1716).Fix missing callstack and error case while logging in
spark_apply()
.Fix regression in
sdf_collect()
failing to collect tables.Fix new connection RStudio selectors colors when running under OS X Mojave.
Support for launching Livy logs from connection pane.
Sparklyr 0.9.2
Removed
overwrite
parameter inspark_read_table()
(#1698).Fix regression preventing using R 3.2 (#1695).
Additional jar search paths under Spark 2.3.1 (#1694)
Sparklyr 0.9.1
Terminate streams when Shiny app terminates.
Fix
dplyr::collect()
with Spark streams and improve printing.Fix regression in
sparklyr.sanitize.column.names.verbose
setting which would cause verbose column renames.Fix to
stream_write_kafka()
andstream_write_jdbc()
.
Sparklyr 0.9.0
Streaming
Support for
stream_read_*()
andstream_write_*()
to read from and to Spark structured streams.Support for
dplyr
,sdf_sql()
,spark_apply()
and scoring pipeline in Spark streams.Support for
reactiveSpark()
to create ashiny
reactive over a Spark stream.Support for convenience functions
stream_*()
to stop, change triggers, print, generate test streams, etc.
Monitoring
Support for interrupting long running operations and recover gracefully using the same connection.
Support cancelling Spark jobs by interrupting R session.
Support for monitoring job progress within RStudio, required RStudio 1.2.
Progress reports can be turned off by setting
sparklyr.progress
toFALSE
inspark_config()
.
Kubernetes
Added config
sparklyr.gateway.routing
to avoid routing to ports since Kubernetes clusters have unique spark masters.Change backend ports to be choosen deterministically by searching for free ports starting on
sparklyr.gateway.port
which default to8880
. This allows users to enable port forwarding withkubectl port-forward
.Added support to set config
sparklyr.events.aftersubmit
to a function that is called afterspark-submit
which can be used to automatically configure port forwarding.
Batches
- Added support for
spark_submit()
to assist submitting non-interactive Spark jobs.
Spark ML
-
(Breaking change) The formula API for ML classification algorithms no longer indexes numeric labels, to avoid the confusion of
0
being mapped to"1"
and vice versa. This means that if the largest numeric label isN
, Spark will fit aN+1
-class classification model, regardless of how many distinct labels there are in the provided training set (#1591). - Fix retrieval of coefficients in
ml_logistic_regression()
(@shabbybanks, #1596). -
(Breaking change) For model objects,
lazy val
anddef
attributes have been converted to closures, so they are not evaluated at object instantiation (#1453). - Input and output column names are no longer required to construct pipeline objects to be consistent with Spark (#1513).
- Vector attributes of pipeline stages are now printed correctly (#1618).
- Deprecate various aliases favoring method names in Spark.
- Deprecate incompatible signatures for
sdf_transform()
andml_transform()
families of methods; the former should take atbl_spark
as the first argument while the latter should take a model object as the first argument. - Input and output column names are no longer required to construct pipeline objects to be consistent with Spark (#1513).
Data
Implemented support for
DBI::db_explain()
(#1623).Fixed for
timestamp
fields when usingcopy_to()
(#1312, @yutannihilation).Added support to read and write ORC files using
spark_read_orc()
andspark_write_orc()
(#1548).
Livy
Fixed
must share the same src
error forsdf_broadcast()
and other functions when using Livy connections.Added support for logging
sparklyr
server events and logging sparklyr invokes as comments in the Livy UI.Added support to open the Livy UI from the connections viewer while using RStudio.
Improve performance in Livy for long execution queries, fixed
livy.session.command.timeout
and support forlivy.session.command.interval
to control max polling while waiting for command response (#1538).Fixed Livy version with MapR distributions.
Removed
install
column fromlivy_available_versions()
.
Distributed R
Added
name
parameter tospark_apply()
to optionally name resulting table.Fix to
spark_apply()
to retain column types when NAs are present (#1665).spark_apply()
now supportsrlang
anonymous functions. For example,sdf_len(sc, 3) %>% spark_apply(~.x+1)
.Breaking Change:
spark_apply()
no longer defaults to the input column names when thecolumns
parameter is nos specified.Support for reading column names from the R data frame returned by
spark_apply()
.Fix to support retrieving empty data frames in grouped
spark_apply()
operations (#1505).Added support for
sparklyr.apply.packages
to configure default behavior forspark_apply()
parameters (#1530).Added support for
spark.r.libpaths
to configure package library inspark_apply()
(#1530).
Connections
Default to Spark 2.3.1 for installation and local connections (#1680).
ml_load()
no longer keeps extraneous table views which was cluttering up the RStudio Connections pane (@randomgambit, #1549).Avoid preparing windows environment in non-local connections.
Extensions
The
ensure_*
family of functions is deprecated in favor of forge which doesn’t use NSE and provides more informative errors messages for debugging (#1514).Support for
sparklyr.invoke.trace
andsparklyr.invoke.trace.callstack
configuration options to trace allinvoke()
calls.Support to invoke methods with
char
types using single character strings (@lawremi, #1395).
Serialization
- Fixed collection of
Date
types to support correct local JVM timezone to UTC ().
Documentation
- Many new examples for
ft_binarizer()
,ft_bucketizer()
,ft_min_max_scaler
,ft_max_abs_scaler()
,ft_standard_scaler()
,ml_kmeans()
,ml_pca()
,ml_bisecting_kmeans()
,ml_gaussian_mixture()
,ml_naive_bayes()
,ml_decision_tree()
,ml_random_forest()
,ml_multilayer_perceptron_classifier()
,ml_linear_regression()
,ml_logistic_regression()
,ml_gradient_boosted_trees()
,ml_generalized_linear_regression()
,ml_cross_validator()
,ml_evaluator()
,ml_clustering_evaluator()
,ml_corr()
,ml_chisquare_test()
andsdf_pivot()
(@samuelmacedo83).
Broom
- Implemented
tidy()
,augment()
, andglance()
forml_aft_survival_regression()
,ml_isotonic_regression()
,ml_naive_bayes()
,ml_logistic_regression()
,ml_decision_tree()
,ml_random_forest()
,ml_gradient_boosted_trees()
,ml_bisecting_kmeans()
,ml_kmeans()
andml_gaussian_mixture()
models (@samuelmacedo83)
Configuration
Deprecated configuration option
sparklyr.dplyr.compute.nocache
.Added
spark_config_settings()
to list allsparklyr
configuration settings and describe them, cleaned all settings and grouped by area while maintaining support for previous settings.Static SQL configuration properties are now respected for Spark 2.3, and
spark.sql.catalogImplementation
defaults tohive
to maintain Hive support (#1496, #415).spark_config()
values can now also be specified asoptions()
.Support for functions as values in entries to
spark_config()
to enable advanced configuration workflows.
Sparklyr 0.8.4
Added support for
spark_session_config()
to modify spark session settings.Added support for
sdf_debug_string()
to print execution plan for a Spark DataFrame.Fixed DESCRIPTION file to include test packages as requested by CRAN.
Support for
sparklyr.spark-submit
asconfig
entry to allow customizing thespark-submit
command.Changed
spark_connect()
to give precedence to theversion
parameter overSPARK_HOME_VERSION
and other automatic version detection mechanisms, improved automatic version detection in Spark 2.X.Fixed
sdf_bind_rows()
withdplyr 0.7.5
and prepend id column instead of appending it to match behavior.broom::tidy()
for linear regression and generalized linear regression models now give correct results (#1501).
Sparklyr 0.8.3
- Support for Spark 2.3 in local windows clusters (#1473).
Sparklyr 0.8.2
Support for resource managers using
https
inyarn-cluster
mode (#1459).Fixed regression for connections using Livy and Spark 1.6.X.
Sparklyr 0.8.1
- Fixed regression for connections using
mode
withdatabricks
.
Sparklyr 0.8.0
Spark ML
Added
ml_validation_metrics()
to extract validation metrics from cross validator and train split validator models.ml_transform()
now also takes a list of transformers, e.g. the result ofml_stages()
on aPipelineModel
(#1444).Added
collect_sub_models
parameter toml_cross_validator()
andml_train_validation_split()
and helper functionml_sub_models()
to allow inspecting models trained for each fold/parameter set (#1362).Added
parallelism
parameter toml_cross_validator()
andml_train_validation_split()
to allow tuning in parallel (#1446).Added support for
feature_subset_strategy
parameter in GBT algorithms (#1445).Added
string_order_type
toft_string_indexer()
to allow control over how strings are indexed (#1443).Added
ft_string_indexer_model()
constructor for the string indexer transformer (#1442).Added
ml_feature_importances()
for extracing feature importances from tree-based models (#1436).ml_tree_feature_importance()
is maintained as an alias.Added
ml_vocabulary()
to extract vocabulary from count vectorizer model andml_topics_matrix()
to extract matrix from LDA model.ml_tree_feature_importance()
now works properly with decision tree classification models (#1401).Added
ml_corr()
for calculating correlation matrices andml_chisquare_test()
for performing chi-square hypothesis testing (#1247).ml_save()
outputs message when model is successfully saved (#1348).ml_
routines no longer capture the calling expression (#1393).Added support for
offset
argument inml_generalized_linear_regression()
(#1396).Fixed regression blocking use of response-features syntax in some
ml_
functions (#1302).Added support for Huber loss for linear regression (#1335).
ft_bucketizer()
andft_quantile_discretizer()
now support multiple input columns (#1338, #1339).Added
ft_feature_hasher()
(#1336).Added
ml_clustering_evaluator()
(#1333).ml_default_stop_words()
now returns English stop words by default (#1280).Support the
sdf_predict(ml_transformer, dataset)
signature with a deprecation warning. Also added a deprecation warning to the usage ofsdf_predict(ml_model, dataset)
. (#1287)Fixed regression blocking use of
ml_kmeans()
in Spark 1.6.x.
Extensions
invoke*()
method dispatch now supportsChar
andShort
parameters. Also,Long
parameters now allow numeric arguments, but integers are supported for backwards compatibility (#1395).invoke_static()
now supports calling Scala’s package objects (#1384).spark_connection
andspark_jobj
classes are now exported (#1374).
Distributed R
Added support for
profile
parameter inspark_apply()
that collects a profile to measure perpformance that can be rendered using theprofvis
package.Added support for
spark_apply()
under Livy connections.Fixed file not found error in
spark_apply()
while working under low disk space.Added support for
sparklyr.apply.options.rscript.before
to run a custom command before launching the R worker role.Added support for
sparklyr.apply.options.vanilla
to be set toFALSE
to avoid using--vanilla
while launching R worker role.Fixed serialization issues most commonly hit while using
spark_apply()
with NAs (#1365, #1366).Fixed issue with dates or date-times not roundtripping with `spark_apply() (#1376).
Fixed data frame provided by
spark_apply()
to not provide characters not factors (#1313).
Miscellaneous
Fixed typo in
sparklyr.yarn.cluster.hostaddress.timeot
(#1318).Fixed regression blocking use of
livy.session.start.timeout
parameter in Livy connections.Added support for Livy 0.4 and Livy 0.5.
Livy now supports Kerberos authentication.
Default to Spark 2.3.0 for installation and local connections (#1449).
yarn-cluster
now supported by connecting withmaster="yarn"
andconfig
entrysparklyr.shell.deploy-mode
set tocluster
(#1404).sample_frac()
andsample_n()
now work properly in nontrivial queries (#1299)sdf_copy_to()
no longer gives a spurious warning when user enters a multiline expression forx
(#1386).spark_available_versions()
was changed to only return available Spark versions, Hadoop versions can be still retrieved usinghadoop = TRUE
.spark_installed_versions()
was changed to retrieve the full path to the installation folder.cbind()
andsdf_bind_cols()
don’t use NSE internally anymore and no longer output names of mismatched data frames on error (#1363).
Sparklyr 0.7.0
Added support for Spark 2.2.1.
Switched
copy_to
serializer to use Scala implementation, this change can be reverted by setting thesparklyr.copy.serializer
option tocsv_file
.Added support for
spark_web()
for Livy and Databricks connections when using Spark 2.X.Fixed
SIGPIPE
error underspark_connect()
immediately after aspark_disconnect()
operation.spark_web()
is is more reliable under Spark 2.X by making use of a new API to programmatically find the right address.Added support in
dbWriteTable()
fortemporary = FALSE
to allow persisting table across connections. Changed default value fortemporary
toTRUE
to matchDBI
specification, for compatibility, default value can be reverted back toFALSE
using thesparklyr.dbwritetable.temp
option.ncol()
now returns the number of columns instead ofNA
, andnrow()
now returnsNA_real_
.Added support to collect
VectorUDT
column types with nested arrays.Fixed issue in which connecting to Livy would fail due to long user names or long passwords.
Fixed error in the Spark connection dialog for clusters using a proxy.
Improved support for Spark 2.X under Cloudera clusters by prioritizing use of
spark2-submit
overspark-submit
.Livy new connection dialog now prompts for password using
rstudioapi::askForPassword()
.Added
schema
parameter tospark_read_parquet()
that enables reading a subset of the schema to increase performance.Implemented
sdf_describe()
to easily compute summary statistics for data frames.Fixed data frames with dates in
spark_apply()
retrieved asDate
instead of doubles.Added support to use
invoke()
with arrays of POSIXlt and POSIXct.Added support for
context
parameter inspark_apply()
to allow callers to pass additional contextual information to thef()
closure.Implemented workaround to support in
spark_write_table()
formode = 'append'
.Various ML improvements, including support for pipelines, additional algorithms, hyper-parameter tuning, and better model persistence.
Added
spark_read_libsvm()
for reading libsvm files.Added support for separating struct columns in
sdf_separate_column()
.Fixed collection of
short
,float
andbyte
to properly return NAs.Added
sparklyr.collect.datechars
option to enable collectingDateType
andTimestampTime
ascharacters
to support compatibility with previos versions.Fixed collection of
DateType
andTimestampTime
fromcharacter
to properDate
andPOSIXct
types.
Sparklyr 0.6.4
Added support for HTTPS for
yarn-cluster
which is activated by settingyarn.http.policy
toHTTPS_ONLY
inyarn-site.xml
.Added support for
sparklyr.yarn.cluster.accepted.timeout
underyarn-cluster
to allow users to wait for resources under cluster with high waiting times.Fix to
spark_apply()
when package distribution deadlock triggers in environments where multiple executors run under the same node.Added support in
spark_apply()
for specifying a list ofpackages
to distribute to each worker node.Added support in
yarn-cluster
forsparklyr.yarn.cluster.lookup.prefix
,sparklyr.yarn.cluster.lookup.username
andsparklyr.yarn.cluster.lookup.byname
to control the new application lookup behavior.
Sparklyr 0.6.3
Enabled support for Java 9 for clusters configured with Hadoop 2.8. Java 9 blocked on ‘master=local’ unless ‘options(sparklyr.java9 = TRUE)’ is set.
Fixed issue in
spark_connect()
where usingset.seed()
before connection would cause session ids to be duplicates and connections to be reused.Fixed issue in
spark_connect()
blocking gateway port when connection was never started to the backend, for isntasnce, while interrupting the r session while connecting.Performance improvement for quering field names from tables impacting tables and
dplyr
queries, most noticeable inna.omit
with several columns.Fix to
spark_apply()
when closure returns adata.frame
that contains no rows and has one or more columns.Fix to
spark_apply()
while usingtryCatch()
within closure and increased callstack printed to logs when error triggers within closure.Added support for the
SPARKLYR_LOG_FILE
environment variable to specify the file used for log output.Fixed regression for
union_all()
affecting Spark 1.6.X.Added support for
na.omit.cache
option that when set toFALSE
will preventna.omit
from caching results when rows are dropped.Added support in
spark_connect()
foryarn-cluster
with hight-availability enabled.Added support for
spark_connect()
withmaster="yarn-cluster"
to query YARN resource manager API and retrieve the correct container host name.Fixed issue in
invoke()
calls while using integer arrays that containNA
which can be commonly experienced while usingspark_apply()
.Added
topics.description
underml_lda()
result.Added support for
ft_stop_words_remover()
to strip out stop words from tokens.Feature transformers (
ft_*
functions) now explicitly requireinput.col
andoutput.col
to be specified.Added support for
spark_apply_log()
to enable logging in worker nodes while usingspark_apply()
.Fix to
spark_apply()
forSparkUncaughtExceptionHandler
exception while running over large jobs that may overlap during an, now unnecesary, unregister operation.Fix race-condition first time
spark_apply()
is run when more than one partition runs in a worker and both processes try to unpack the packages bundle at the same time.spark_apply()
now adds generic column names when needed and validatesf
is afunction
.Improved documentation and error cases for
metric
argument inml_classification_eval()
andml_binary_classification_eval()
.Fix to
spark_install()
to use the/logs
subfolder to store locallog4j
logs.Fix to
spark_apply()
when R is used from a worker node since worker node already contains packages but still might be triggering different R session.Fix connection from closing when
invoke()
attempts to use a class with a method that contains a reference to an undefined class.Implemented all tuning options from Spark ML for
ml_random_forest()
,ml_gradient_boosted_trees()
, andml_decision_tree()
.Avoid tasks failing under
spark_apply()
and multiple concurrent partitions running while selecting backend port.Added support for numeric arguments for
n
inlead()
for dplyr.Added unsupported error message to
sample_n()
andsample_frac()
when Spark is not 2.0 or higher.Fixed
SIGPIPE
error underspark_connect()
immediately after aspark_disconnect()
operation.Added support for
sparklyr.apply.env.
underspark_config()
to allowspark_apply()
to initializae environment varaibles.Added support for
spark_read_text()
andspark_write_text()
to read from and to plain text files.Addesd support for RStudio project templates to create an “R Package using sparklyr”.
Fix
compute()
to trigger refresh of the connections view.Added a
k
argument toml_pca()
to enable specification of number of principal components to extract. Also implementedsdf_project()
to project datasets using the results ofml_pca()
models.Added support for additional livy session creation parameters using the
livy_config()
function.
Sparklyr 0.6.2
- Fix connection_spark_shinyapp() under RStudio 1.1 to avoid error while listing Spark installation options for the first time.
Sparklyr 0.6.1
Fixed error in
spark_apply()
that may triggered when multiple CPUs are used in a single node due to race conditions while accesing the gateway service and another in theJVMObjectTracker
.spark_apply()
now supports explicit column types using thecolumns
argument to avoid sampling types.spark_apply()
withgroup_by
no longer requires persisting to disk nor memory.Added support for Spark 1.6.3 under
spark_install()
.Added support for Spark 1.6.3 under
spark_install()
spark_apply()
now logs the current callstack when it fails.Fixed error triggered while processing empty partitions in
spark_apply()
.Fixed slow printing issue caused by
print
calculating the total row count, which is expensive for some tables.Fixed
sparklyr 0.6
issue blocking concurrentsparklyr
connections, which required to setconfig$sparklyr.gateway.remote = FALSE
as workaround.
Sparklyr 0.6.0
Distributed R
Added
packages
parameter tospark_apply()
to distribute packages across worker nodes automatically.Added
sparklyr.closures.rlang
as aspark_config()
value to support generic closures provided by therlang
package.Added config options
sparklyr.worker.gateway.address
andsparklyr.worker.gateway.port
to configure gateway used under worker nodes.Added
group_by
parameter tospark_apply()
, to support operations over groups of dataframes.Added
spark_apply()
, allowing users to use R code to directly manipulate and transform Spark DataFrames.
External Data
Added
spark_write_source()
. This function writes data into a Spark data source which can be loaded through an Spark package.Added
spark_write_jdbc()
. This function writes from a Spark DataFrame into a JDBC connection.Added
columns
parameter tospark_read_*()
functions to load data with named columns or explicit column types.Added
partition_by
parameter tospark_write_csv()
,spark_write_json()
,spark_write_table()
andspark_write_parquet()
.Added
spark_read_source()
. This function reads data from a Spark data source which can be loaded through an Spark package.Added support for
mode = "overwrite"
andmode = "append"
tospark_write_csv()
.spark_write_table()
now supports saving to default Hive path.Improved performance of
spark_read_csv()
reading remote data wheninfer_schema = FALSE
.Added
spark_read_jdbc()
. This function reads from a JDBC connection into a Spark DataFrame.Renamed
spark_load_table()
andspark_save_table()
intospark_read_table()
andspark_write_table()
for consistency with existingspark_read_*()
andspark_write_*()
functions.Added support to specify a vector of column names in
spark_read_csv()
to specify column names without having to set the type of each column.Improved
copy_to()
,sdf_copy_to()
anddbWriteTable()
performance underyarn-client
mode.
dplyr
Support for
cumprod()
to calculate cumulative products.Support for
cor()
,cov()
,sd()
andvar()
as window functions.Support for Hive built-in operators
%like%
,%rlike%
, and%regexp%
for matching regular expressions infilter()
andmutate()
.Support for dplyr (>= 0.6) which among many improvements, increases performance in some queries by making use of a new query optimizer.
sample_frac()
takes a fraction instead of a percent to match dplyr.Improved performance of
sample_n()
andsample_frac()
through the use ofTABLESAMPLE
in the generated query.
Databases
Added
src_databases()
. This function list all the available databases.Added
tbl_change_db()
. This function changes current database.
DataFrames
Added
sdf_len()
,sdf_seq()
andsdf_along()
to help generate numeric sequences as Spark DataFrames.Added
spark_set_checkpoint_dir()
,spark_get_checkpoint_dir()
, andsdf_checkpoint()
to enable checkpointing.Added
sdf_broadcast()
which can be used to hint the query optimizer to perform a broadcast join in cases where a shuffle hash join is planned but not optimal.Added
sdf_repartition()
,sdf_coalesce()
, andsdf_num_partitions()
to support repartitioning and getting the number of partitions of Spark DataFrames.Added
sdf_bind_rows()
andsdf_bind_cols()
– these functions are thesparklyr
equivalent ofdplyr::bind_rows()
anddplyr::bind_cols()
.Added
sdf_separate_column()
– this function allows one to separate components of an array / vector column into separate scalar-valued columns.sdf_with_sequential_id()
now supportsfrom
parameter to choose the starting value of the id column.Added
sdf_pivot()
. This function provides a mechanism for constructing pivot tables, using Spark’s ‘groupBy’ + ‘pivot’ functionality, with a formula interface similar to that ofreshape2::dcast()
.
MLlib
Added
vocabulary.only
toft_count_vectorizer()
to retrieve the vocabulary with ease.GLM type models now support
weights.column
to specify weights in model fitting. (#217)ml_logistic_regression()
now supports multinomial regression, in addition to binomial regression [requires Spark 2.1.0 or greater]. (#748)Implemented
residuals()
andsdf_residuals()
for Spark linear regression and GLM models. The former returns a R vector while the latter returns atbl_spark
of training data with aresiduals
column added.Added
ml_model_data()
, used for extracting data associated with Spark ML models.The
ml_save()
andml_load()
functions gain ameta
argument, allowing users to specify where R-level model metadata should be saved independently of the Spark model itself. This should help facilitate the saving and loading of Spark models used in non-local connection scenarios.ml_als_factorization()
now supports the implicit matrix factorization and nonnegative least square options.Added
ft_count_vectorizer()
. This function can be used to transform columns of a Spark DataFrame so that they might be used as input toml_lda()
. This should make it easier to invokeml_lda()
on Spark data sets.
Broom
R Compatibility
- Implemented
cbind.tbl_spark()
. This method works by first generating index columns usingsdf_with_sequential_id()
then performinginner_join()
. Note that dplyr_join()
functions should still be used for DataFrames with common keys since they are less expensive.
Connections
Increased default number of concurrent connections by setting default for
spark.port.maxRetries
from 16 to 128.Support for gateway connections
sparklyr://hostname:port/session
and usingspark-submit --class sparklyr.Shell sparklyr-2.1-2.11.jar <port> <id> --remote
.Added support for
sparklyr.gateway.service
andsparklyr.gateway.remote
to enable/disable the gateway in service and to accept remote connections required for Yarn Cluster mode.Added support for Yarn Cluster mode using
master = "yarn-cluster"
. Either, explicitly setconfig = list(sparklyr.gateway.address = "<driver-name>")
or implicitlysparklyr
will read thesite-config.xml
for theYARN_CONF_DIR
environment variable.Added
spark_context_config()
andhive_context_config()
to retrieve runtime configurations for the Spark and Hive contexts.Added
sparklyr.log.console
to redirect logs to console, useful to troubleshootingspark_connect
.Added
sparklyr.backend.args
as config option to enable passing parameters to thesparklyr
backend.Improved logging while establishing connections to
sparklyr
.Improved
spark_connect()
performance.Implemented new configuration checks to proactively report connection errors in Windows.
While connecting to spark from Windows, setting the
sparklyr.verbose
option toTRUE
prints detailed configuration steps.Added
custom_headers
tolivy_config()
to add custom headers to the REST call to the Livy server
Compilation
Added support for
jar_dep
in the compilation specification to support additionaljars
throughspark_compile()
.spark_compile()
now prints deprecation warnings.Added
download_scalac()
to assist downloading all the Scala compilers required to build usingcompile_package_jars
and provided support for using anyscalac
minor versions while looking for the right compiler.
Backend
- Improved backend logging by adding type and session id prefix.
Miscellaneous
copy_to()
andsdf_copy_to()
auto generate aname
when an expression can’t be transformed into a table name.Implemented
type_sum.jobj()
(from tibble) to enable better printing of jobj objects embedded in data frames.Added the
spark_home_set()
function, to help facilitate the setting of theSPARK_HOME
environment variable. This should prove useful in teaching environments, when teaching the basics of Spark and sparklyr.Added support for the
sparklyr.ui.connections
option, which adds additional connection options into the new connections dialog. Therstudio.spark.connections
option is now deprecated.Implemented the “New Connection Dialog” as a Shiny application to be able to support newer versions of RStudio that deprecate current connections UI.
Bug Fixes
When using
spark_connect()
in local clusters, it validates thatjava
exists underJAVA_HOME
to help troubleshoot systems that have an incorrectJAVA_HOME
.Improved
argument is of length zero
error triggered while retrieving data with no columns to display.Fixed
Path does not exist
referencinghdfs
exception duringcopy_to
under systems configured withHADOOP_HOME
.Fixed session crash after “No status is returned” error by terminating invalid connection and added support to print log trace during this error.
compute()
now caches data in memory by default. To revert this beavior usesparklyr.dplyr.compute.nocache
set toTRUE
.spark_connect()
withmaster = "local"
and a givenversion
overridesSPARK_HOME
to avoid existing installation mismatches.Fixed
spark_connect()
under Windows issue whennewInstance0
is present in the logs.Fixed collecting
long
type columns when NAs are present (#463).Fixed backend issue that affects systems where
localhost
does not resolve properly to the loopback address.Fixed issue collecting data frames containing newlines
\n
.Spark Null objects (objects of class NullType) discovered within numeric vectors are now collected as NAs, rather than lists of NAs.
Fixed warning while connecting with livy and improved 401 message.
Fixed issue in
spark_read_parquet()
and other read methods in whichspark_normalize_path()
would not work in some platforms while loading data using custom protocols likes3n://
for Amazon S3.Resolved issue in
spark_save()
/load_table()
to support saving / loading data and added path parameter inspark_load_table()
for consistency with other functions.
Sparklyr 0.5.5
- Implemented support for
connectionViewer
interface required in RStudio 1.1 andspark_connect
withmode="databricks"
.
Sparklyr 0.5.4
- Implemented support for
dplyr 0.6
and Spark 2.1.x.
Sparklyr 0.5.3
- Implemented support for
DBI 0.6
.
Sparklyr 0.5.2
Fix to
spark_connect
affecting Windows users and Spark 1.6.x.Fix to Livy connections which would cause connections to fail while connection is on ‘waiting’ state.
Sparklyr 0.5.0
Implemented basic authorization for Livy connections using
livy_config_auth()
.Added support to specify additional
spark-submit
parameters using thesparklyr.shell.args
environment variable.Renamed
sdf_load()
andsdf_save()
tospark_read()
andspark_write()
for consistency.The functions
tbl_cache()
andtbl_uncache()
can now be using without requiring thedplyr
namespace to be loaded.spark_read_csv(..., columns = <...>, header = FALSE)
should now work as expected – previously,sparklyr
would still attempt to normalize the column names provided.Support to configure Livy using the
livy.
prefix in theconfig.yml
file.Implemented experimental support for Livy through:
livy_install()
,livy_service_start()
,livy_service_stop()
andspark_connect(method = "livy")
.The
ml
routines now acceptdata
as an optional argument, to support calls of the form e.g.ml_linear_regression(y ~ x, data = data)
. This should be especially helpful in conjunction withdplyr::do()
.Spark
DenseVector
andSparseVector
objects are now deserialized as R numeric vectors, rather than Spark objects. This should make it easier to work with the output produced bysdf_predict()
with Random Forest models, for example.Implemented
dim.tbl_spark()
. This should ensure thatdim()
,nrow()
andncol()
all produce the expected result withtbl_spark
s.Improved Spark 2.0 installation in Windows by creating
spark-defaults.conf
and configuringspark.sql.warehouse.dir
.Embedded Apache Spark package dependencies to avoid requiring internet connectivity while connecting for the first through
spark_connect
. Thesparklyr.csv.embedded
config setting was added to configure a regular expression to match Spark versions where the embedded package is deployed.Increased exception callstack and message length to include full error details when an exception is thrown in Spark.
Improved validation of supported Java versions.
The
spark_read_csv()
function now accepts theinfer_schema
parameter, controlling whether the columns schema should be inferred from the underlying file itself. Disabling this should improve performance when the schema is known beforehand.Added a
do_.tbl_spark
implementation, allowing for the execution ofdplyr::do
statements on Spark DataFrames. Currently, the computation is performed in serial across the different groups specified on the Spark DataFrame; in the future we hope to explore a parallel implementation. Note thatdo_
always returns atbl_df
rather than atbl_spark
, as the objects produced within ado_
query may not necessarily be Spark objects.Improved errors, warnings and fallbacks for unsupported Spark versions.
sparklyr
now defaults totar = "internal"
in its calls tountar()
. This should help resolve issues some Windows users have seen related to an inability to connect to Spark, which ultimately were caused by a lack of permissions on the Spark installation.Resolved an issue where
copy_to()
and other R => Spark data transfer functions could fail when the last column contained missing / empty values. (#265)Added
sdf_persist()
as a wrapper to the Spark DataFramepersist()
API.Resolved an issue where
predict()
could produce results in the wrong order for large Spark DataFrames.Implemented support for
na.action
with the various Spark ML routines. The value ofgetOption("na.action")
is used by default. Users can customize thena.action
argument through theml.options
object accepted by all ML routines.On Windows, long paths, and paths containing spaces, are now supported within calls to
spark_connect()
.The
lag()
window function now accepts numeric values forn
. Previously, only integer values were accepted. (#249)Added support to configure Ppark environment variables using
spark.env.*
config.Added support for the
Tokenizer
andRegexTokenizer
feature transformers. These are exported as theft_tokenizer()
andft_regex_tokenizer()
functions.Resolved an issue where attempting to call
copy_to()
with an Rdata.frame
containing many columns could fail with a Java StackOverflow. (#244)Resolved an issue where attempting to call
collect()
on a Spark DataFrame containing many columns could produce the wrong result. (#242)Added support to parameterize network timeouts using the
sparklyr.backend.timeout
,sparklyr.gateway.start.timeout
andsparklyr.gateway.connect.timeout
config settings.Improved logging while establishing connections to
sparklyr
.Added
sparklyr.gateway.port
andsparklyr.gateway.address
as config settings.The
spark_log()
function now accepts thefilter
parameter. This can be used to filter entries within the Spark log.Increased network timeout for
sparklyr.backend.timeout
.Moved
spark.jars.default
setting from options to Spark config.sparklyr
now properly respects the Hive metastore directory with thesdf_save_table()
andsdf_load_table()
APIs for Spark < 2.0.0.Added
sdf_quantile()
as a means of computing (approximate) quantiles for a column of a Spark DataFrame.Added support for
n_distinct(...)
within thedplyr
interface, based on call to Hive functioncount(DISTINCT ...)
. (#220)
Sparklyr 0.4.0
- First release to CRAN.