Taylor, Ronald C
2017-Mar-14 21:53 UTC
[R] problem with sparkl_connect() in the sparklyr package for parallelizing R in the Spark environment - "Gateway in port (8880) did not respond"
Hi folks,
I posted the message below as a new issue on the sparklyr web page at github
over a week ago, but have not gotten any reply back. So I am posting here, in
the hope somebody on this list can provide guidance. I really want to get R
working in Spark on our local Linux cluster. Eager to get going, if I can get
out of the gate. But have to get past this spark_connect() problem. Please see
the below.
- Ron Taylor
%%%%%%%%%%%%%
problem with spark_connect() using sparklyr on a Cloudera CDH 5.10.0 Hadoop
cluster #534
rtaylor24<https://github.com/rtaylor24> commented 8 days
ago<https://github.com/rstudio/sparklyr/issues/534>
Hello folks,
I am trying to use sparklyr for the first time on a Hadoop cluster. I have used
R with sparklyr on a "local" copy of Spark on my Mac laptop, but this
is the first time that I am trying to run it as a "yarn-client" on a
true cluster, to actually get some parallelization out of sparklyr use.
We have a small Linux cluster at our lab running Cloudera CDH 5.10.0. When I try
to do the spark_connect() from an R session started on a command line on the
Hadoop cluster's name (master) node, I get the same msg as in an earlier
CLOSED issue. That is, my error msg is:
"Failed while connecting to sparklyr to port (8880) for sessionid (2423):
Gateway in port (8880) did not respond."
I am thus reopening that issue here, since I still need help even after reading
that older issue (#394<https://github.com/rstudio/sparklyr/issues/394>).
At bottom is the record of my R session on the Hadoop cluster's name node,
with all the details that I can think of printed out to the screen.
I note that the version of Spark used by CDH is 1.6.0, which is different than
what is in spark_home_dir (1.6.2). I cannot seem to change the spark_home_dir by
setting SPARK_HOME to the Spark location used by the CDH distribution.
spark_home_dir does not get altered by my setting of SPARK_HOME (as you can see
below). So one question (perhaps the critical question?) is: how do I force
sparklyr to connect to the Spark version being used by the CDH distribution?
As you can see at the Cloudera web page at
https://www.cloudera.com/documentation/enterprise/release-notes/topics/cdh_vd_cdh_package_tarball_510.html
the 5.10.0 distribution has Spark 1.6.0, not 1.6.2. So I am trying to tell
sparklyr code to use the Spark 1.6.0 distribution that is located here:
/opt/cloudera/parcels/CDH-5.10.0-1.cdh5.10.0.p0.41/bin/spark-submit
/opt/cloudera/parcels/CDH-5.10.0-1.cdh5.10.0.p0.41/bin/spark-shell
/opt/cloudera/parcels/CDH-5.10.0-1.cdh5.10.0.p0.41/lib/spark
and so I was trying to set SPARK_HOME as follows:
Sys.setenv(SPARK_HOME =
"/opt/cloudera/parcels/CDH-5.10.0-1.cdh5.10.0.p0.41")
Sys.getenv("SPARK_HOME")
[1] "/opt/cloudera/parcels/CDH-5.10.0-1.cdh5.10.0.p0.41"
However! I note that part of the error msg (see bottom) says that the correct
path was used to spark-submit:
"Path:
/opt/cloudera/parcels/CDH-5.10.0-1.cdh5.10.0.p0.41/bin/spark-submit"
So maybe sparklyr is indeed accessing the Spark 1.6.0 distribution as it should
in the cluster, and the problem lies elsewhere??
One other note: there was an earlier version of sparklyr installed by support
here on the Hadoop name node. I have bypassed that, installed the latest version
of sparklyr (0.5.1) into
/people/rtaylor/Rpackages
as you can see below.
Would very much appreciate some guidance to get me over this initial hurdle.
* Ron Taylor
Pacific Northwest National Laboratory
email: ronald.taylor at pnnl.gov<mailto:ronald.taylor at pnnl.gov>
%%%%%%%%%%%%%%
screen output from my failed run:
[rtaylor at bigdatann Rwork]$ R
R version 3.3.2 (2016-10-31) -- "Sincere Pumpkin Patch"
Copyright (C) 2016 The R Foundation for Statistical Computing
Platform: x86_64-pc-linux-gnu (64-bit)
'help.start()' for an HTML browser interface to help.
Type 'q()' to quit R.
ls()
character(0)
(.packages())
[1] "stats" "graphics" "grDevices"
"utils" "datasets" "methods"
[7] "base"
install.packages("sparklyr",
lib="/people/rtaylor/Rpackages/")
--- Please select a CRAN mirror for use in this session ---
trying URL
'https://cran.cnr.berkeley.edu/src/contrib/sparklyr_0.5.2.tar.gz'
Content type 'application/x-gzip' length 732806 bytes (715 KB)
=================================================downloaded 715 KB
* installing source package 'sparklyr' ...
** package 'sparklyr' successfully unpacked and MD5 sums checked
** R
** inst
** preparing package for lazy loading
** help
*** installing help indices
** building package indices
** testing if installed package can be loaded
* DONE (sparklyr)
The downloaded source packages are in
'/tmp/RtmpQUB4IE/downloaded_packages'
library(sparklyr, lib.loc="/people/rtaylor/Rpackages/")
(.packages())
[1] "sparklyr" "stats" "graphics"
"grDevices" "utils" "datasets"
[7] "methods" "base"
sessionInfo()
R version 3.3.2 (2016-10-31)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Red Hat Enterprise Linux Workstation release 6.4 (Santiago)
locale:
[1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
[3] LC_TIME=en_US.UTF-8 LC_COLLATE=en_US.UTF-8
[5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8
[7] LC_PAPER=en_US.UTF-8 LC_NAME=C
[9] LC_ADDRESS=C LC_TELEPHONE=C
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
attached base packages:
[1] stats graphics grDevices utils datasets methods base
other attached packages:
[1] sparklyr_0.5.1
loaded via a namespace (and not attached):
[1] Rcpp_0.12.9 withr_1.0.2 digest_0.6.12 dplyr_0.5.0
[5] rprojroot_1.2 assertthat_0.1 rappdirs_0.3.1 R6_2.2.0
[9] jsonlite_1.2 DBI_0.5-1 backports_1.0.5 magrittr_1.5
[13] httr_1.2.1 config_0.2 tools_3.3.2 parallel_3.3.2
[17] yaml_2.1.14 base64enc_0.1-3 tcltk_3.3.2 tibble_1.2
Sys.getenv("JAVA_HOME")
[1] "/usr/java/latest"
spark_installed_versions()
spark hadoop dir
1 1.6.2 2.6 spark-1.6.2-bin-hadoop2.6
spark_home_dir()
[1] "/people/rtaylor/.cache/spark/spark-1.6.2-bin-hadoop2.6"
R.home(component = "home")
[1] "/share/apps/R/3.3.2/lib64/R"
path.expand("~")
[1] "/people/rtaylor"
Sys.setenv(SPARK_HOME =
"/opt/cloudera/parcels/CDH-5.10.0-1.cdh5.10.0.p0.41")
Sys.getenv("SPARK_HOME")
[1] "/opt/cloudera/parcels/CDH-5.10.0-1.cdh5.10.0.p0.41"
spark_home_dir()
[1] "/people/rtaylor/.cache/spark/spark-1.6.2-bin-hadoop2.6"
ls()
character(0)
config <- spark_config()
ls()
[1] "config"
sc <- spark_connect(master = "yarn-client", config = config,
version = "1.6.0")
Error in force(code) :
Failed while connecting to sparklyr to port (8880) for sessionid (2423): Gateway
in port (8880) did not respond.
Path: /opt/cloudera/parcels/CDH-5.10.0-1.cdh5.10.0.p0.41/bin/spark-submit
Parameters: --class, sparklyr.Backend, --jars,
'/share/apps/R/3.3.2/lib64/R/library/sparklyr/java/spark-csv_2.11-1.3.0.jar','/share/apps/R/3.3.2/lib64/R/library/sparklyr/java/commons-csv-1.1.jar','/share/apps/R/3.3.2/lib64/R/library/sparklyr/java/univocity-parsers-1.5.1.jar',
'/share/apps/R/3.3.2/lib64/R/library/sparklyr/java/sparklyr-1.6-2.10.jar',
8880, 2423
---- Output Log ----
/opt/cloudera/parcels/CDH-5.10.0-1.cdh5.10.0.p0.41/bin/../lib/spark/bin/spark-submit:
line 27: /opt/cloudera/parcels/CDH-5.10.0-1.cdh5.10.0.p0.41/bin/spark-class: No
such file or directory
/opt/cloudera/parcels/CDH-5.10.0-1.cdh5.10.0.p0.41/bin/../lib/spark/bin/spark-submit:
line 27: exec:
/opt/cloudera/parcels/CDH-5.10.0-1.cdh5.10.0.p0.41/bin/spark-class: cannot
execute: No such file or directory
---- Error Log ----
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
[[alternative HTML version deleted]]