From OPeNDAP Documentation

<< back to HowTo Guides

How to configure an Amazon Linux AMI for EC2 to build Hyrax

This describes how to set up a vanilla Amazon Linux AMI virtual machine so that we can build the Hyrax data server.

Amazon Linux AMI versions

These instructions have been amended to cover:

  • Amazon Linux AMI 2012.09

Initial configuration

Set up general build stuff that you need:

  • yum install java-1.6.0-openjdk java-1.6.0-openjdk-devel
  • yum install junit
  • yum install ant
  • yum install ant-junit.noarch
  • yum install make
  • yum install subversion
  • yum install gcc-c++
  • yum install flex
  • yum install bison
  • yum install curl-devel
  • yum install libxml2-devel
  • yum install libjpeg-devel
  • yum install zlib-devel
  • yum install readline-devel
  • yum install libuuid-devel
  • yum install openssl-devel

Other dependencies

  • yum install libicu-devel


  • download the latest versions of autoconf, automake and libtool
  • ./configure, make, make install

Building Hyrax from the shrew project

In a (ba)sh shell:

Check out the shrew project
svn co
Change your working directory to the shrew directory.
cd shrew
Copy Tomcat from src/dependencies/downloads and expand the tar file
cp src/dependencies/downloads/apache-tomcat-*.tar.gz .
tar -xzf apache-tomcat-*
Source the spath file
. spath
Build the dependancies
cd src/dependencies

Note There's another way to handle the 'dependencies' needed to build the server and that is to get the software you need from yum. The problem you'll run into is that most of these 'scientific data format' packages are not in yum proper and you need to configure an additional yum repository. That's not that bad, and it has the advantage of basing the build on 'stock' software. The word 'stock' is in quotes because those RPMs come from a somewhat non-stock repo... Be that as it may, see How to configure CentOS for the production of RPM binaries for the low-down on that option.

Using the shrew top level Makefile

Change your working directory to the shrew directory.
cd $prefix
Copy the to
Edit the to be certain that the various dependancies are correctly configured.
Since I am trying to get a new package up I edited this file, mostly I think you should just be able to make the copy and proceed. However, be sure to check the versions/directories named in the --with... options listed at teh start of the with the actual directories they reference in $prefix/deps. Mismatches will break the build, especially down in the handlers where most of those dependencies are used.
Engage the autotools process and see how far you get...
autoreconf -vif
./configure --prefix=$prefix
make world

Note Adding -j N where N is an integer that is the number of processors on your host + 1 will enable a parallel build. You'll be happy about that if you have more than one processor...

If the above fails then you need to build each Hyrax component by hand.

Note But all hope for an easy time is not lost. If the BES and libdap built but the handlers fail, it's often because one of the handlers' configure script failed and that's often because it could not find a dependency (e.g., hdf4 didn't find the correct hdf4 library). There are two targets you should know about: modules and modules-configure. The latter runs configure for all of the modules. If the build fails running configure for one of the modules (which is fairly obvious because of the messages that include the word 'configure'), then look for the name of thing that failed. Then looks at the configure params in the and see that the name of the package matches the name of the dependency in $prefix/deps. For example, if $prefix/deps has hdf-4.2.8 but the says hdf4_arg = --with-hdf4=$(prefix)/deps/hdf-4.2.10 the target modules-configure is going to fail when it goes to configure the hdf4 handler. Edit the line so that instead of .../hdf-4.2.10 it matches what's in $prefix/deps, hdf-4.2.8 in this example. Once hacked, you can run make modules-configure, make modules just like that. Since the world target stopped, you need to run the tests by hand. Either read the to see how or run make daemon-check. Note that you'll also need to build the OLFS using its targets if you go this route. Using make olfs should work.

Building the OLFS

Some targets in the also build the OLFS, but the world target, which otherwise builds the whole server, does not. However, ant was already loaded using the RPMs, so the OLFS build is easy:

Build and install the OLFS
cd src/olfs
ant server
cp build/dist/opendap/.war $prefix/apache-tomcat-7.0.29/webaps
cd $prefix

Start/Test the server

Start and Test the server
besctl start

In a web browser, you should see the tomcat page at http://localhost:8080/ and the Hyrax initial page at http://localhost:8080/opendap/

Building by hand

If the make world target fails and work-around in the Note doesn't work, here's how to build the BES piece by piece. It's actually pretty simple, but involves repetition and baby-sitting the build machine for the whole process.

Build and install libdap
cd $prefix/src/libdap
autoreconf -vif
./configure --prefix=$prefix
make install
Build and install the BES
cd $prefix/src/bes
autoreconf -vif
./configure --prefix=$prefix
make install
Build and install the dap-server module
cd $prefix/src/modules/dap-server
autoreconf -vif
./configure --prefix=$prefix
make install
Build and install the csv_handler module
cd $prefix/src/modules/csv_handler
autoreconf -vif
./configure --prefix=$prefix
make install
Build and install the fileout_gdal module
cd $prefix/src/modules/fileout_gdal
autoreconf -vif
./configure --prefix=$prefix --with-gdal=$prefix/deps/gdal-12.15.12
make install
Build and install the fileout_netcdf module
cd $prefix/src/modules/fileout_netcdf
autoreconf -vif
./configure --prefix=$prefix --with-netcdf=$prefix/deps/netcdf-4.1.2
make install
Build and install the fits_handler module
cd $prefix/src/modules/fits_handler
autoreconf -vif
./configure --prefix=$prefix --with-cfits=$prefix/deps/cfitsio
make install
Build and install the freeform_handler module
cd $prefix/src/modules/freeform_handler
autoreconf -vif
./configure --prefix=$prefix
make install
Build and install the gateway_module module
cd $prefix/src/modules/gateway_module
autoreconf -vif
./configure --prefix=$prefix
make install
Build and install the gdal_handler module
cd $prefix/src/modules/fileout_netcdf
autoreconf -vif
./configure --prefix=$prefix --with-gdal=$prefix/deps/gdal-12.15.12
make install
Build and install the hdf4_handler module
cd $prefix/src/modules/hdf4_handler
autoreconf -vif
./configure --prefix=$prefix --with-hdf4=$prefix/deps/hdf-4.2.8
make install
Build and install the hdf5_handler module
cd $prefix/src/modules/hdf5_handler
autoreconf -vif
./configure --prefix=$prefix --with-hdf5=$prefix/deps/hdf5-1.8.6
make install
Build and install the ncml_module module
cd $prefix/src/modules/ncml_module
autoreconf -vif
./configure --prefix=$prefix
make install
Build and install the netcdf_handler module
cd $prefix/src/modules/fileout_netcdf
autoreconf -vif
./configure --prefix=$prefix --with-netcdf=$prefix/deps/netcdf-4.1.2
make install
Build and install the xml_data_handler module
cd $prefix/src/modules/xml_data_handler
autoreconf -vif
./configure --prefix=$prefix
make install
Build and install the ugrid_functions module
cd $prefix/src/modules/ugrid_functions
autoreconf -vif
./configure --prefix=$prefix --with-gridfields=$prefix/deps/gridfields-1.0.1
make install

Building the ncml_handler when incompatible versions of ICU are resident on the system

Is there a system ICU?

Check to see what's on the machine in terms of ICU:

[root@ip-10-152-156-244 shrew]# rpm -qa | grep icu
[root@ip-10-152-156-244 shrew]#

(no need to be root)

So we can see that icu is there but it's the version 4 code and that is massively incompatible with the version 3 code. You probably knew all of that, I think we talked about it, but I just thought I'd show the useful 'rpm -qa' (query all) command.

Where for art thou, ICU files?

Here's how to see where stuff is:

[ec2-user@ip-10-152-156-244 shrew]$ rpm -qil libicu-devel-4.2.1-9.9.amzn1.x86_64Name  : libicu-devel Relocations: (not relocatable)
Version  : 4.2.1 Vendor:
Release  : 9.9.amzn1 Build Date: Mon 09 Jan 2012 04:52:37 PM UTC
Install Date: Wed 30 Jan 2013 05:39:05 AM UTC Build Host:
Group  : Development/Libraries Source RPM: icu-4.2.1-9.9.amzn1.src.rpm
Description :
Includes and definitions for developing with icu.

So if configure looks for icu-config, it'll find the one for icu 4.x which our handler cannot use - at least not without some tweaking things. I think the includes are all different (nice, eh?)

Make the build find the ICU in our dependencies

Put our icu-3.6/bin dir on the PATH in front of everything:

[ec2-user@ip-10-152-156-244 shrew]$ export PATH=/home/ec2-user/opendap/shrew/deps/icu-3.6/bin:$PATH

And configure in the ncml_handler should work just fine

But when you try to start the server, massive library not found crap. That's because on linux you have to run ldconfig for a library to be found 'automatically'. As a convenience to users, icu left this step out…

Make the dynamic library load find our depencencies ICU at runtime

Add our icu-3.6/lib dir to LD_LIBRARY_PATH

[root@ip-10-152-156-244 shrew]# export LD_LIBRARY_PATH=/home/ec2-user/opendap/shrew/deps/icu-3.6/lib/
[root@ip-10-152-156-244 shrew]# besctl start
Starting the BES
OK: Successfully started the BES
PID: 11975 UID: 0
[root@ip-10-152-156-244 shrew]#

Start tomcat and the server is up!

Why we don't have this these issues on CentOS is likely that they use icu-3.6 and even if we build our own version, it's not being used. In fact, the reason we used 3.6 and not 4.2 was that most linux distros came with 3.6.