ESGF Installer Known Issues

Installer Known Issues

This is a list of know issues that might be encountered after an ESGF 2.X installation or upgrade:

v2.8.1

• OpenIds from esgf-node.llnl.gov (namely the LLNL selection) are not recognized in the esg-orp on a datanode. An old version of /esg/config/esgf_idp_static.xml is included in this release that has the old domain name for llnl. Solution: Please clone https://github.com/ESGF/config and copy the file from esgf-prod.

v2.6.9

• The publisher doesn't give any output, but is actually running. This is due to a correction to the logging settings to allow for INFO level messages to be suppressed. The log_level setting in esg.ini under [DEFAULT] should be changed to INFO in order to continue with messages.

v2.6.8c

• lxml ImportError for libiconv.

(esgf-pub) # conda install -c conda-forge libiconv

• If you have a KeyError with the cmip6_handler then you'll need to manually upgrade correct both cdms and cmor installations. First You can test the cause of the KeyError with this:

(esgf-pub) $ python -c "import esgcet.config.cmip6_handler"

If there's this error:

ImportError: libnetcdf.so.11: cannot open shared object file: No such file or directory

Do the following:

(esgf-pub) $conda remove cdms2 cdtime cmor cdutil

(esgf-pub) $ conda install -c conda-forge cmor cdms2 cdtime

v2.6.7

• The issue with the esgf_dashboard schema listed under v2.5.13-16 remains unchanged. Refer to the fix there.

• Some users noted a libssh2 error while attempting to publish data; this is due to manually sourcing the /etc/esg.env file which is now deprecated. Please activate the conda virtual-env by executing source /usr/local/conda/bin/activate esgf-pub instead, after initializing a "clean" shell environment.

• The latest version of the Publisher requires trust-certificates to be present in $HOME/.globus/certificates directory. If this directory exists, remove it. When you are sure there is no directory called 'certificates' under $HOME/.globus, create a symbolic link called certificates, under $HOME/.globus, and point it to /etc/grid-security/certificates

• Some users noted a publication-time error about missing 'hessian_service_certificates_location'. To fix this, place the following line in your /esg/config/esgcet/esg.ini file: hessian_service_certs_location = %(home)s/.globus/certificates

• "yum clean all" will break the esgf.repo file used by the installer. This becomes an issue after trying to install on a node following a purge.

v2.5.13-16

• Error encountered during installation of esgf_dashboard schema.
  • You will need to manually remove the old schema first:

psql -U dbsuper esgcet # enter your database password

drop schema esgf_dashboard cascade;

delete from esgf_migrate_version where repository_id like 'ESGF Dashboard%';

drop LANGUAGE plpgsql;

• Retry the installation

• esgprep appears to hang during mapfile generation (as root) source /usr/local/conda/bin/activate esgf-pub pip install esgprep==2.7.9

• After installing the latest 2.5.14 release, and fixing the ORP problem, data download from the TDS would not work: the TDS kept asking for username/password through the basic authentication pop-up dialog.

The problem can be solved by editing the TDS application configuration file:

/usr/local/tomcat/webapps/thredds/WEB-INF/applicationContext.xml

commenting out this section:

<bean id="restrictedDatasetAuthorizer" class="thredds.servlet.restrict.TomcatAuthorizer"> <property name="useSSL" value="false"/> <property name="sslPort" value="8443"/> </bean>

and uncommenting this section:

<bean id="restrictedDatasetAuthorizer" class="esg.orp.app.tds.TDSAuthorizer" />

• v2.4.24 ez_setup deprecated. pip won't install as a prerequisite for the esgpublisher installation. We will work on a fix for this. Also apache_frontend is impacted but a correction may already be in place. [this appears to be corrected]

• v2.3.8 - CoG doesn't load (500 error, sanitizer error reported in httpd error_log)

  Solution: manually upgrade CoG
  See instructions here: https://acme-climate.atlassian.net/wiki/display/ESGF/Manual+Installation+of+ESGF+Software+Components

• For compute nodes: Issue with LAS in 2.3.8. LAS is broken after installation due to tomcat upgrade. The fix is the following:

rm -Rf /usr/local/tomcat/content/las
ln -s /esg/content/las /usr/local/tomcat/content/las
esg-node restart

• On freshly installed datanodes, your esg.ini file may contain globus:#DEFAULTENDPONTNAME#

  • #DEFAULTENDPONTNAME# needs to be replaced with the UUID for your endpoint (assigned automatically to the gridftp server instance)
  • You can log into globus.org and find the endpoint associated with your node's FQDN. The UUID should be listed in a the endpoints properties page
  • or you can remove the entry from esg.ini altogether (not recommended) but globus urls can be added to existing publications later (once an endpoint has been established)

• Publishing: upgrading to TLS v1.2. If your site requires TLS v1.2 to be configured in apache, then you also will need to modify the publisher. With sudo privs, edit /usr/local/uvcdat/2.2.0/lib/python2.7/site-packages/esgcet-3.0.1-py2.7.egg/esgcet/publish/thredds.py. Change ssl.PROTOCOL_TLSv1 to ssl.PROTOCOL_TLSv1_2. For older versions of the publisher, find the latest esgcet-2.X.X instead of 3.0.1.

---

Also, check out the FAQ: https://github.com/ESGF/esgf.github.io/wiki/ESGFNode|FAQ

---