Dear ROC, CIC, site managers, and otherwise affected friends,
this is the release notification for the long awaited LCG-2_6_0 release.
---> PLEASE READ THE SECTION ABOUT THE BDII <---
The release has been given 2 weeks ago to the ROCs in Italy, UK/IE, and
SE to do a deployment test. During the following week we received many
problem reports. In addition some application specific testing was
going on which, as expected, uncovered some more issues. This was a
very useful step in the process. Finding solutions for all the
reported problems took the better part of this week.
There is no longer a deadline for upgrading. The ROC managers will
coordinate the upgrade in their region. Please clarify with them what
the plan for moving to LCG-2_6_0 is.
Sites that are at 2_5_0 should update, because there are several
improvements in this release.
We wish you a smooth and successful upgrade
the release preparation team
Read this if you opt to ignore all documentation:
=================================================
The glue schema has changed to GLUE-1.2.
(http://infnforge.cnaf.infn.it/glueinfomodel/index.php/Spec/V12)
This affects ALL sites, including those that opt not to upgrade in the
near future but run their own top-level BDII for their UIs, WNs, and RBs
at their site.
A LCG-2_4_0 BDII can't see the information that is published by a
LCG-2_6_0 site. As more and more sites upgrade to 2.6.0, the resources
advertised by a 2.4.0 top-level BDII will disappear.
If you use your own top-level BDII, there are 3 solutions:
----------------------------------------------------------
1) Upgrade only your BDII. This is a simple operation and is
recommended. Please see this document
http://grid-deployment.web.cern.ch/grid-deployment/documentation/
LCG2-Manual-Upgrade/
2) Manually update your WNs, UIs and RB to reference a 2.6.0 BDII, such
as lcg-bdii.cern.ch.
3) If you use yaim, update BDII_HOST in your site-info.def to a 2.6.0
machine and rerun the yaim configuration scripts on your WNs, UIs
and
the RB. Use the version of yaim distributed with your middleware, eg
lcg-yaim-2.4.0.
Note that 'site BDIIs' (usually found on a CE) do not have
to be upgraded.
Sites that upgrade to LCG-2_6_0:
================================
If you upgrade any node that provides information for the information
system (BDII) make sure that you FIRST upgrade the elements higher in
the hierarchy. First the BDII, then the Site BDII (CE) and at the end
the other service nodes.
The schema is backward compatible, but a pre 1.2 element will ignore
entries that use the new schema.
Do not configure the FTS clients on your T1 or T2 center unless you are
participating in the SC3 activity and you know who your T0/T1 center is!
If in doubt, have a look at the SC3 page.
FTS services are configured with direct help by the SC3 team at CERN.
DO NOT INSTALL A FTS SERVER UNLESS YOU ARE A T1 IN SC3
Make sure you understand how to configure the LFC as a local catalogue
or as a replica catalogue and that you know what you want to setup.
Only install one of the new VO-boxes after you have clarified with the
VO that you want to support on it that they will use this as a pilot
service. Documentation on how to use this node will be provided shortly
via the GOC-wiki pages.
Upgrades paths that are supported:
----------------------------------
2_4_0 -> 2_6_0
and installation from scratch.
Since 2_5_0 is very close to 2_4_0 we are quite confident that this
works too.
If your site is on 2_5_0 you still have to upgrade because 2_6_0
contains major changes.
Where can I find the link to the new release?
---------------------------------------------
here: http://lcg.web.cern.ch/LCG/Sites/releases.html
Now some more details
=====================
GLUE 1.2
--------
The new version of the Glue Schema is backwards compatible although the
top level BDIIs need to be upgraded first. If a site update to the new
information providers before the top level BDII has been upgrade the
site will be rejected by the BDII.
This affects ALL nodes that provide data for the information system. In
addition several of the client tools to access the information system
have been adapted, like the popular lcg-infosites.
Details on the new schema can be seen here:
http://infnforge.cnaf.infn.it/glueinfomodel/index.php/Spec/V12
LFC 1.3.7
---------
The LFC has now two different usages. As a central replica catalogue
and as a local file catalogue. Make sure you don't get confused when
setting up a node. Depending if you install a local or a central LFC,
the information to publish in the Information System is different.
You can set LFC_TYPE in yaim and it will configure the box
appropriately.
It nows supports sessions, and a Python interface is provided as a
separate RPM. In addition the DLI interface is supported.
The LFC Administrator's Guide is available at:
http://goc.grid.sinica.edu.tw/gocwiki/How_to_set_up_an_LFC_service
The archived LFC support mailing list to contact is:
[log in to unmask]
The release includes YAIM install-scripts for use with MySQL backends.
For the ORACLE backend, refer to the LFC Administrator's Guide. For
both backends, meta-rpms are provided.
The LFC service is relevant for all those sites that either host and
LFC for VOs, or provide the service of a local LFC catalogue.
DPM 1.3.7
---------
The DPM Administrator's Guide is available at:
http://goc.grid.sinica.edu.tw/gocwiki/
How_to_install_the_Disk_Pool_Manager_%28DPM%29
The archived DPM support mailing list to contact is:
[log in to unmask]
The DPM includes secure rfio, and a DPM-enabled standard GridFTP server.
For SEs based on DPM, YAIM installation and configuration scripts are
provided for the MySQL based version. For the ORACLE backend version,
refer to the DPM Administrator's Guide. For both backends, meta-rpms are
provided.
The minimal configuration, where the disk server and the DPM servers are
located on the same machine, is supported by YAIM.
Yaim also provides SE_dpm_disk for a separate DPM disk server.
Bugs fixed : #9293, #9393, #9510
VO-BOX 1.0.0-5
--------------
This is a pilot version of the node type that allows VOs to run their
own user level agents. VO's software managers can use their
certificates to get interactive access to these nodes and install the
agents. Mechanisms for proxy renewal are provided. This uses gsi-ssh.
We don't suggest that sites set these nodes up immediately unless a VO
approaches them directly.
Documentation on the usage will be provided at the beginning of the
week on the GOC-wiki page.
FTS
---
The clients are part of the UIs and WNs and supported by YAIM scripts.
The server installation is described on the service challenge WIKI page:
https://uimon.cern.ch/twiki/bin/view/LCG/LCGServiceChallenges
See the section "Read this if you opt to ignore all documentation:" for
details on who should install/configure this.
New and updated software
========================
Data Management Client:
-----------------------
lcg_util 1.3.1-1
lcg-gfal 1.7.3
lcg-info 1.0.11
There are now LCG data management libs with timeouts. These should make
it easier to handle the all to frequent problems with hanging calls.
Lcg_util and GFAL have been modified to accept a timeout parameter (-t
SECS). This is used to timeout both gridftp transfers and interactions
with the SRM. CTRL-C is now caught during gridftp transfers too, and the
transfer is cleaned up properly.
This requires an update of the WNs and UIs.
VDT 1.2.2:
----------
Updated because of gLite dependency. Globus identical to VDT 1.2.0.
Workload Management 1.6.8:
--------------------------
bug #5549 NS cannot handle being addressed through RB host alias
bug #6295 RB problem if the OutputData attribut is too big
bug #6653 LogMonitor can abort on removing job directory
bug #7902 memory leak in edg-wl-ns_daemon
bug #7611 NS can get into state where all connections fail
bug #7923 voms authentication error with edg-job-list-matchand
edg-job-submit
bug #7965 WL startup scripts must not implicitly refer to /opt/edg
bug #7974 LD_LIB_PATH too long with tar installation
bug #8821 LB locallogger endianess patch has not been applied to LCG
version
plus some changes, currently without bug number:
* add result caching to brokerinfoGlueImpl::retrieveCloseSEsInfo
(), to
reduce the number of information system queries
* generalise inputDataType setting for DLI
* add max_rows attrib to logging/etc/server.sql
Information System:
-------------------
BDII now supports the new GLUE 1.2 schema and no longer uses iptables.
The information providers also publish GLUE 1.2 information.
The RB now has a dynamic plug-in that gets information about the load.
dCache:
-------
The configuration contains several fixes and enhancements provided by
GridPP
R-GMA:
------
The new gLite 1.2 version of R-GMA. This is setup in
a compatibility mode offering secure and in-secure services.
Monitoring:
-----------
Significant new or improved components are now available.
There is a sensor for gridFTP logfiles that allows to monitor the
transfers from a globus gridFTP server There is a new sensor for the job
state found in the LB server on the RB. There is a new sensor that
monitors the resource usage of running jobs. All of the above are based
on R-GMA.
In addition monitoring for stdout and error while jobs are running is
now available. Documentation on the usage will be provided shortly on
the GOC-Wiki
Accounting:
-----------
Updated to work with the gLite version of R-GMA and now supports LSF.
VOMS:
-----
Version 1.5.4 is compatible with new gLite and old 1.3.7 VOMS servers.
YAIM:
-----
Bug fixes, new services and adaptation to the new glue schema.
The install and configure node types have been unified. The list of
node types can be found in the file:
/opt/lcg/yaim/scripts/node-info.def
To extract the list do:
grep FUNC /opt/lcg/yaim/scripts/node-info.def | sed "s/_FUNC.*//g"
Other:
======
- The SFTs have been packaged and are now available on the UI and WN.
- gsi version of ssh is included on the UIs.
YAIM site-info.def changes
==========================
DPMUSER_PWD
password of the database user for DPM.
DPMMGR
username of the database user for DPM. Can be chosen by the site
FTS_SERVER_URL
Set by those involved in the Service Challenge 3.
LEAVE THIS EMPTY IF YOU ARE NOT IN SC3!!!!
LFC_DB_PASSWORD
password of the database user for LFC.
VOBOX_HOST, host for the new VO box.
VOBOX_PORT, port used by the VO box.
GSSKLOG
set to yes if your site provides an X509 to KERBEROS Authentication
Server
(AFS used kerberos)
GSSKLOG_SERVER
Name of the local Authentication Server
LFC_TYPE
local or central
TORQUE_SERVER
host running the torque server.
APEL_DB_PASSWORD
password of the database user for apel.
SITE_LOC
the location of the site, (Nearest town/city).
SITE_LAT
latitude of the site.
SITE_LONG
longitude of the site.
SITE_WEB
web url for the site.
SITE_TIER
the site tier, (Tier0, Tier1 or Tier2)
SITE_SUPPORT_SITE
the site name of your Tier1 site. Tier1 sites should put CERN-CIC.
EDG_WL_SCRATCH
scratch directory for jobs.
VO_<VONAME>_VOMS_SERVERS
VOMS server for the VO.
VO_<VONAME>_VOMS_POOL_PATH
path on VOMS server for pool accounts.
VO_<VONAME>_VOMS_EXTRA_MAPS
for accounts other than sgm or pool.
Notice also that the following variables are not needed any more:
DPMCONFIG
DPMDB_PWD
DPMLOGS
DPM_POOLS
CA_WGET
SITE_VERSION
YAIM bugs that have been fixed:
-------------------------------
#6167 USERS_CONF required to exist
#6522 Password of DB with accounting information set to fixed string
#7612 Duplicated mis-used /etc/init.d RGMA service files
#7800 Have unique identifier for node/service types
#7852 hardcoded value for APEL database password
#8039 CE hostname is made PBS server name
#8286 Please install the LFC-client RPM as well on the LFC server
when
installed via YAIM
#8294 GIP/BDII do not publish correctly Close SE bindings with 4 SEs
#8492 lcg-expiregridmapdir expire period not long enough for Resource
Brokers
#8570 No way to set EDG_WL_SCRATCH via yaim site-def file.
#8576 LCMAPS gridmapfile and groupmapfile should be managed by YAIM
#8581 Default YAIM configuration for a BDII doesn't include
config_fmon_client
#8584 Add http_proxy env vars to lcgenv.sh
#8586 Request for Feature: lcg-bdii processes monitored on the CE
#8590 LFC server installation incomplete + /opt/lcg/bin not in $PATH
#8614 The lcg-expiregridmapdir.pl cron job should have argument '-v'
not '-v 1'
#8617 Request for Feature: new processes to monitor on a CE-node
#8629 yaim DCache and multiple pool nodes
#8688 Yaim General apt-get issue with source duplications
#8705 config_mkgridmap : permissions on /etc/grid-security/gridmapdir
not set
#8727 JAVA_OPTS="-Xmx???M -server" required on MON
#8733 config_bdii: incorrect 'chkconfig' usage causes lcg-bdii not to
restart at boot
#8770 MemSize for JAVA_OPTS in config_rgma_server must be an integer
#8900 sshd gets killed during configuration phase
#8931 gatekeeper does not have logrotate entry
#8961 YAIM UI config should not report as a problem that there is no
host cert found
#8962 YAIM UI config: spurious errors about qstat
#8973 config_gip requires CE_CLOSE_SE1_ACCESS_POINT even if
CE_CLOSE_SE does not contain "SE1"
#9141 Default Machine for Grid Ice Server
#9196 yaim and VOMS based VOs
#9226 config_rgma_server (JAVA_OPTS)
#9229 gridftp log rotate does not have to restart gridftp
#9250 gLite directory must be created for config_fts_client (tar WN
only?)
#9285 lfcmgr:x:33009:33009:LFC user:/home/lfcmgr:/bin/bash
#9286 apt repository entry for CASTOR-client rpm seems corrupt
#9292 config_mkgridmap makes bad VOMS gridmapfile and groupmapfile
#9311 RB-rpm is missing rgma gin rpm
#9414 LFC : please add the possibility to run the DLI
#9416 Tags for WNs shared areas to be published in the IS
#9546 Add secure rgma
#9547 edg-rgma clean-up
#9666 Yaim is not configuring /opt/glite/etc/rgma/rgma.conf on the
rgma server
#9717 change order of connectors in server.xml file
|
