Overview
This document is made up of four sections, describing installation , post-installation
work , performance and the admin interface
Installation
Prerequisites
- The icat distribution: icat.server-4.7.0-distro.zip
- Java 8
- A suitable deployed container. The installation scripts
support Glassfish and to a lesser extent WildFly. Testing has been
carried out with Glassfish 4.0. Glassfish
installation instructions are available.
- A database as described in Database
installation instructions installed on the server
- Deployed ICAT authenticators.
- Python (version 2.4 to 2.7) and the python-suds package
installed on the server.
- MySQL-python must be installed if you have an MySQL 4.2.x
schema you need to upgrade.
- cx_Oracle must be installed (in addition to an Oracle client)
if you have an Oracle 4.2.x schema you need to upgrade.
- icat-setup
version 1.1.x or higher must be installed if you have a 4.2.x schema
you need to upgrade and you wish to preserve existing rules.
Installing a group of ICATs
If your facility depends upon a single ICAT instance then
ingestion of data can be held up by a user making an expensive query.
To avoid this it is suggested that you install multiple servers
each running a Glassfish with an ICAT but all sharing one database.
However there are a number of opportunities to get things wrong in the
setup so it is recommended that you install the central machine first
and make sure that it works before adding in the satellites. Ingestion
can be directed to one node and the other nodes can be load balanced
for user access by, for example, an Apache web server. In this
documentation one machine is referred to as the central one and the
others are referred to as satellites.
All machines must use the same database. If the central machine
is running the database then ports must be opened to the satellite
machines. By default this is 3306 for MySQL and 1521 for Oracle. Ports
must also be opened on the central machine for JMS and IIOP which are
by default 7676 and 3700 and respectively. In addition it seems that
the ORB (IIOP) makes unusual use of the ephemeral ports which require
that they are also open on the central machine. The set of ephemeral
ports is large - and for unix may be found at
/proc/sys/net/ipv4/ip_local_port_range. Security is not enabled for
either JMS or IIOP communication so the firewalls should normally be
configured to only allow traffic from the set of satellite machines.
Authentication can either be carried out on the central machine
- which has the advantage that you only have authenticator logs
building up on that machine or it can be handled by the satellites so
distributing the load better. Even if you choose to use central
authentication the authenticators must, unfortunately still be
installed on the satellites. Glassfish seems to get confused if the
JNDI resources are not recognised locally on the satellites. To enable
central authentication you need a line in the icat.properties file on
each satellite machine for each authenticator of the form:
authn.XXX.hostPort = <central machine>:3700 where XXX is one of
the values listed in the authn.list property.
Lucene calls must also be directed to the central machine with a
line in each satellite's icat.properties file of the form:
lucene.hostPort = <central machine>:3700. This is the only
lucene property that should be set on the satellites. It is essential
that this property is set otherwise each node will only see changes
made via that machine.
JMS messages must all be sent to the central machine. This is
important otherwise cache refresh messages may be missed. All machines
may produce these messages and all machines are listening for them. To
configure this an entry is added to the icat-setup.properties file
(not the icat.properties file) of the satellite machines of the form
jmsHostPort = <central machine>:7676
You could then set up an Apache front end to do load balancing. This
will probably just connect to the satellites leaving the central
machine to handle ingestion of data. See Apache
front end for one way of doing this.
Schema upgrade
Any existing lucene database should be removed. See the
icat.properties file for the value of lucene.directory and ensure that
the directory specified there is empty.
The database schema must be upgraded in steps depending upon how
old your icat installation is
Upgrade 4.2.5 schema to 4.3.x
This is for upgrading a 4.2.5 schema to 4.3.2. If you have
already upgraded to 4.3.x skip this step. Do not attempt to use this
procedure on a 4.3.x schema!
- Back up the database in case it should get into a state from
which recovery is impractical.
- Run the get_rules program to save the rules in the format
accepted by icat-setup. This must be run as somebody who has read
access to the rules. For example: ./get_rules.py
https://example.com:8181 db username root password password The
program should report how many rules it has saved and where.
- Ensure that nobody tries to use ICAT while it is being
upgraded - the simplest approach is to undeploy the old one which can
be done from the command line or by using a web browser and
connecting on the admin port (typically 4848) and undeploying from
there.
- For MySQL edit username, password, schema and dbhost at the
top of the file ./upgrade_mysql_4_2_5.py and run it or or
for Oracle, edit username , password and db at the top of the file ./upgrade_oracle_4_2_5.py
and run that. Note that the procedure has been tested on ICAT 4.2.5
but should work for earlier 4.2 versions. The script will first check
that everything should go work. If it reports problems fix them and
try again. Once it gets past the checking stage it starts the
conversion which can take a long time (many hours for a production
system). At the end you should have a 4.3.2 database. Any indices
which had been created manually will have been removed.
- Install the new icat
- Restore the rules using the icat-setup tool. For
example:
icat-setup -f rules.authz https://example.com:8181 db
username root password secret
This assumes that you are in the directory where you ran
get_rules.py which will have created a file rules.authz. The
credentials (keyword value pairs following the authenticator
mnemonic) should be those of one of the users specified in the
rootUserNames of the icat.properties file.
Please check the rules.authz first as it will not work if it
references entities that no longer exist. For example "Group" must
be replaced by "Grouping" and InputDatafile is no longer part of
ICAT. Also because some problems have been found with conditions
containing dots (such as "InvestigationUser [user.name='fred']") in
rules these must now be re-expressed without dots.
Upgrade 4.3.x schema to 4.4.0
You may increase the size of the "what" column of the Rule table to
1024 to match the size the column has on a brand new installation and
you must modify the INVESTIGATIONUSER table as role values
may no longer be null. Choose a name to use for the default role - for
example 'member' and for MySQL:
UPDATE INVESTIGATIONUSER SET ROLE = 'member' WHERE ROLE IS NULL;
ALTER TABLE INVESTIGATIONUSER MODIFY COLUMN ROLE varchar(255) NOT
NULL;
ALTER TABLE INVESTIGATIONUSER DROP FOREIGN KEY
FK_INVESTIGATIONUSER_USER_ID;
ALTER TABLE INVESTIGATIONUSER
DROP INDEX UNQ_INVESTIGATIONUSER_0;
ALTER TABLE
INVESTIGATIONUSER ADD CONSTRAINT UNQ_INVESTIGATIONUSER_0 UNIQUE
(USER_ID, INVESTIGATION_ID, ROLE);
ALTER TABLE
INVESTIGATIONUSER ADD CONSTRAINT FK_INVESTIGATIONUSER_USER_ID FOREIGN
KEY (USER_ID) REFERENCES USER_ (ID)
or for Oracle:
UPDATE INVESTIGATIONUSER SET ROLE = 'member' WHERE ROLE IS NULL;
ALTER TABLE INVESTIGATIONUSER MODIFY (ROLE varchar2(255) NOT NULL);
ALTER TABLE DROP CONSTRAINT UNQ_INVESTIGATIONUSER_0;
ALTER
TABLE INVESTIGATIONUSER ADD CONSTRAINT UNQ_INVESTIGATIONUSER_0 UNIQUE
(USER_ID, INVESTIGATION_ID, ROLE);
Upgrade 4.4.0 schema to 4.5.0
The mechanism for assigning unique id values for each entity in ICAT
has been changed. Previously a sequence table (called SEQUENCE) was
used to hold the last value used. This has been changed to make use of
the native DBMS mechanism which is making the id columns
AUTO_INCREMENT for MySQL or using a Sequence (rather than a table
called SEQUENCE) in the case of Oracle. Do not omit this step
otherwise the id values for new rows will not be set correctly and you
will run into problems with duplicate values. For MySQL run
mysql -u icat -p icat < upgrade_mysql_4_4.sql
or for Oracle run
sqlplus icat @upgrade_oracle_4_4.sql
where in both cases it is assumed that the tables are owned by user
"icat". The MySQL script is simply a list of alter table statements
for each table in the 4.4.0 schema. The oracle script takes the last
sequence number from the SEQUENCE table and uses this to initialize a
sequence. Note that the increment for the sequence must be exactly 50
and the start value must be at least 51 more than the number in the
old SEQUENCE table.
Upgrade 4.5.x or 4.6.x schema to 4.7.0
The Rule and DataCollection tables have changed. First dump the
existing rules with:
./rules.py dump https://example.com:8181 db username root
password secret > rules.ie
i.e. run the rules.py script in the unpacked distribution directory
with the "dump" parameter, the url of your icat and then the
authenticator plugin and credentials to identify a user specified in
the rootUserNames list in icat.properties. This will redirect the dump
into a file called rules.ie. Then undeploy the existing icat
./setup uninstall -v
Then for MySQL run
mysql -u icat -p icat < upgrade_mysql_4_7.sql
or for Oracle run
sqlplus icat @upgrade_oracle_4_7.sql
where in both cases it is assumed that the tables are owned by user
"icat". The script will drop the Rule table and add a DOI column to
the DataCollection table. Next install the new ICAT (which will
recreate an empty Rule table) Finally run
./rules.py load https://example.com:8181 db username root
password secret < rules.ie
which will populate the Rule table. Only run this script once
or you will get duplicate entries in the table.
In addition the
Log
table is no longer used and may be dropped after you have extracted
any information from it that you need.
The icat-setup.properties file
- container
-
Values must be chosen from: TargetServer
Though only Glassfish is working properly at the moment.
- home
- is the top level of the container installation. For Glassfish
it must contain "glassfish/domains" and for JBoss (wildfly) it must
contain jboss-modules.jar.
- port
- is the administration port of the container which is
typically 4848 for Glassfish and 9990 for JBoss.
- secure
- must be set to true or false. If true then only https and not
http connections will be allowed.
- db.driver
- is the name of the jdbc driver which must match the jar file
installed in the container and matching your database.
- db.url
- url to connect to your database. For example:
jdbc:mysql://localhost:3306/icat
- db.username
- username to connect to your database.
- db.password
- password to connect to your database.
- db.target
-
This is optional and may be used to control the SQL generated by the
JPA. Values must be chosen from: TargetDatabase
- db.logging
-
This is optional and if set to one of the values in Eclipse
Link logging.level controls the logging of JPA generated SQL
statements.
- jmsHostPort
-
This is optional. It should only be specified on satellite machines
when you have a group of machines working together as described at Installing a group of ICATs . It takes the form
machine:port where the port will normally be 7676.
The icat.properties and icat.logback.xml files
If you wish to modify the provided logging levels then rename
icat.logback.xml.example to icat.logback.xml and update the
icat.properties file to reference it as explained below. The
icat.properties file may need other changes:
- lifetimeMinutes
- Defines the lifetime of an ICAT sessionid. You should avoid
making it have a long duration as this increases the risk if it is
intercepted, lost or stolen.
- rootUserNames
- Is a space separated list of user identifiers having full
access to all tables. The format of the user identifier is determined
by the chosen authentication plugin. The authn_db and authn_ldap
plugins may be configured to either return the simple user name or to
prepend it with a name identifying the mechanism. For example if
there is a an entry "root" in the database then if the authn_db
authenticator is configured without a mechanism then the user name to
consider will be just "root", however if it has been configured with
a mechanism of "db" then the string "db/root" must be specified.
- maxEntities
- Restrict total number of entities to return in a search or
get call. This should be set as small as possible to protect the
server from running out of memory. However if you set it too small it
may prevent users from doing reasonable things.
- maxIdsInQuery
- For handling INCLUDEs, ICAT may generate queries which are
not acceptable to the database system. To avoid this problem such
queries are broken down. This is the maximum size of each chunk which
must not exceed 1000 for Oracle.
- importCacheSize
- the size of a cache used during import to avoid an excessive
number of calls to the database. The cache is dropped after each call
to import to ensure that authorization rules are enforced. As the
cache is short-lived, modifications to ICAT are unlikely to result in
stale information being used from the cache.
- exportCacheSize
- the size of a cache used during export to avoid an excessive
number of calls to the database. The cache is dropped after each call
to export to ensure that authorization rules are enforced. As the
cache is short-lived, modifications to ICAT are unlikely to result in
stale information being used from the cache.
- authn.list
- is a space separated set of mnemonics for user to select the
plugin in the login call. This must not reference plugins which are
not installed as plugins are checked when ICAT performs its
initialisation; if plugins are missing ICAT will not start.
- authn.<mnemonic>.jndi
- is the jndi name to locate the plugin. When you installed the
plugin a message would have appeared in the server.log stating the
JNDI names. For example for authn_db you would expect to see
java:global/authn_db.ear-1.0.0/authn_db.ejb-1.0.0/DB_Authenticator.
There must be one such entry for each plugin.
- authn.<mnemonic>.friendly
- is optional. It gives a name that a tool might use to label
the plugin.
- authn.<mnemonic>.admin
- is optional. Set to true if you wish to indicate that this
authenticator should only be advertised to administration tools.
- authn.<mnemonic>.hostPort
-
is optional. It should only be specified on satellite machines when
you have a group of machines working together as described at Installing a group of ICATs and when you want to
perform authentication on the central machine. The value takes the
form machine:port where the port will normally be 3700.
- notification.list
- is optional. It is a space separated set of Entity names for
which you with to generate notifications. For each one there must be
another line saying under what conditions you wish to generate a
notification for the entity.
- notification.<entity name>
- a string of letters taken from the set "C" and "U" indicating
for which operations (create and update) you wish to be notified for
that kind of operation on the entity.
- log.list
- is optional. If present it specifies a set of call types to
log via JMS calls. The types are specified by a space separated list
of values taken from READ, WRITE, SESSION, INFO.
- logback.xml
- This is optional. If present it must specify the path to a
logback.xml file. The path may be absolute or relative to the config
directory.
- lucene.directory
- This is optional. It is the path to a directory (whose parent
must exist) in which to store the lucene index. If this is specified
then lucene.commitSeconds and lucene.commitCount must both be
specified. If it is omitted and lucene.hostPort is also omitted then
lucene indices will not be created and the searchText() call will
return nothing.
- lucene.commitSeconds
- the interval in seconds between committing lucene changes to
disk and updating the index. If you set it to 300 then searchText()
calls will see what was available at some time in the past (up to 5
minutes ago) and which is also currently present.
- lucene.commitCount
- the number of changes to accumulate before committing them to
disk. If the number is too high there can be memory problems.
Currently this is only used by the call to lucenePopulate.
- lucene.hostPort
-
This is optional and if set any other lucene settings will be
ignored. It should only be specified on satellite machines when you
have a group of machines working together as described at Installing a group of ICATs The value takes the
form machine:port where the port will normally be 3700.
Check that ICAT works
A small test program,
testicat,
will have been installed for you. This is a python script which
requires that the suds client is available. This connects as one of
the root users you defined as 'rootUserNames' in the icat.properties
file. Invoke the script specifying the url of the machine on which the
ICAT service is deployed (something like https://example.com:8181),
the mnemonic for the chosen authentication plugin followed by the
credentials for one of the root user names supported by that plugin.
These credentials should be passed in as pairs of parameters with key
followed by value. For example:
testicat https://example.com:8181 db username root
password secret
It should report:
Logged in as ... with 119.9... minutes to go
Login, search,
create, delete and logout operations were all successful.
This script can be run at any time as it is almost harmless - it
simply creates a "Group" with an unlikely name and removes it again.
In case of problems, first erase the directory /tmp/suds and try
the testicat again. If it still fails, look at the log files:
server.log and icat.log which can both be found in the logs directory
below your domain. Look also at the relevant authenticator log.
Post-installation work
Fresh Install
If this is a fresh install then you can use the import facility
to do the initial icat population or you could use the icat manager to
create rules, a Facility and other high level entities.
If you are using Oracle the type NUMBER(38, 19) will have been used
for all floating point numbers. This constrains the values that can be
stored - they may be truncated or rejected. To fix this please execute
the SQL statements in
fix_floats_oracle.sql
In all cases
Populate the lucene index by using the
icatadmin tool.
Performance
To improve performance:
- Consider creating the indices defined in indices.sql. Indices
can make a huge difference to the database performance but there is
also a small cost for each index.
- Make entities readable by anyone if they contain no sensitive
information. This is generally the case for those entities that
implement an many-to-many relationship. For example InvestigationUser
relates Investigation to User but has no attributes. By making it
world readable no access to Investigation or User is granted. An in
memory cache of world readable entities is maintained by ICAT.
- Add entries to PublicStep to allow the INCLUDE mechanism to
be less costly. PublicStep is explained in the ICAT Java Client User
Manual. Its contents are also held in an in-memory cache for
performance.
The icatadmin tool
Administration operations have been added to the ICAT API and are
accessible via the icatadmin tool which will have been installed by
the
setup.py
script. It should be invoked as:
icatadmin <url> <plugin>
<credentials>... -- <command> <args>...
to run a single command or
icatadmin <url> <plugin>
<credentials>...
to be prompted for a series of commands as shown below. In
either case if you specify '-' as the password you will be prompted
for it. Note that in the single command case the "--" marker is needed
to terminate the list of credentials. For example:
icatadmin https://example.com:8181 db username root
password secret -- properties
Only users mentioned in the rootUserNames of the icat.properties file
are authorized to use this command.
- populate [<entity name>]
- re-populates lucene for the specified entity name. This is
useful if the database has been modified directly rather than by
using the ICAT API. This call is asynchronous and simply places the
request in a set of entity types to be populated. When the request is
processed all lucene entries of the specified entity type are first
cleared then the corresponding icat entries are scanned to
re-populate lucene. To find what it is doing please use the
"populating" operation described below. It may also be run without an
entity name in which case it will process all entities. The new
lucene index will not be seen until it is completely rebuilt. While
the index is being rebuilt ICAT can be used as normal as any lucene
updates are stored to be applied later.
- populating
- returns a list of entity types to be processed for populating
lucene. Normally the first item returned will be being processed
currently.
- commit
- instructs lucene to update indices. Normally this is not
needed as it is will be done periodically according to the value of
lucene.commitSeconds
- clear
- stops any population and clears all the lucene indices.