This user manual provides information on running and using the UNICORE gateway. Please note also the following places for getting more information:
UNICORE Website: https://www.unicore.eu
Support list: unicore-support@lists.sf.net
Developer’s list: unicore-devel@lists.sf.net
1. Introduction
The Gateway is the entry point into a UNICORE site, routing HTTPS traffic to servers like UNICORE/X. It is installed in front of any networking firewall. It (optionally) authenticates incoming messages and forwards them to their intended destination. The Gateway receives the reply and sends it back to the client. In this way, only a single open port in a site’s firewall has to be configured.
Note
|
LIMITATIONS
This forwarding process only works for “most” HTTP requests, and is not a complete HTTP reverse proxy implementation. For example, it is not possible to run a full, complex web application like the UNICORE portal "behind" the gateway. Check the respective components’s manual whether it can be run behind the Gateway. |
In effect, traffic to a virtual URL, e.g. https://mygateway:8088/Alpha is forwarded to the real URL, e.g. https://host1:7777.
The mappings of virtual URL to real URL for the available sites are listed in a configuration file connections.properties. Additionally, the gateway supports dynamic registration of sites.
The second functionality of the gateway is (optional) authentication of incoming requests. Connections to the gateway are made using SSL, so the gateway can be configured to check whether the caller presents a certificate issued by a trusted authority. Information about the client is forwarded to services behind the gateway in UNICORE proprietary format (as a SOAP or HTTP header).
The Gateway will forward the IP address of the client to the back-end server.
Last not least, the Gateway can be configured as a HTTP load balancer.
Note
|
IMPORTANT NOTE ON PATHS
The UNICORE Gateway is distributed either as a platform independent and portable bundle (as a part of the UNICORE core server package) or as an installable, platform dependent package format such as RPM. Depending on the installation method, the paths to various Gateway files are different. If installing using a distribution-specific package the following paths are used: CONF=/etc/unicore/gateway BIN=/usr/sbin LOG=/var/log/unicore/gateway If installing using the portable bundle all Gateway files are installed under a single directory. Path prefixes then are as follows, where INST is a directory where the Gateway was installed: CONF=INST/conf BIN=INST/bin LOG=INST/log The above variables (CONF, BIN and LOG) are used throughout the rest of this manual. |
2. Installation
The UNICORE Gateway is distributed in the following formats:
-
As a part of platform independent installation bundle called UNICORE core server bundle. The UNICORE core server bundle is provided in two forms: one with a graphical installer and one with a command line installer.
-
As a binary, platform-specific package available currently for RedHat (Centos) and Debian platforms. Those packages are not tested on all possible platforms, but should work without any problems with other versions of similar distributions, e.g. SL6, Centos, or Fedora.
2.1. Installation from the core server bundle
Download the core server bundle from the UNICORE project website.
If you use the graphical installer, follow the on-screen instructions and do not forget to enable the Gateway checkbox when prompted.
If you use the console installer, please review the README file available after extracting the bundle. You don’t have to change any defaults as the Gateway is installed by default.
2.2. Installation from a Linux package (rpm or deb)
Use your distribution’s package manager to install.
3. Upgrading
The general update procedure is presented below, with possible variations:
-
Stop the old Gateway.
-
Update the server package. This step mostly applies for RPM/DEB managed installations. For Quickstart installation it is enough to replace the *.jar files with the new ones.
-
Start the newly installed Gateway.
-
Verify log file and fix any problems reported.
4. Configuration
The Gateway is configured using a set of configuration files, which reside in the CONF subdirectory.
4.1. Java and environment settings: startup.properties
This file contains settings related to the Java VM, such as the Java command to use, memory settings, library paths etc.
4.2. Configuring sites: connections.properties
This is a simple list connecting the names of sites and their physical addresses. An example is:
DEMO-SITE = https://localhost:7777 REGISTRY = https://localhost:7778
If this file is modified, the Gateway will re-read it at runtime, so there is no need to restart the Gateway in order to add or remove sites.
Optionally administrator can enable a possibility for dynamic site registration at runtime, see [dyn-reg] for details. Then this file should contain only the static entries (or none if all sites register dynamically).
Further options for back-end sites configuration are presented in [loadbalance].
4.3. Main server settings: gateway.properties
Use the gateway.hostname property to configure the network interface and port the Gateway will listen on. You can also select between https and http protocol, though in almost all cases https will be used.
Example:
gateway.hostname = https://192.168.100.123:8080
Note
|
If you set the host to 0.0.0.0, the Gateway will listen on all network interfaces of the host machine, else it will listen only on the specified one. |
If the scheme of the hostname URL is set to https, the Gateway uses the configuration data from security.properties to configure the HTTPS settings.
4.4. Certificate-less end-user access
With UNICORE 7, it is possible that end-users do not have client certificates. To enable them to connect, the Gateway needs to accept TLS connections without a client certificate. To configure this, set the following in gateway.properties
gateway.httpServer.requireClientAuthn=false
4.4.1. Scalability settings
To fine-tune the operational parameters of the embedded Jetty server, you can set advanced HTTP server parameters. See [ref-jetty] for details. Among others you can use the non-blocking IO connector offered by Jetty, which will scale up to higher numbers of concurrent connections than the default connector.
The Gateway acts as a https client for the VSites behind it. The number of concurrent calls is limited, and controlled by two parameters:
# maximum total number of concurrent calls to Vsites gateway.client.maxTotal=100 # total number of concurrent calls per site gateway.client.maxPerService=20
You can also control the limit on the maximum SOAP header size which is allowed by the Gateway. Typically you don’t have to touch this parameter. However if your clients do produce very big SOAP headers and the Gateway blocks them, you can increase the limit. Note that such a giant SOAP header usually means that the client is not behaving in a sane way, e.g. is trying to perform a DoS attack.
# maximum size of an accepted SOAP header, in bytes gateway.soapMaxHeader=102400
Note that Gateway may consume this amount of memory (plus some extra amount for other data) for each opened connection. Therefore, this value multiplied by the number of maximum allowed connections, should be significantly lower, then the total memory available for the Gateway.
4.4.2. Dynamic registration of Vsites
Dynamic registration is controlled by three properties in CONF/gateway.properties file:
gateway.registration.enable=true
If set to true, the Gateway will accept dynamic registrations which are made by sending a HTTP POST request to the URL /VSITE_REGISTRATION_REQUEST
Filters can be set to forbid access of certain hosts, or to require certain strings in the Vsite addresses. For example:
gateway.registration.deny=foo.org example.org
will deny registration if the remote hostname contains foo.org or example.org. Conversely,
gateway.registration.allow=mydomain.org
will only accept registrations if the remote address contains mydomain.org. These two (deny and allow) can be combined.
4.4.3. Web interface ("monkey page")
For testing and simple monitoring purposes, the Gateway displays a website showing detailed site information (the details view can be disabled). Once the Gateway is running, open up a browser and navigate to https://<gateway_host>:8080 (or whichever URL the gateway is running on). If the Gateway is configured to do SSL authentication, you will need to import a suitable client certificate into your web browser.
A HTML form for testing the dynamic registration is available as well, by clicking the link in the footer of the main Gateway page.
To disable the Vsite details page, set
gateway.disableWebpage=true
4.4.4. Main options reference
Property name | Type | Default value / mandatory | Description |
---|---|---|---|
gateway.hostname |
string |
mandatory to be set |
external gateway bind address |
gateway.registration.allow |
string |
- |
Space separated list of allowed hosts for dynamic registration. |
gateway.registration.deny |
string |
- |
Space separated list of denied hosts for dynamic registration. |
gateway.registration.enable |
[true, false] |
false |
Whether dynamic registration of sites is enabled. |
--- Passing Consignor info --- |
|||
gateway.consignorTokenTimeTolerance |
integer >= 0 |
30 |
The validity time of the authenticated client information passed to backend sites will start that many seconds before the real authentication. It is used to mask time synchronization problems between machines. |
gateway.consignorTokenValidity |
integer >= 1 |
60 |
What is the validity time of the authenticated client information passed to backend sites. Increase it if there machines clocks are not synhronized. |
gateway.signConsignorToken |
[true, false] |
false |
Controls whether information about the authenticated client (the consignor) passed to backend sites should be signed, or not. Signing is slower, but is required when sites may be reached directly, bypassing the Gateway. |
--- Gateway → Site client --- |
|||
gateway.client.chunked |
[true, false] |
true |
Controls whether chunked passing of HTTP requests to backend sites is supported. |
gateway.client.connectionTimeout |
integer number |
30000 |
Connection timeout, used when connecting to backend sites. |
gateway.client.expectContinue |
[true, false] |
true |
Controls whether the HTTP expec-continue mechanism is enaled on connections to backend sites. |
gateway.client.gzip |
[true, false] |
true |
Controls whether support for compression is announced to backend sites. |
gateway.client.keepAlive |
[true, false] |
true |
Whether to keep alive the connections to backend sites. |
gateway.client.maxPerService |
integer number |
20 |
Maximum allowed number of connections per backend site. |
gateway.client.maxTotal |
integer number |
100 |
Maximum total number of connections to backend sites allowed. |
gateway.client.socketTimeout |
integer number |
30000 |
Connection timeout, used when connecting to backend sites. |
--- Advanced --- |
|||
gateway.disableWebpage |
[true, false] |
false |
Whether the (so called monkey) status web page should be disabled. |
gateway.externalHostname |
string |
not set |
External address of the gateway, when it is accessible through a frontend server as Apache HTTP. |
gateway.soapMaxHeader |
integer [1024 — 1024000000] |
102400 |
Size in bytes of the accepted SOAP header. In the most cases you don’t need to change it. |
Property name | Type | Default value / mandatory | Description |
---|---|---|---|
gateway.httpServer.CORS_allowedHeaders |
string |
* |
CORS: comma separated list of allowed HTTP headers (default: any) |
gateway.httpServer.CORS_allowedMethods |
string |
GET,PUT,POST,DELETE,HEAD |
CORS: comma separated list of allowed HTTP verbs. |
gateway.httpServer.CORS_allowedOrigins |
string |
* |
CORS: allowed script origins. |
gateway.httpServer.CORS_chainPreflight |
[true, false] |
false |
CORS: whether preflight OPTION requests are chained (passed on) to the resource or handled via the CORS filter. |
gateway.httpServer.CORS_exposedHeaders |
string |
Location,Content-Type |
CORS: comma separated list of HTTP headers that are allowed to be exposed to the client. |
gateway.httpServer.disabledCipherSuites |
string |
empty string |
Space separated list of SSL cipher suites to be disabled. Names of the ciphers must adhere to the standard Java cipher names, available here: http://docs.oracle.com/javase/8/docs/technotes/guides/security/SunProviders.html#SupportedCipherSuites |
gateway.httpServer.enableCORS |
[true, false] |
false |
Control whether Cross-Origin Resource Sharing is enabled. Enable to allow e.g. accesing REST services from client-side JavaScript. |
gateway.httpServer.enableHsts |
[true, false] |
false |
Control whether HTTP strict transport security is enabled. It is a good and strongly suggested security mechanism for all production sites. At the same time it can not be used with self-signed or not issued by a generally trusted CA server certificates, as with HSTS a user can’t opt in to enter such site. |
gateway.httpServer.fastRandom |
[true, false] |
false |
Use insecure, but fast pseudo random generator to generate session ids instead of secure generator for SSL sockets. |
gateway.httpServer.gzip.enable |
[true, false] |
false |
Controls whether to enable compression of HTTP responses. |
gateway.httpServer.gzip.minGzipSize |
integer number |
100000 |
Specifies the minimal size of message that should be compressed. |
gateway.httpServer.highLoadConnections |
integer number |
200 |
If the number of connections exceeds this amount, then the connector is put into a special low on resources state. Existing connections will be closed faster. Note that the server will also go to the low on resources mode if there are no available threads in the pool. You can set this to 0 to disable the connections limit (and have only thread pool size governed limit). If set to a negative number then the low on resources mode won’t be used at all. |
gateway.httpServer.lowResourceMaxIdleTime |
integer >= 1 |
100 |
In low resource conditions, time (in ms.) before an idle connection will time out. |
gateway.httpServer.maxIdleTime |
integer >= 1 |
200000 |
Time (in ms.) before an idle connection will time out. It should be large enough not to expire connections with slow clients, values below 30s are getting quite risky. |
gateway.httpServer.maxThreads |
integer number |
255 |
Maximum number of threads to have in the thread pool for processing HTTP connections. Note that this number will be increased with few additional threads to handle connectors. |
gateway.httpServer.minThreads |
integer >= 1 |
1 |
Minimum number of threads to have in the thread pool for processing HTTP connections. Note that this number will be increased with few additional threads to handle connectors. |
gateway.httpServer.requireClientAuthn |
[true, false] |
true |
Controls whether the SSL socket requires client-side authentication. |
gateway.httpServer.soLingerTime |
integer number |
-1 |
Socket linger time. |
gateway.httpServer.useNIO |
[true, false] |
true |
DEPRECATED, no effect |
gateway.httpServer.wantClientAuthn |
[true, false] |
true |
Controls whether the SSL socket accepts (but does not require) client-side authentication. |
gateway.httpServer.xFrameAllowed |
string |
http://localhost |
URI origin that is allowed to embed web interface inside a (i)frame. Meaningful only if the xFrameOptions is set to allowFrom. The value should be in the form: http[s]://host[:port] |
gateway.httpServer.xFrameOptions |
[deny, sameOrigin, allowFrom, allow] |
deny |
Defines whether a clickjacking prevention should be turned on, by insertionof the X-Frame-Options HTTP header. The allow value disables the feature. See the RFC 7034 for details. Note that for the allowFrom you should define also the xFrameAllowed option and it is not fully supported by all the browsers. |
4.5. security.properties
In the security.properties file, the trust store and Gateway credential is configured.
Property name | Type | Default value / mandatory | Description |
---|---|---|---|
gateway.credential.path |
filesystem path |
mandatory to be set |
Credential location. In case of jks, pkcs12 and pem store it is the only location required. In case when credential is provided in two files, it is the certificate file path. |
gateway.credential.format |
[jks, pkcs12, der, pem] |
- |
Format of the credential. It is guessed when not given. Note that pem might be either a PEM keystore with certificates and keys (in PEM format) or a pair of PEM files (one with certificate and second with private key). |
gateway.credential.password |
string |
- |
Password required to load the credential. |
gateway.credential.keyPath |
string |
- |
Location of the private key if stored separately from the main credential (applicable for pem and der types only), |
gateway.credential.keyPassword |
string |
- |
Private key password, which might be needed only for jks or pkcs12, if key is encrypted with different password then the main credential password. |
gateway.credential.keyAlias |
string |
- |
Keystore alias of the key entry to be used. Can be ignored if the keystore contains only one key entry. Only applicable for jks and pkcs12. |
Property name | Type | Default value / mandatory | Description |
---|---|---|---|
gateway.truststore.allowProxy |
[ALLOW, DENY] |
ALLOW |
Controls whether proxy certificates are supported. |
gateway.truststore.type |
[keystore, openssl, directory] |
mandatory to be set |
The truststore type. |
gateway.truststore.updateInterval |
integer number |
600 |
How often the truststore should be reloaded, in seconds. Set to negative value to disable refreshing at runtime. (runtime updateable) |
--- Directory type settings --- |
|||
gateway.truststore.directoryConnectionTimeout |
integer number |
15 |
Connection timeout for fetching the remote CA certificates in seconds. |
gateway.truststore.directoryDiskCachePath |
filesystem path |
- |
Directory where CA certificates should be cached, after downloading them from a remote source. Can be left undefined if no disk cache should be used. Note that directory should be secured, i.e. normal users should not be allowed to write to it. |
gateway.truststore.directoryEncoding |
[PEM, DER] |
PEM |
For directory truststore controls whether certificates are encoded in PEM or DER. Note that the PEM file can contain arbitrary number of concatenated, PEM-encoded certificates. |
gateway.truststore.directoryLocations.* |
list of properties with a common prefix |
- |
List of CA certificates locations. Can contain URLs, local files and wildcard expressions. (runtime updateable) |
--- Keystore type settings --- |
|||
gateway.truststore.keystoreFormat |
string |
- |
The keystore type (jks, pkcs12) in case of truststore of keystore type. |
gateway.truststore.keystorePassword |
string |
- |
The password of the keystore type truststore. |
gateway.truststore.keystorePath |
string |
- |
The keystore path in case of truststore of keystore type. |
--- Openssl type settings --- |
|||
gateway.truststore.opensslNewStoreFormat |
[true, false] |
false |
In case of openssl truststore, specifies whether the trust store is in openssl 1.0.0+ format (true) or older openssl 0.x format (false) |
gateway.truststore.opensslNsMode |
[GLOBUS_EUGRIDPMA, EUGRIDPMA_GLOBUS, GLOBUS, EUGRIDPMA, GLOBUS_EUGRIDPMA_REQUIRE, EUGRIDPMA_GLOBUS_REQUIRE, GLOBUS_REQUIRE, EUGRIDPMA_REQUIRE, EUGRIDPMA_AND_GLOBUS, EUGRIDPMA_AND_GLOBUS_REQUIRE, IGNORE] |
EUGRIDPMA_GLOBUS |
In case of openssl truststore, controls which (and in which order) namespace checking rules should be applied. The REQUIRE settings will cause that all configured namespace definitions files must be present for each trusted CA certificate (otherwise checking will fail). The AND settings will cause to check both existing namespace files. Otherwise the first found is checked (in the order defined by the property). |
gateway.truststore.opensslPath |
filesystem path |
/etc/grid-security/certificates |
Directory to be used for opeenssl truststore. |
--- Revocation settings --- |
|||
gateway.truststore.crlConnectionTimeout |
integer number |
15 |
Connection timeout for fetching the remote CRLs in seconds (not used for Openssl truststores). |
gateway.truststore.crlDiskCachePath |
filesystem path |
- |
Directory where CRLs should be cached, after downloading them from remote source. Can be left undefined if no disk cache should be used. Note that directory should be secured, i.e. normal users should not be allowed to write to it. Not used for Openssl truststores. |
gateway.truststore.crlLocations.* |
list of properties with a common prefix |
- |
List of CRLs locations. Can contain URLs, local files and wildcard expressions. Not used for Openssl truststores. (runtime updateable) |
gateway.truststore.crlMode |
[REQUIRE, IF_VALID, IGNORE] |
IF_VALID |
General CRL handling mode. The IF_VALID setting turns on CRL checking only in case the CRL is present. |
gateway.truststore.crlUpdateInterval |
integer number |
600 |
How often CRLs should be updated, in seconds. Set to negative value to disable refreshing at runtime. (runtime updateable) |
gateway.truststore.ocspCacheTtl |
integer number |
3600 |
For how long the OCSP responses should be locally cached in seconds (this is a maximum value, responses won’t be cached after expiration) |
gateway.truststore.ocspDiskCache |
filesystem path |
- |
If this property is defined then OCSP responses will be cached on disk in the defined folder. |
gateway.truststore.ocspLocalResponders.<NUMBER> |
list of properties with a common prefix |
- |
Optional list of local OCSP responders |
gateway.truststore.ocspMode |
[REQUIRE, IF_AVAILABLE, IGNORE] |
IF_AVAILABLE |
General OCSP ckecking mode. REQUIRE should not be used unless it is guaranteed that for all certificates an OCSP responder is defined. |
gateway.truststore.ocspTimeout |
integer number |
10000 |
Timeout for OCSP connections in miliseconds. |
gateway.truststore.revocationOrder |
[CRL_OCSP, OCSP_CRL] |
OCSP_CRL |
Controls overal revocation sources order |
gateway.truststore.revocationUseAll |
[true, false] |
false |
Controls whether all defined revocation sources should be always checked, even if the first one already confirmed that a checked certificate is not revoked. |
4.5.1. Proxy certificate support
The UNICORE Gateway optionally accepts proxy certificates as used by other Grid middleware systems. In general, we think proxies are a bad idea, but for interoperability purposes, proxies support can be enabled. If enabled, the clients using proxies are authenticated as the initial issuer of the presented proxy certificates chain. See the above reference properties table for the actual setting.
4.6. Logging
UNICORE uses the Log4j logging framework. It is configured using a config file. By default, this file is found in components configuration directory and is named logging.properties. The config file is specified with a Java property log4j.configuration (which is set in startup script).
Several libraries used by UNICORE also use the Java utils logging facility (the output is two-lines per log entry). For convenience its configuration is also controlled in the same logging.properties file and is directed to the same destination as the main Log4j output.
Note
|
You can change the logging configuration at runtime by editing the logging.properties file. The new configuration will take effect a few seconds after the file has been modified. |
By default, log files are written to the the LOGS directory.
The following example config file configures logging so that log files are rotated daily.
# Set root logger level to INFO and its only appender to A1. log4j.rootLogger=INFO, A1 # A1 is set to be a rolling file appender with default params log4j.appender.A1=org.apache.log4j.DailyRollingFileAppender log4j.appender.A1.File=logs/uas.log #configure daily rollover: once per day the uas.log will be copied #to a file named e.g. uas.log.2008-12-24 log4j.appender.A1.DatePattern='.'yyyy-MM-dd # A1 uses the PatternLayout log4j.appender.A1.layout=org.apache.log4j.PatternLayout log4j.appender.A1.layout.ConversionPattern=%d [%t] %-5p %c{1} %x - %m%n
Note
|
In Log4j, the log rotation frequency is controlled by the DatePattern. Check http://logging.apache.org/log4j/1.2/apidocs/org/apache/log4j/DailyRollingFileAppender.html for the details. |
For more info on controlling the logging we refer to the log4j documentation:
Log4j supports a very wide range of logging options, such as date based or size based file rollover, logging different things to different files and much more. For full information on Log4j we refer to the publicly available documentation, for example the Log4j manual.
4.6.1. Logger categories, names and levels
Logger names are hierarchical. In UNICORE, prefixes are used (e.g. "unicore.security") to which the Java class name is appended. For example, the XUUDB connector in UNICORE/X logs to the "unicore.security.XUUDBAuthoriser" logger.
Therefore the logging output produced can be controlled in a fine-grained manner. Log levels in Log4j are (in increasing level of severity):
# TRACE on this level huge pieces of unprocessed information are dumped, # DEBUG on this level UNICORE logs (hopefully) admin-friendly, verbose information, useful for hunting problems, # INFO standard information, not much output, # WARN warnings are logged when something went wrong (so it should be investigated), but recovery was possible, # ERROR something went wrong and operation probably failed, # FATAL something went really wrong - this is used very rarely for critical situations like server failure.
For example, to debug a security problem in the UNICORE security layer, you can set:
log4j.logger.unicore.security=DEBUG
If you are just interested in details of credentials handling, but not everything related to security you can use the following:
log4j.logger.unicore.security=INFO log4j.logger.unicore.security.CredentialProperties=DEBUG
so the XUUDBAuthoriser will log on DEBUG level, while the other security components log on INFO level.
Note
|
(so the full category is printed) and turn on the general DEBUG logging for a while (on unicore). Then interesting events can be seen and subsequently the logging configuration can be fine tuned to only show them. |
4.6.2. Logging
The most important, root log categories used by the Gateway’s logging are:
unicore.gateway |
General Gateway logging |
unicore.connections |
Log IPs of clients, and the DN after the SSL handshake |
unicore.httpserver |
HTTP processing, Jetty server |
unicore.security |
Certificate details and other security |
The Gateway uses the so called MDC (Mapped Diagnostic Context) to provide additional information on the client which is served. In the MDC the client’s IP address and client’s Distinguished Name is stored. You can control whether to attach MDC to each line by the %X{entry} log4j pattern entry. As the entry you can use: clientIP or clientName. For example:
log4j.appender.A1.layout.ConversionPattern=%d [%t] [%X{clientIP} %X{clientName}] %-5p %c{1} - %m%n
5. Using Apache httpd as a frontend
You may wish to use the Apache webserver (httpd) as a frontent for the Gateway (e.g. for security or fault-tolerance reasons).
Requirements
-
Apache httpd
-
mod_proxy for Apache httpd
External references
6. Using the Gateway for failover and/or loadbalancing of UNICORE sites
The Gateway can be used as a simple failover solution and/or loadbalancer to achieve high availability and/or higher scalability of UNICORE/X sites without additional tools.
A site definition (in CONF/connections.properties) can be extended, so that multiple physical servers are used for a single virtual site.
An example for such a so-called multi-site declaration in the connections.properties file looks as follows:
#declare a multisite with two physical servers MYSITE=multisite:vsites=https://localhost:7788 https://localhost:7789
This will tell the Gateway that the virtual site "MYSITE" is indeed a multi-site with the two given physical sites.
6.1. Configuration
Configuration options for the multi-site can be passed in two ways. On the one hand they can go into the connections.properties file, by putting them in the multi-site definition, separated by ";" characters:
#declare a multisite with parameters MYSITE=multisite:param1=value1;param2=value2;param3=value3;...
The following general parameters exist
vsites |
List of physical sites |
strategy |
Class name of the site selection strategy to use (see below) |
config |
Name of a file containing additional parameters |
Using the "config" option, all the parameters can be placed in a separate file for enhanced readability. For example you could define in connections.properties:
#declare a multisite with parameters read from a separate file MYSITE=multisite:config=conf/mysite-cluster.properties
and give the details in the file "conf/mysite-cluster.properties":
#example multisite configuration vsites=https://localhost:7788 https://localhost:7789 #check site health at most every 5 seconds strategy.healthcheck.interval=5000
6.2. Available Strategies
A selection strategy is used to decide where a client request will be routed. By default, the strategy is "Primary with fallback", i.e. the request will go to the first site if it is available, otherwise it will go to the second site.
Primary with fallback
This strategy is suitable for a high-availability scenario, where a secondary site takes over the work in case the primary one goes down for maintenance or due to a problem. This is the default strategy, so nothing needs to be configured to enable it. If you want to explicitely enable it anyway, set
strategy=primaryWithFallback
The strategy will select from the first two defined physical sites. The first, primary one will be used if it is available, else the second one. Health check is done on each request, but not more frequently as specified by the "strategy.healthcheck.interval" parameter. By default, this parameter is set to 5000 milliseconds.
Changes to the site health will be logged at "INFO" level, so you can see when the sites go up or down.
Round robin
This strategy is suitable for a load-balancing scenario, where a random site will be chosen from the available ones. To enable it, set
strategy=roundRobin
Changes to the site health will be logged at "INFO" level, so you can see when the sites go up or down.
It is very important to be aware that this strategy requires that all backend sites used in the pool, share a common persistence. It is because Gateway does not track clients, so particular client requests may land at different sites. This is typically solved by using a non-default, shared database for sites, such as MySQL.
Note
|
Currently loadbalancing of target sites is an experimental feature and is not yet fully functional. It will be improved in future UNICORE versions. |
Custom strategy
You can implement and use your own failover strategy, in this case, use the name of the Java class as strategy name:
strategy=your_class_name
7. Gateway failover and migration
The [loadbalance] covered usage of the Gateway to provide failover of backend services. However it may be needed to guarantee high-availabilty for the Gateway itself or to move it to other machine in case of the original one’s failure.
7.1. Gateway’s migration
The Gateway does not store any state information, therefore its migration is easy. It is enough to install the Gateway at the target machine (or even to simply copy it in the case of installation from the core server bundle) and to make sure that the original Gateway’s configuration is preserved.
If the new machine uses a different address, it needs to be reflected in the server’s configuration file (the listen address). Also, the configuration of sites behind the Gateway must be updated accordingly.
7.2. Failover and loadbalancing of the Gateway
Gateway itself doesn’t provide any features related to its own redundancy. However as it is stateless, the standard redundancy solutions can be used.
The simpliest solution is to use Round Robin DNS, where DNS server routes the Gateway’s DNS address to a pool of real IP addresses. While easy to set up this solution has a significant drawback: DNS server doesn’t care about machines being down.
The much better choice is to use the Linux-HA software suite, often known under the name of its principal component, the heartbeat. For details see http://www.linux-ha.org
Additionally a more advanced HTTP-aware software can be used, such as HA-Proxy (http://haproxy.1wt.eu). Currently Gateway and UNICORE don’t maintain HTTP sessions so usage of the HTTP-aware load-balancer is not strictly required, but such solutions generally provide more general purpose features.
8. Building the Gateway from source
To checkout the latest version of the Gateway source code, do
svn co http://svn.code.sf.net/p/unicore/svn/gateway/trunk gateway
You will need to install Maven from http://maven.apache.org Compile using
mvn install
which compiles the code and runs the tests.
The file "README-building.txt" contains instructions for building distributable packages.