URL Rewriting

(2Q19)


This article describes the URL rewriting mechanism in eXist-db.

Warning:

This article must be rewritten completely. It contains a lot of outdated information. Be careful!

Basics

URL rewriting is done by a standard Java servlet filter called XQueryURLRewrite. Its main job is to intercept incoming requests and forward them to the appropriate handlers, which are again standard servlets. In fact, there's nothing eXist-specific to the servlet filter, except that it uses XQuery scripts to configure the forwarding and URL rewriting. Like any other servlet filter, it is configured in etc/webapp/WEB-INF/web.xml.

A controller XQuery is executed once for every requests. It must return an XML fragment, which tells the servlet filter how to proceed with the request. The returned XML fragment can range from simply define forwarding up to describing complex pipelines involving multiple steps.

The main advantage of using XQuery for the controller is that we have the whole power of the language available. The controller can look at request parameters or headers, add new parameters or attributes, rewrite the request URI or access the database. There's basically no limit.

URL Rewriting

When designing RESTful web applications, a common rule is to provide meaningful URIs to the user. For example, our eXist wiki implements a hierarchical document space. The user can directly browse to a document by entering the path to it into the browser's location bar. The URL http://atomic.exist-db.org/HowTo/OxygenXML/eXistXmlRpcChanged will directly lead to the corresponding document.

Internally all document views are handled by the same XQuery script. Above URL will be forwarded to an XQuery called index.xql as follows:

index.xql?feed=HowTo/OxygenXML/&ref=eXistXmlRpcChanged

The XQuery code which does the rewrite magic is shown below:

let $params := subsequence(analyze-string($exist:path, '^/?(.*)/([^/]+)$')//fn:group, 2)
return
    <dispatch xmlns="http://exist.sourceforge.net/NS/exist">
        <forward url="/index.xql">
            <add-parameter name="feed" value="{$params[1]}"/>
            <add-parameter name="ref" value="{$params[2]}"/>
        </forward>
    </dispatch>

The <forward> element tells XQueryURLRewrite to pass the request to the specified URL. The forwarding is done via the RequestDispatcher of the servlet engine and is thus invisible to the user.

Relative URLs within forward or redirect elements are interpreted relative to the request URI, absolute paths relative to the root of the current controller hierarchy. If the controller which processes the request is stored in the database, all absolute and relative paths will be resolved against the database. This is explained in more detail below.

If you want the user to see the rewritten URL, replace the <forward> action with a <redirect>. A common use for <redirect> is to send the user to a default page:

if ($exist:path eq '/') then
    <dispatch xmlns="http://exist.sourceforge.net/NS/exist">
    	<redirect url="index.xml"/>
    </dispatch>

If no action is specified within the dispatch element, the request will be passed through the filter chain and handled normally. The same happens if the action is the <ignore> element. For example, the simplest controller script consist of a single ignore:

eXist's URL rewriting mechanism lets you specify an <error handler> element for each dispatch directive in your controller.xq. NOTE: In eXist-db versions prior to 5.3.0, controller.xql must be used instead.

<ignore xmlns="http://exist.sourceforge.net/NS/exist">
  <cache-control cache="yes"/>
</ignore>

It is important to understand that only one (!) controller will ever be applied to a given request. It is not possible to forward from one controller to another (or to the same). Once you either ignored or forwarded a request in the controller, it will be passed to the servlet which handles it or, if it references a resource, it will be processed by the servlet engine itself. The controller will not be called again for the same request.

Redirects are different in this respect: they cause the client (the web browser) to send a second request. This will again be filtered by XQueryURLRewrite. It is therefore possible to create redirect loops this way!

Variables

Within a controller.xq file, you have access to the entire XQuery function library, including the functions in the request, response and session modules. You could therefore use a function like request:get-uri() to get the current URI of the request. However, to simplify things, XQueryURLRewrite passes a few variables to the controller script:

exist:path

The last part of the request URI after the section leading to the controller.

For instance: If the resource example.xml resides within the same directory as the controller query, $exist:path will be /example.xml.

exist:resource

The section of the URI after the last /, usually pointing to a resource.

For instance: example.xml.

exist:controller

The part of the URI leading to the current controller script.

For example: if the request path is /xquery/test.xql and the controller is in the xquery collection, $exist:controller will contain /xquery.

exist:prefix

If the current controller hierarchy is mapped to a certain path prefix, $exist:prefix returns that prefix.

For example: the default configuration maps the path /tools to a collection in the database (see below). In this case, $exist:prefix would contain /tools.

exist:root

The root of the current controller hierarchy. This may either point to the file system or to a collection in the database. Use this variable to locate resources relative to the root of the application.

For example: assume you want to process a request using stylesheet db2xhtml.xsl, which could either be stored in the /stylesheets directory in the root of the webapp or, if the app is running from within the database, the corresponding /stylesheets collection. You want your app to be able to run from either location. The solution is to use exist:root:

<forward servlet="XSLTServlet">
  <set-attribute name="xslt.stylesheet" value="{$exist:root}/stylesheets/db2xhtml.xsl"/>
</forward>

To summarize: if the request path is /exist/tools/sandbox/get-examples.xql:

$exist:prefix

/tools

$exist:controller

/sandbox

$exist:path

/get-examples.xql

$exist:resource

get-examples.xml

You do not need to explicitly declare the variables or the namespace. However you can add an external declaration for these variables at the top of your XQuery. For instance:

declare variable $exist:path as external;

Locating Controller Scripts and Configuring Base Mappings

By convention, the controller XQueries are called controller.xq.

XQueryURLRewrite will try to guess the path to the controller by looking at the request path.

In fact, one web application may have more than one controller hierarchy. For example, you may want to keep the main webapp within the file system, while some tools and scripts should be served from a database collection. This can be done by configuring two roots within the controller-config.xml file in etc/webapp/WEB-INF. controller-config.xml defines the base mappings used by XQueryURLRewrite.

It basically has two components:

  • <forward> actions which map patterns to servlets

  • <root> elements define the root for a file system or db collection hierarchy

The <forward> tags specify path mappings for common servlets, similar to a servlet mapping in web.xml. The advantage is that XQueryURLRewrite becomes a single point of entry for the entire web application and we don't need to handle any of the servlet paths in the main controller. For example, if we registered a servlet mapping for /rest in web.xml, we would need to make sure that this path is ignored in our main controller.xq. However, if the mapping is done via controller-config.xml, it will already been known to XQueryURLRewrite and we don't need take care of the path in our controller.

The <root> elements define the roots of a directory or database collection hierarchy, mapped to a certain base path. For example, the default controller-config.xml uses two roots:

<!-- 
 Default configuration: main web application is served from the
 webapp directory. 
-->
<root pattern="/tools" path="xmldb:exist:///db/www"/>
<root pattern=".*" path="/"/>

This means that paths starting with /tools will be mapped to the collection hierarchy below /db/www. Everything else is handled by the catch all pattern pointing to the root directory of the webapp (by default corresponding to $EXIST_HOME/etc/webapp). For example, the URI

http://localhost:8080/exist/tools/admin/admin.xql

will be handled by the controller stored in database collection /db/www/admin/ (if there is one) or will directly resolve to /db/www/admin/admin.xql. In this case, all relative or absolute URIs within the controller will be resolved against the database, not the file system. However, there's a possibility to escape this path interpretation, described below.

MVC and Pipelines

XQueryURLRewrite does more than just forward or redirect requests: the response can be processed by passing it to a pipeline of views. "Views" are again just plain Java servlets. The most common use of a view would be to post-processes the XML returned from the primary URL, either through another XQuery or an XSLT stylesheet (XSLTServlet). XQueryURLRewrite passes the HTTP response stream of the previous servlet to the HTTP request received by the next servlet.

Views may also directly exchange information through the use of request attributes (more on that below).

You define a view pipeline by adding a <view> element to the <dispatch> fragment returned by the controller. The <view> element is a wrapper around another sequence of <forward> or <rewrite> actions.

For example, assume we have XML written in docbook format and want to show this as HTML by sending this through an XSLT stylesheet webapp/stylesheets/db2html.xsl. This can be done by returning the following <dispatch> fragment by controller.xq:

<dispatch xmlns="http://exist.sourceforge.net/NS/exist">
  <view>
    <forward servlet="XSLTServlet">
      <set-attribute name="xslt.stylesheet" value="stylesheets/db2html.xsl"/>
    </forward>
  </view>
  <cache-control cache="no"/>
</dispatch>

In this example there's no forwarding action except for the view So the request will be handled by the servlet engine the normal way. The response is then passed to XSLTServlet. A new HTTP POST request is created whose body is set to the response data of the previous step. XSLTServlet gets the path to the stylesheet from the request attribute xslt.stylesheet and applies it to the data.

If any step in the pipeline generates an error or returns an HTTP status code >= 400, the pipeline processing stops and the response is send back to the client immediately. The same happens if the first step returns with an HTTP status 304 (NOT MODIFIED), which indicates that the client can use the version it has cached.

We can also pass a request through more than one view. The following fragment applies two stylesheets in sequence:

if ($name eq 'acronyms.xql') then
	<dispatch xmlns="http://exist.sourceforge.net/NS/exist">
		<!-- query results are passed to XSLT servlet via request attribute -->
		<set-attribute name="xquery.attribute"
			value="model"/>
		<view>
			<forward servlet="XSLTServlet">
				<set-attribute name="xslt.input"
					value="model"/>
				<set-attribute name="xslt.stylesheet"
					value="xquery/stylesheets/acronyms.xsl"/>
			</forward>
			<forward servlet="XSLTServlet">
				<clear-attribute name="xslt.input"/>
				<set-attribute name="xslt.stylesheet" 
					value="stylesheets/db2html.xsl"/>
			</forward>
		</view>
	</dispatch>

The example also demonstrates how information can be passed between actions. XQueryServlet (which is called implicitly because the URL ends with .xql) can save the results of the called XQuery to a request attribute instead of writing them to the HTTP output stream. It does so if it finds a request attribute xquery.attribute, which should contain the name of the attribute the output should be saved to.

In the example above, xquery.attribute is set to model. This causes XQueryServlet to fill the request attribute model with the results of the XQuery it executes. The query result will not be written to the HTTP response as you would normally expect, the HTTP response body will just be empty.

Likewise, XSLTServlet can take its input from a request attribute instead of parsing the HTTP request body. The name of the request attribute should be given in attribute xslt.model. XSLTServlet discards the current request content (which is empty anyway) and uses the data in the attribute's value as input for the transformation process.

XSLTServlet will always write to the HTTP response. The second invocation of XSLTServlet therefore needs to read its input from the HTTP request body which contains the response of the first servlet. Since request attributes are preserved throughout the entire pipeline, we need to clear the xslt.input with an explicit call to clear-attribute.

What benefits does it have to exchange data through request attributes: We save one serialization step: XQueryServlet directly passes the node tree of its output as a valid XQuery value, so XSLTServlet does not need to parse it again.

The advantages become more obvious if you have two or more XQueries which need to exchange information: XQuery 1 can use the XQuery extension function request:set-attribute() to save an arbitrary XQuery sequence to an attribute. XQuery 2 then calls request:get-attribute() to retrieve this value. It can directly access the data passed in from XQuery 1. No time is lost serializing/deserializing the data.

Let's have a look at a more complex example: the XQuery sandbox web application needs to execute a user-supplied XQuery fragment. The results should be retrieved in an asynchronous way, so the user doesn't need to wait and the web interface remains usable.

Older versions of the sandbox used the util:eval function to evaluate the query. However, this has side-effects because util:eval executes the query within the context of another query. Some features like module imports will not work properly this way. To avoid util:eval, the controller code below passes the user-supplied query to XQueryServlet first, then post-processes the returned result and stores it into a session for later use by the ajax frontend:

if (starts-with($path, '/sandbox/execute')) then
    let $query := request:get-parameter("qu", ())
    let $startTime := util:system-time()
    return
        <dispatch xmlns="http://exist.sourceforge.net/NS/exist">
    		<!-- Query is executed by XQueryServlet -->
            <forward servlet="XQueryServlet">
    			<!-- Query is passed via the attribute 'xquery.source' -->
                <set-attribute name="xquery.source" value="{$query}"/>
    			<!-- Results should be written into attribute 'results' -->
    			<set-attribute name="xquery.attribute" value="results"/>
    			<!-- Errors should be passed through instead of terminating the request -->
    			<set-attribute name="xquery.report-errors" value="yes"/>
            </forward>
    		<view>
    			<!-- Post process the result: store it into the HTTP session
    				and return the number of hits only. -->
    			<forward url="session.xql">
    				<clear-attribute name="xquery.source"/>
    				<clear-attribute name="xquery.attribute"/>
    				<set-attribute name="elapsed" 
    					value="{string(seconds-from-duration(util:system-time() - $startTime))}"/>
    			</forward>
    		</view>
        </dispatch>
    (: Retrieve an item from the query results stored in the HTTP session. The
    format of the URL will be /sandbox/results/X, where X is the number of the
    item in the result set :)
    else if (starts-with($path, '/sandbox/results/')) then
    <dispatch xmlns="http://exist.sourceforge.net/NS/exist">
    	<forward url="../session.xql">
    		<add-parameter name="num" value="{$name}"/>
    	</forward>
    </dispatch>

The client passes the user-supplied query string in a request parameter, so the controller has to forward this to XQueryServlet somehow. XQueryServlet has an option to read the XQuery source from a request attribute, xquery.source. The query result will be saved to the attribute results. The second XQuery, session.xql, takes the result and stores it into a HTTP session, returning only the number of hits and the elapsed time.

When called through retrieve, session.xql looks at parameter num and returns the item at the corresponding position from the query results stored in the HTTP session.

Controller XML Format

A controller XQuery is expected to return a single XML element: <dispatch> in the eXist namespace: http://exist.sourceforge.net/NS/exist. <dispatch> may contain a single action element, followed by an optional <view> element. Two action elements are currently allowed:

<redirect>

Redirects the client to another URL, indicating that the other URL must be used for subsequent requests. The URL to redirect to is given in attribute url. A redirect will be visible to the user.

<forward>

Forwards the current request to another request path or servlet. The forwarding is done on the server only (via the RequestDispatcher of the servlet engine). The client can't see where the request was forwarded to.

The request can either be forwarded to a servlet or to another request path, depending on which attribute is specified:

url

The new request path, which will be processed by the servlet engine in the normal way, as if it were directly called. A relative path will be relative to the current request path. Absolute path will be resolved relative to the current web context.

For example, if the current web context is /exist and the supplied attribute reads url="/admin", the resulting path will be /exist/admin.

servlet

The name of a servlet as given in the <servlet-name> element in the corresponding servlet definition of the web descriptor web.xml.

For example, valid names within the eXist standard setup would be XQueryServlet or XSLTServlet.

absolute

To be used in combination with url. If set to "yes", the url will be interpreted as an absolute path within the current servlet context. See below for an example.

method

The HTTP method (POST, GET, PUT ...) to use when passing the request to the pipeline step (does not apply to the first step). This is important if the servlet or URL does not support all methods. The default method for pipeline steps in the view section is always POST.

In addition to the action, an element <cache-control> may appear:

<cache-control>

The cache-control element is used to tell XQueryURLRewrite if the current URL rewrite should be cached. It has a single attribute cache="yes|no". Internally, XQueryURLRewrite keeps a map of input URIs to dispatch rules. With the cache enabled, the controller XQuery only needs to be executed once for every input URI. Subsequent requests will use the cache.

Watch out: only the URL rewrite rule is cached, not the HTTP response. The cache-control setting has nothing to do with the corresponding HTTP cache headers or client-side caching within the browser.

Within an action element, parameters and attributes can be set as follows:

<add-parameter>

Adds (or overwrites) a request parameter. The name of the parameter is taken from attribute name, the value from attribute value. The original HTTP request will be copied before the change is applied. Subsequent steps in the pipeline will not see the parameter.

<set-attribute>

Set a request attribute to the given value. The name of the attribute is read from attribute name, the value from attribute value. You can set arbitrary request attributes, for instance to pass information between XQueries. Some attributes may be reserved by called servlets.

<clear-attribute>

Clears a request attribute. The name of the attribute is read from attribute name. Unlike parameters, request attributes will be visible to subsequent steps in the processing pipeline. They need to be cleared once they are no longer needed.

<set-header>

Sets an HTTP response header field. The HTTP response is shared between all steps in the pipeline, so all following steps will be able to see the change.

Accessing resources not stored in the database

If your controller.xq is stored in a database collection, all relative and/or absolute URIs within the controller will be resolved against the database, not the file system. This can be a problem if you need to access common resources, which should be shared with other applications residing on the file system or in the database.

The <forward> directive accepts an optional attribute absolute="yes|no" to handle this. If one sets absolute="yes", an absolute path (starting with a /) in the url attribute will resolve relative to the current servlet context, not the controller context.

For example, to forward all requests starting with a path /libs/ to a directory within the webapp folder of eXist, you can use the following snippet:

if (starts-with($exist:path, "/libs/")) then
    <dispatch xmlns="http://exist.sourceforge.net/NS/exist">
        <forward url="/{substring-after($exist:path, '/libs/')}" absolute="yes"/>
    </dispatch>

This simply removes the /libs/ prefix and sets absolute="yes", so the path will be resolved relative to the main context of the servlet engine, usually /exist/. In your HTML, you can now write:

<script type="text/javascript" src="/libs/scripts/jquery/jquery-1.7.1.min.js"/>

This will locate the jquery file in webapp/scripts/jquery/..., even if the rest of your application is stored in the db and not on the file system.

Special Attributes Accepted by eXist Servlets

eXist's XQueryServlet as well as the XSLTServlet will listen to a few predefined request attributes. The names of these attributes are listed below and should not be used for other purposes.

XQueryServlet

xquery.attribute

Contains the name of a request attribute. Instead of writing query results to the response output stream, XQueryServlet will store them into the named attribute. The value of the attribute will be an XQuery Sequence (org.exist.xquery.Sequence). If no query results were returned, the attribute will contain an empty sequence.

xquery.source

If set, the value of this attribute must contain the XQuery code to execute. Normally, XQueryServlet reads the XQuery from the file given in the request path. xquery.source is a way to overwrite this behaviour, e.g. if you want to evaluate an XQuery which was generated within the controller.

xquery.module-load-path

The path which will be used for locating modules. This is only relevant in combination with xquery.source and tells the XQuery engine where to look for modules imported by the query. For example, if you stored required modules into the database collection /db/test, you can set xquery.module-load-path to xmldb:exist:///db/test. If the query contains an expression:

import module namespace test="http://exist-db.org/test" at "test.xql";

the XQuery engine will try to find the module test.xql in the filesystem by default, which is not what you want. Setting xquery.module-load-path fixes this.

xquery.report-errors

If set to yes, an error in the XQuery will not result in an HTTP error. Instead, the string message of the error is enclosed in an element <error> which is then written to the response stream. The HTTP status is not changed.

XSLTServlet

xslt.stylesheet

The path to the XSL stylesheet. Relative paths will be resolved against the current request URI, absolute paths against the context of the web application (/exist). To reference a stylesheet which is stored in the database, use an XML:DB URI like xmldb:exist:///db/styles/myxsl.xsl.

xslt.input

Contains the name of a request attribute from which the input to the transformation process should be taken. The input has to be a valid eXist XQuery sequence.

This attribute is usually combined with xquery.attribute provided by XQueryServlet and allows passing data between the two without additional serialization/parsing overhead.

xslt.user

The name of the eXist user to read and apply the stylesheet.

xslt.password

Password for the user given in xslt.user

XSLTServlet will attempt to map all other request attributes starting with the prefix xslt. into stylesheet parameters. So, for example, if you set a request attribute xslt.myattr it will be available within the stylesheet as parameter $xslt.myattr. For security reasons, this is the only way to pass request parameters into the stylesheet: use the controller query to transform the request parameter into a request attribute and pass that to the view.

However, depending on the XSLT engine used, automatic conversion of types between eXist/Java and the XSLT processor may not always work. Best to limit your attribute values to strings.