The DynamicHTTPMetadataProvider fetches entity metadata just-in-time from a remote HTTP server. The metadata request URL is constructed by applying a transform to the entityID. The transform strategy is configured in a child element.
Metadata is cached in memory subject to a complex set of interacting settings and the cache indicators within the metadata itself, and also can be saved to disk and reloaded back into memory at reload or startup time to restore the state of the cache. This isn't a fully redundant safety net but can be used as part of an overall strategy to reduce the risk of relying on remote sources in real-time. Ultimately, remote sources have to be bulletproof or there will be outages. This can be mitigated but not fully eliminated as a risk.
The DynamicHTTPMetadataProvider is used with remote metadata. See the MetadataManagementBestPractices topic for more information. |
Contents
The <MetadataProvider> element and the type DynamicHTTPMetadataProvider are defined by the urn:mace:shibboleth:2.0:metadata schema, which can be located at http://shibboleth.net/schema/idp/shibboleth-metadata.xsd.
Any of the Common Attributes, the Dynamic Attributes, or the HTTP Attributes may be configured.
Configure the Dynamic Attributes for the desired cache behavior. In particular, the minCacheDuration attribute and/or the maxCacheDuration attribute should be adjusted based on the life cycle of the metadata. Note that the cacheDuration attribute in metadata (if any) contributes to the overall cache behavior.
The following HTTP attributes are exclusive to the DynamicHTTPMetadataProvider type:
| Name | Type | Default | Description |
|---|---|---|---|
maxConnectionsTotal 3.3 | Integer | 100 | The maximum total number of simultaneous connections allowed by the HTTP client's connection pool manager. This attribute is incompatible with httpClientRef. |
maxConnectionsPerRoute 3.3 | Integer | 100 | The maximum number of simultaneous connections per route allowed by the HTTP client's connection pool manager. This attribute is incompatible with httpClientRef. |
| List of String (comma-separated) | "application/samlmetadata+xml, application/xml, text/xml" | The MIME types supported by this provider when requesting metadata from the HTTP server. The |
For a DynamicHTTPMetadataProvider, the HTTP Connection Attributes have aggressive default timeout values:
| Name | Default |
|---|---|
connectionRequestTimeout | PT5S (5 seconds) |
connectionTimeout | PT5S (5 seconds) |
socketTimeout | PT5S (5 seconds) |
These may be tightened further if desired.
By default, a DynamicHTTPMetadataProvider caches metadata in memory (httpCaching="memory"). The default values of the HTTP Caching Attributes are optimized for numerous, relatively small metadata files (i.e., single entity descriptors).
| A file cache will not survive a restart and so there is little (if any) benefit in overriding the default in-memory caching strategy. |
Any of the following child elements may be specified (in order).
| Name | Cardinality | Description |
|---|---|---|
<MetadataFilter> | 0 or more | A metadata filter applied to candidate metadata as it flows through the metadata pipeline |
<TLSTrustEngine>3.1 | 0 or 1 | A custom TrustEngine used to evaluate TLS server certificates. This element conflicts with and is overridden by the httpClientSecurityParametersRef attribute. |
<MetadataQueryProtocol> | 0 or 1 | Constructs the metadata request URL based on the requirements of the Metadata Query Protocol |
<Template> | 0 or 1 | Constructs the metadata request URL by means of a simple transform based on substitution |
<Regex> | 0 or 1 | Constructs the metadata request URL by means of a complex transform based on a regular expression |
The <MetadataFilter> child element is common to all metadata providers while the <TLSTrustEngine> child element is common to all HTTP metadata providers. The remaining child elements are exclusive to the DynamicHTTPMetadataProvider type.
At most one of the <MetadataQueryProtocol>, <Template>, or <Regex> child elements is allowed. If none are configured, the provider constructs the metadata request URL directly from the entityID. This corresponds to the "well-known location" mechanism defined in the SAML 2.0 Metadata specification, section 4.1.
If you forget to configure a child element, the provider will default to the well-known location strategy. This constrains the entityID to be an URL (not an URN) but the provider does not check the URL scheme. If the scheme on the entityID is "http:", the metadata exchange will be vulnerable to a man-in-the-middle attack. For this reason, the well-known location strategy should be avoided in most cases. |
If the <MetadataQueryProtocol> child element is used, the metadata request URL is constructed according to the SAML Profile for the Metadata Query Protocol, which itself is based on the Metadata Query Protocol specification. The content of the <MetadataQueryProtocol> child element will be used as the "Base URL" defined in that specification.
The <MetadataQueryProtocol> child element has the following optional attribute:
| Name | Type | Default | Description |
|---|---|---|---|
transformRef | Bean ID | A reference to a transform function for the entityID. If used, the child element must be empty. |
The transformRef attribute may be used if (and only if) the child element is empty (i.e., it has no content).
If the <Template> child element is used, the metadata request URL is constructed by means of a simple transform. Specifically, the value of the entityID is substituted into the template parameter "${entityID}".
The <Template> child element has the following attributes:
| Name | Type | Default | Description |
|---|---|---|---|
encodingStyle 3.4 | "none", "form", "path", or "fragment" | "form" | Determines whether and how the entityID value will be URL encoded prior to replacement. Allowed values are:
The precise definition of these terms is defined in the documentation for the methods of the Guava library's UrlEscapers class. |
| Boolean | true | Deprecated. Use 'encodingStyle instead as of v3.4. If the element contains an encoded attribute set to "false", the value will be replaced directly, otherwise it will be URL form encoded. |
transformRef | Bean ID | A reference to a transform function for the entityID. If used, the child element must be empty. | |
velocityEngine | Bean ID | shibboleth.VelocityEngine | This attribute may be used to specify the name of the Velocity engine defined within the application. |
The transformRef attribute may be used if (and only if) the child element is empty (i.e., it has no content).
If the <Regex> child element is used, the metadata request URL is constructed by means of a complex transform. The entityID value is first matched against the regular expression contained in the <Regex> element's match attribute. Then, the <Regex> element's content is treated as a replacement expression to run based on the results of the match.
The <Regex> child element has the following required attribute:
| Name | Type | Default | Description |
|---|---|---|---|
match | String | required | A regular expression against which the entityID is evaluated. |
Note that only numeric/positional group references (e.g., $1) are supported.
A typical use case is to load entity metadata dynamically from a metadata query server (i.e., a server that supports the Metadata Query Protocol). Here is a complete example:
Note that the <MetadataQueryProtocol> child element encodes the base URL of the Metadata Query Protocol. For example, consider the following child element:
<!-- Specify the base URL for the Metadata Query Protocol --> <MetadataQueryProtocol>https://mdq.example.org/global/</MetadataQueryProtocol> |
The previous <MetadataQueryProtocol> child element is equivalent to the following <Template> child element:
<!--
The Template element specifies a simple template with a single parameter.
By default, the entityID is percent-encoded before substitution.
-->
<Template>https://mdq.example.org/global/entities/${entityID}</Template> |
The above configuration explicitly formulates an MDQ protocol URL. This example is for illustration purposes only. If the server supports the Metadata Query Protocol, a <MetadataQueryProtocol> child element should be used instead. This intentionally hides the details of the Metadata Query Protocol.
Finally, here is an example of the well-known location strategy:
<MetadataProvider id="WellKnownEntityMetadata" xsi:type="DynamicHTTPMetadataProvider">
<!--
Use the well-known location strategy to get SP metadata. The
entityID is not configured here; it is determined from the
AuthnRequest's Issuer element, as sent by the requester.
In this case, the entityID MUST be in the form of a URL (rather
than a URN). It is STRONGLY RECOMMENDED that https URLs be used
to protect against man-in-the-middle attacks.
-->
</MetadataProvider> |
A DynamicHTTPMetadataProvider fetches entity metadata as needed. We say that the IdP queries for SP metadata just-in-time.
Compare this to a FileBackedHTTPMetadataProvider that batch loads all of the entity descriptors in a metadata file whether or not the individual entity descriptors are actually needed. In contrast, a DynamicHTTPMetadataProvider loads exactly those entities that are needed—no more, no less. In this sense, a DynamicHTTPMetadataProvider is much more efficient.
OTOH, all metadata query protocols are synchronous protocols by definition. Basically the IdP is blocked until it obtains the metadata it needs.
When an IdP receives a SAML protocol request from a particular SP, the IdP must first obtain entity metadata for that SP. If the IdP has no such metadata in its possession, metadata resolution proceeds sequentially according to a configured chain of metadata providers. Upon encountering a DynamicHTTPMetadataProvider in the chain, the IdP consults an HTTP client that acts as an intermediary between the IdP and the query server.
The HTTP client implements a shared HTTP cache. (RFC 7234) If the desired metadata is already cached, and the stored response is fresh, the client immediately returns the cached metadata to the IdP. Otherwise the client issues an HTTP request to the query server. Upon receiving a response from the server, the client caches the response and finally returns the metadata to the IdP.
In either case, the IdP parses the metadata and applies any metadata filters configured on the DynamicHTTPMetadataProvider. The metadata that ultimately emerges from the configured metadata pipeline is cached locally (in memory) for future use.
The next time the IdP receives a SAML protocol request from this SP, it again traverses the chain of providers until it encounters the DynamicHTTPMetadataProvider. This time, however, the IdP does not bother to consult the HTTP client since the needed metadata is in the IdP’s local cache.
The IdP’s local cache is governed by the Dynamic Attributes. In particular, the minCacheDuration and maxCacheDuration attributes strongly influence the life cycle of metadata in the local cache. Any cacheDuration and validUntil attributes in the metadata itself also influence the behavior of the local cache.
Yes, by default the HTTP client caches responses in memory (httpCaching=”memory”). Consequently, two copies of each entity descriptor reside in memory, one managed by the HTTP client and another one managed directly by the IdP.
The HTTP client may be configured for file caching but a file cache will not survive a restart so the overall benefit of file caching is reduced. In most cases, a memory cache is preferred, at least for systems with adequate memory,
Yes, the HTTP client supports HTTP conditional GET (RFC 7232) for optimal performance but the inner workings of the HTTP client are opaque to the IdP. If the IdP does in fact consult the HTTP client, and the client returns metadata to the IdP, the IdP blindly parses the metadata and applies the metadata filters. There are no optimizations implemented on the IdP side.
When the HTTP client sends an HTTP request to a metadata query server, the SAML protocol exchange is blocked until a response is received from the server and returned to the IdP. If the client reports a failed request, the IdP continues with the next provider in the configured chain of providers. If the offending DynamicHTTPMetadataProvider is the last provider in the chain, metadata resolution fails.
There are at least three things you can do to help minimize the impact of metadata query:
Configure minCacheDuration and/or maxCacheDuration
Configure the HTTP Connection Attributes
Configure a robust chain of metadata providers
As noted above, the minCacheDuration and maxCacheDuration attributes strongly influence the life cycle of metadata in the local cache. The goal is to avoid needless interaction with the HTTP server. To achieve this goal, you need to understand the life cycle of the metadata on the server. For this reason, it is best to ask your federation operator for specific recommendations.
OTOH, the federation operator may influence the life cycle of metadata in the IdP’s local cache by including a cacheDuration attribute in the metadata itself. In that case, the deployer has fewer configuration options to consider, by design.
The HTTP Connection Attributes include the following attributes:
connectionRequestTimeout (default: PT5S): The maximum amount of time to wait for a connection to be returned from the HTTP client's connection pool manager.
connectionTimeout (default: PT5S): The maximum amount of time to wait to establish a connection with the remote server.
socketTimeout (default: PT5S): The maximum amount of time to wait between two consecutive packets while reading from the socket connected to the remote server.
As noted above, each of these attributes defaults to 5 seconds. You may want to tighten these timeout values further, depending on what you know about the route to the server or the server itself.
Regardless of the IdP configuration or the service-level agreement you have with the server operator, things will go wrong. One thing you can do to hedge your bets is to deploy a local query server as backup. Alternatively, one or more high-value SPs can be pre-loaded into memory.