Prepare For Load Balancing and Failover

For a high scale or highly available deployment, you can prepare a pool of IG servers with nearly identical configurations, and then load balance requests across the pool, routing around any servers that become unavailable. Load balancing allows the service to handle more load.

Before you spread requests across multiple servers, however, you must determine what to do with state information that IG saves in the context, or retrieves locally from the IG server system. If information is retrieved locally, then consider setting up failover. If one server becomes unavailable, another server in the pool can take its place. The benefit of failover is that a server failure can be invisible to client applications.

IG saves state information in the following ways:

  • By using a handler, such as a SamlFederationHandler or a custom ScriptableHandler, that can store information in the context. Most handlers depend on information in the context, some of which is first stored by IG.

  • By using filters, such as AssignmentFilters, HeaderFilters, OAuth2ClientFilters, OAuth2ResourceServerFilters, ScriptableFilters, SqlAttributesFilters, and StaticRequestFilters, that can store information in the context. Most filters depend on information in the request, response, or context, some of which is first stored by IG.

IG retrieves information locally in the following ways:

  • By using filters and handlers, such as FileAttributesFilters, ScriptableFilters, ScriptableHandlers, and SqlAttributesFilters, that depend on local system files or container configuration.

By default, the context data, including storage of the default session implementation, resides in memory. For information about whether to store session data on the user-agent instead, see JwtSession.

When using JwtSession with a cookie domain, share the encryption keys and the signature symmetric secret across all IG configurations so that any server can read or update JWT cookies from any other server in the same cookie domain.

If your data does not fit in an HTTP cookie, for example, because when encrypted it is larger than 4 KB, consider storing a reference in the cookie, and then retrieve the data by using another filter. IG logs warning messages if the JwtSession cookie is too large. Using a reference can also work when a server becomes unavailable, and the load balancer must fail requests over to another server in the pool.

If some data attached to a context must be stored on the server-side, then you have additional configuration steps to perform for session stickiness and for session replication. Session stickiness means that the load balancer sends all requests from the same client session to the same server. Session stickiness helps to ensure that a client request goes to the server holding the original session data. Session replication involves writing session data either to other servers or to a data store, so that if one server goes down, other servers can read the session data and continue processing. Session replication helps when one server fails, allowing another server to take its place without having to start the session over again. If you set up session stickiness but not session replication, when a server crashes, the client session information for that server is lost, and the client must start again with a new session.