Coyote Point Systems E350GX, Installation And Administration Manual

Page: 149 / 344

Managing Servers
Equalizer Installation and Administration Guide 149
Maximum Connections Limits, Responders, and Hot Spares
When a maximum connections limit is set on all the servers in a cluster, it is often desirable to define either a
responder or a hot spare server for the cluster, so that any attempted connections to the cluster that occur after the
max connections
limit has been reached are directed to the responder or hot spare instead of being refused or sent
to the server anyway because of a persistent connection.
In general, a Responder is easier to configure than setting up a separate server as a hot spare, since the responder
runs on Equalizer. However, while Responders are capable of returning only a single HTML page, a hot spare can be
configured to return multiple HTML pages and images. See the section “Automatic Cluster Responders” on page
152 for information on configuring a responder.
To use a hot spare, you would usually configure it on Equalizer as follows:
• Set
max connections
to zero (0), so that all connection requests sent to the hot spare are accepted.
• Enable the
hot spare
flag. This specifies that any requests refused by all the other servers in the cluster
because they reached their
max connections
limit (or are down) will be forwarded to the hot spare server.
• Enable the
dont persist
flag so that connections made to the hot spare don’t persist. Each connection to the
cluster must first be load balanced amongst the other servers in the cluster and only go to the hot spare if all
the other servers have reached their
max connections
limit.
Interaction of Server Options and Connection Processing
Server option settings have a direct influence on connection and request processing, particularly Layer 4 and Layer 7
persistence. (Note that persistence is set at the cluster level, but can be disabled for individual servers using the
dont
persist
option.) The hierarchy of server option settings is shown in the table below:
Shutting Down a Server Gracefully
To avoid interrupting user sessions, make sure that a server to be shut down or deleted from a cluster no longer has
any active connections. When a server’s initial weight is zero, Equalizer will not send new requests to that server.
server disabled
(initial weight = 0)
An initial weight of 0 tells Equalizer that no traffic should be sent to
the server, disabling the server. This option setting takes precedence
over all other options (including persistence, hot spare, etc.).
max connections > 0
If set to a non-zero value, Equalizer limits the total number of
simultaneously open connections to the server to that value. This limit is
not overridden if the
hot spare option is enabled on a server, and is not
overidden by a Layer 4 sticky record or Layer 7 persistence cookie for
the server in an incoming request.
quiesce enabled
The server is not included in load balancing decisions, so that no new
connections will be made to this server. If a request in an incoming
connection has an existing Layer 4 sticky record or Layer 7 cookie for a
server, however, the request will be sent to that server even when
quiesce is enabled. [Note that if dont persist is also enabled on the
server, the sticky record or cookie is ignored.]
hot spare enabled
The server is not included in load balancing decisions, so that traffic is
sent to this server only when no other server in the cluster is available
to accept client connections. If a request in an incoming connection has
an existing Layer 4 sticky record or Layer 7 cookie for a server, however,
the request will be sent to that server even when
hot spare is enabled.
[Note that if dont persist is also enabled on the server, the sticky
record or cookie is ignored.]