Cluster
From CauchoWiki
Cluster configuration documentation is at http://www.caucho.com/resin-3.0/config/balance.xtp.
Contents |
[edit] <cluster> directive
| directive | description | default |
|---|---|---|
| id | unique identifier of the cluster | "" |
| cluster-ref | use the values from a cluster-definition directive | |
| client-live-time | how long the load-balancer should keep an idle socket open | 30s |
| client-dead-time | how long the load-balancer should treat a backend as dead | 30s |
| client-read-timeout | timeout of load-balancer reads | 30s |
| client-write-timeout | timeout of load-balancer writes | |
| <srun> | configures an srun port |
[edit] <cluster-definition> directive
| directive | description | default |
|---|---|---|
| id | unique identifier of the cluster definition | "" |
| client-live-time | how long the load-balancer should keep an idle socket open | 30s |
| client-dead-time | how long the load-balancer should treat a backend as dead | 30s |
| client-read-timeout | timeout of load-balancer reads | 30s |
| client-write-timeout | timeout of load-balancer writes | |
| <srun> | configures an srun port |
[edit] example
Clustering and load balancing is configured in a <cluster> directive inside the <server> block of the resin.conf.
A cluster with two machines might look like:
[edit] resin.conf
<resin xmlns="http://caucho.com/ns/resin"> ... <server> <cluster> <client-live-time>30s</client-live-time> <srun id="a" host="192.168.0.10" port="6802"/> <srun id="b" host="192.168.0.11" port="6803"/> </cluster> ... </server> </resin>
[edit] Load Balancing Timeouts
Load balancing has two sets of time outs, one set from the client (front-end) perspective, and one set from the server (back-end) perspective.
The client timeouts are configured directly in the <cluster> directive. The server timeouts are configured in the individual <srun> directives.
| client-read-timeout | How long the frontend should wait on a read before timing out | 60s |
| client-write-timeout | How long the frontend should wait on a write before timing out | 60s |
| client-live-time | How long the frontend should keep a socket alive before discarding it | 60s |
| client-dead-time | How long the frontend should treat a backend as dead before retrying | 15s |
[edit] resin.conf
<cluster> <client-live-time>60s</client-live-time> <srun id="a" host="192.168.0.10" port="6802"/> </cluster>
The server timeouts specify how long the backend will wait for new data from a frontend before giving up. This timeout should be longer than the client-live-time.
| read-timeout | How long the backend should wait on a read before timing out | 30s |
| write-timeout | How long the backend should wait on a write before timing out | 30s |
[edit] resin.conf
<cluster> ... <srun id="c" host="196.168.0.14" port="6802" read-timeout="120s"/> ... </cluster>
