Use the following set of configuration settings to enforce your content integrity policy.
This option allows an administrator to force a second replica of a stream to be written to another node prior to returning a success status to the client. If either write operation fails, the client receives an error status.
Normally, when a stream is written to a node, the replication constraints are checked during the normal course of health checking all streams on the node. While the behavior should be acceptable for most installations, there may be business requirements that immediately duplicate the stream to another node as quickly as possible. The
scsp.replicateOnWrite (ROW) option forces a second immediate copy of an object to another node. If the stream has a constraint for creating more than two copies, the additional copies are made during the normal health checking process.
During normal operations, the Health Processor automatically creates two stream replicas in the background on a regular basis. Using
scsp.replicateOnWrite, all stream replicas are created during the initial write. While this process may temporarily restrict client responsiveness as streams are created and replicated at the same time, your streams are protected from a single volume failure right away.
This option allows an administrator to control how aggressively to handle asynchronous replications of an object after it is initially written into the cluster.
The default setting is 2, which makes the replication secondary to handling user requests. Setting the parameter to 1 gives the replication task equal priority with user requests.
scsp.replicateOnWrite option can be used to make replication synchronous – the highest priority.
This option allows an administrator to require validation of all content reads before returning a successful read completion.
Although this option can be specified on a per-read basis, setting the value to 1 in the configuration file forces all reads to use validation.
During the read from the disk, the content hash is computed. If the hash is wrong, indicating logical disk corruption, the socket will be closed before the last block is transmitted, forcing an error to the client.
Note: Using this option creates additional CPU load on the node.