Perform the tasks in the following order when starting a system where this product is installed.
Figure 8.1 Sequence for starting the system where this product is installed
When mounting the DFS, mount the DFS management server first, then mount the DFS clients.
For this reason, first start the master servers, which are the DFS management server. Start the primary master server and then the secondary master server when operating using a master server replicated configuration.
After starting the master servers, start the slave servers, development servers, and collaboration servers, which are the DFS clients.
When operating an existing system where the collaboration servers are already running, only mounting of the DFS on the collaboration server needs to be performed.
Note
When using replicated configuration for the master server, the system is not running until the DFS management server has started on both the primary master server and the secondary master server.
For this reason, start the secondary master server system without waiting for the primary master server system to finish starting.
If DFS mounting is not set to automatic when starting the servers, mount the DFS manually after the servers have started.
After DFS mounting is finished on all servers, use bdpp_start on the master servers to start the Hadoop of this product (refer to "A.14 bdpp_start" for details).
The following processes start when Hadoop is started:
Master server
JobTracker (Hadoop)
Slave server
TaskTracker (Hadoop)
Note
The "JobTracker" of Apache Hadoop on the master server cannot be started directly - you must start the Hadoop of this product using bdpp_start.
Refer to the Hadoop Project Top Page (http://hadoop.apache.org/) for details on how to use Apache Hadoop.
Information
Set whether to mount the DFS automatically in /etc/fstab. Mounting is automatic when the DFS mount options are as follows:
On a master server: The noatrc option has not been specified
On any other server: The noauto option has not been specified
