Cassandra Frequently asked questions (Ii.) __cassandra

9 can I speed up my large number of writes through bulk submissions?

No, using a bulk commit only leads to a deferred spike, replaces it with an asynchronous insert, or uses a true "bulk load"
Batch update for the same partition key is an exception, as long as a batch size is maintained within a reasonable range, there is still good, but remember not to blindly use the bulk.

10. In Red Hat Enterprise Edition (RHEL), nodes cannot be added to the cluster.
Check to see if the SELinux is turned on and turn it off.

11, how do I unsubscribe from the Apache Cassandra mailing list
Send an email to user-unsubscribe@cassandra.apache.org

12. Why use the top command to see Cassandra using more memory than the Java Virtual machine limits for internal storage.

Cassandra Internal use of memory-mapped files (mmap). That is, we use the operating system's virtual memory system to map a certain number of disk files to the Cassandra process's address space. This will use virtual memory, the address space, which can be found with some tools such as the top command, but in 64-bit operating systems, the virtual address space is infinite, so you don't have to worry about that.

What is the meaning of the memory that is used? It is the memory that is actually occupied with BRC () or mmap, and the point is that the file for mmap does not necessarily reside permanently in memory. So the memory used is just a cache, just as the system IO would use the kernel page cache.

The difference between general IO and Mmap is that the memory of the MMAP engine is mapped to a process, so you can see the virtual memory size of the process with the top command. One of the benefits of mmap than IO is that as long as you have it in memory, you can read it inside, without touching page fault (read the system page cache, the kernel may need to do semi-context switching.) ), more details reference http://www.varnish-cache.org/trac/wiki/ArchitectNotes

13, what is the seed node.

The seed node is used to discover the entire cluster when the node is started.

If you configure your node to point to several nodes as the seed, then the nodes in your cluster will send gossip information to the seed node more than the non seed nodes. In other words: the seed node corresponds to the hub of the gossip network, and each node can quickly know the state of the other nodes through the seed node.

When a node is newly joined to a cluster, you need to specify a seed node to discover other nodes in the cluster, and when you add a new node to the cluster, you need to specify at least one surviving seed node to connect, and once a node joins the cluster, it learns the other nodes, He doesn't need the seed node the next time it starts.

You can set any node as a seed at any time, there is nothing special about the seed node, as long as you write it to the seed list, he is a seed node.

The seed is not bootable (bootstrap) (because if a node adds itself to the seed list, that means he needs to transmit the data himself, and if you need to, you can start the node as a non-seeding node and then add it to the seed list when you start. Of course, if you're initializing a new cluster without any data, you don't have to worry about it.

Remind me of the following two points:
Select more than two nodes for each data center as a seed node
Keep all your nodes in the same seed list.

Turn from: http://zhaoyanblog.com/archives/960.html