The process of replicating data from the cluster to the individual node, bringing the node into sync with the cluster, is known as provisioning. There are two methods available in Galera Cluster to provision nodes:
In a State Snapshot Transfer (SST), the cluster provisions nodes by transferring a full data copy from one node to another. When a new node joins the cluster, the new node initiates a State Snapshot Transfer to synchronize its data with a node that is already part of the cluster.
You can choose from two conceptually different approaches in Galera Cluster to transfer a state from one database to another:
Logical This method uses mysqldump. It requires that you fully initialize the receiving server and ready it to accept connections before the transfer.
This is a blocking method. The donor node becomes READ-ONLY for the duration of the transfer. The State Snapshot Transfer applies the FLUSH TABLES WITH READ LOCK command on the donor node.
mysqldump is the slowest method for State Snapshot Transfers. This can be an issue in a loaded cluster.
Physical This method uses rsync, rsync_wan, xtrabackup and other methods and copies the data files directly from server to server. It requires that you initialize the receiving server after the transfer.
This method is faster than mysqldump, but they have certain limitations. You can only use them on server startup. The receiving server requires very similar configurations to the donor, (for example, both servers must use the same innodb_file_per_table value).
Some of these methods, such as xtrabackup can be made non-blocking on the donor. They are supported through a scriptable SST interface.
For more information on the particular methods available for State Snapshot Transfers, see the Comparison of State Transfer Methods.
You can set which State Snapshot Transfer method a node uses from the confirmation file. For example:
In an Incremental State Transfer (IST), the cluster provisions a node by identifying the missing transactions on the joiner and sends them only, instead of the entire state.
This provisioning method is only available under certain conditions:
When these conditions are met, the donor node transfers the missing transactions alone, replaying them in order until the joiner catches up with the cluster.
For example, say that you have a node in your cluster that falls behind the cluster. This node carries a node state that reads:
Meanwhile, the current node state on the cluster reads:
The donor node on the cluster receives the state transfer request from the joiner node. It checks its write-set cache for the sequence number 197223. If that seqno is not available in the write-set cache, a State Snapshot Transfer initiates. If that seqno is available in the write-set cache, the donor node sends the commits from 197223 through to 201913 to the joiner, instead of the full state.
The advantage of Incremental State Transfers is that they can dramatically speed up the reemerging of a node to the cluster. Additionally, the process is non-blocking on the donor.
The most important parameter for Incremental State Transfers is gcache.size on the donor node. This controls how much space you allocate in system memory for caching write-sets. The more space available the more write-sets you can store. The more write-sets you can store the wider the seqno gaps you can close through Incremental State Transfers.
On the other hand, if the write-set cache is much larger than the size of your database state, Incremental State Transfers begun less efficient than sending a state snapshot.
Galera Cluster stores write-sets in a special cache called the Write-set Cache, or GCache. GCache cache is a memory allocator for write-sets. Its primary purpose is to minimize the write-set footprint on the RAM. Galera Cluster improves upon this through the offload write-set storage to disk.
The write-set cache employs three types of storage:
Permanent In-Memory Store Here write-sets allocate using the default memory allocator for the operating system. This is useful in systems that have spare RAM. The store has a hard size limit.
By default it is disabled.
Permanent Ring-Buffer File Here write-sets pre-allocate to disk during cache initialization. This is intended as the main write-set store.
By default, its size is 128Mb.
On-Demand Page Store Here write-sets allocate to memory-mapped page files during runtime as necessary.
By default, its size is 128Mb, but can be larger if it needs to store a larger write-set. The size of the page store is limited by the free disk space. By default, Galera Cluster deletes page files when not in use, but you can set a limit on the total size of the page files to keep.
When all other stores are disabled, at least one page file remains present on disk.
For more information on parameters that control write-set caching, see the gcache.* parameters on Galera Parameters.
Galera Cluster uses an allocation algorithm that attempts to store write-sets in the above order. That is, first it attempts to use permanent in-memory store. If there is not enough space for the write-set, it attempts to store to the permanent ring-buffer file. The page store always succeeds, unless the write-set is larger than the available disk space.
By default, the write-set cache allocates files in the working directory of the process. You can specify a dedicated location for write-set caching, using the gcache.dir parameter.
Given that all cache files are memory-mapped, the write-set caching process may appear to use more memory than it actually does.