Compose Cache
TDengine uses an insert-driven cache administration plan, often known as initial in, initially out (FIFO). This plan differs from go through-driven “minimum not too long ago made use of (LRU)” cache administration. A FIFO plan shops the most recent facts in cache and flushes the oldest details from cache to disk when the cache use reaches a threshold. In IoT use instances, the most recent info or the current state is most critical. The cache policy in TDengine, like Significantly of the design and architecture of TDengine, is predicated on the character of IoT info.
Once you produce a database, you can configure the scale of your generate cache on Each individual vnode. The vgroups parameter decides the amount of vgroups that procedure details during the database, and the buffer parameter decides the size with the compose cache for every vnode.
Generate Databases db0 VGROUPS 100 BUFFER 16MB
In theory, larger cache dimensions are generally much better. Having said that, at a particular issue, it turns into impossible to enhance overall performance by raising cache size. In many situations, you could keep the default cache settings.
Study Cache
If you develop a databases, you could configure whether or not the latest information from every single subtable is cached. To do so, set the cachemodel parameter as follows:
none: Caching is disabled.
last_row: The latest row of data in Every subtable is cached. This option considerably improves the performance of the LAST_ROW functionality.
last_value: The most up-to-date non-null price in Every single column of each subtable is cached. This selection appreciably enhances the overall performance of the final operate in usual conditions, such as In which, Get BY, Team BY, and INTERVAL statements.
both equally: Rows and columns are both of those cached. This option is similar to simultaneously enabling possibility last_row and last_value.
Metadata Cache
To enhance question and publish effectiveness, Every single vnode caches the metadata that it gets. Once you make a database, you may configure the size on the metadata cache in the webpages and pagesize parameters.
Make read more Databases db0 Internet pages 128 PAGESIZE 16kb
The previous SQL assertion creates 128 pages on Every single vnode in the db0 databases. Each webpage provides a 16 KB metadata cache.
File Technique Cache
TDengine implements facts trustworthiness by the use of a generate-in advance log (WAL). Producing facts to your WAL is basically writing data into the disk within an ordered, append-only fashion. For that reason, the file program cache performs a significant position in create overall performance. When you develop a databases, You should utilize the wal parameter to decide on bigger functionality or higher dependability.
1: This feature writes to the WAL but isn't going to help fsync. New info published to the WAL is stored inside the file process cache although not published to disk. This provides much better functionality.
2: This option writes towards the WAL and enables fsync. New data written towards the WAL is immediately written to disk. This presents greater data reliability.
Client Cache
In combination with the server-side caching talked over Earlier, the core shopper library libtaos.so also tends to make usage of caching. TDengine Consumer caches the metadata of all databases, supertables, and subtables that it's got accessed, along with the cluster topology.
If a customer modifies particular metadata even though multiple customers are concurrently accessing a TDengine cluster, the metadata caches on Just about every shopper may fail or turn into out of sync. If this occurs, operate the reset query cache command on the afflicted purchasers to pressure them to obtain contemporary metadata and reset their caches.