Architecture - Amazon Timestream


Amazon Timestream for Live Analytics has been designed from the ground up to collect, store, and process time series data at scale. Its serverless architecture supports fully decoupled data ingestion, storage, and query processing systems that can scale independently. This design simplifies each subsystem, making it easier to achieve unwavering reliability, eliminate scaling bottlenecks, and reduce the chances of correlated system failures. Each of these factors becomes more important as the system scales.

Write architecture

When writing time-series data, Amazon Timestream for Live Analytics routes writes for a table, partition, to a fault-tolerant memory store instance that processes high throughput data writes. The memory store in turn achieves durability in a separate storage system that replicates the data across three Availability Zones (AZs). Replication is quorum based such that the loss of nodes, or an entire AZ, will not disrupt write availability. In near real-time, other in-memory storage nodes sync to the data in order to serve queries. The reader replica nodes span AZs as well, to ensure high read availability.

Timestream for Live Analytics supports writing data directly into the magnetic store, for applications generating lower throughput late-arriving data. Late-arriving data is data with a timestamp earlier than the current time. Similar to the high throughput writes in the memory store, the data written into the magnetic store is replicated across three AZs and the replication is quorum based.

Whether data is written to the memory or magnetic store, Timestream for Live Analytics automatically indexes and partitions data before writing it to storage. A single Timestream for Live Analytics table may have hundreds, thousands, or even millions of partitions. Individual partitions do not, directly, communicate with each other and do not share any data (shared-nothing architecture). Instead, the partitioning of a table is tracked through a highly available partition tracking and indexing service. This provides another separation of concerns designed specifically to minimize the effect of failures in the system and make correlated failures much less likely.

Storage architecture

When data is stored in Timestream for Live Analytics, data is organized in time order as well as across time based on context attributes written with the data. Having a partitioning scheme that divides "space" in addition to time is important for massively scaling a time series system. This is because most time series data is written at or around the current time. As a result, partitioning by time alone does not do a good job of distributing write traffic or allowing for effective pruning of data at query time. This is important for extreme scale time series processing, and it has allowed Timestream for Live Analytics to scale orders of magnitude higher than the other leading systems out there today in serverless fashion. The resulting partitions are referred to as "tiles" because they represent divisions of a two-dimensional space (which are designed to be of a similar size). Timestream for Live Analytics tables start out as a single partition (tile), and then split in the spatial dimension as throughput requires. When tiles reach a certain size, they then split in the time dimension in order to achieve better read parallelism as the data size grows.

Timestream for Live Analytics is designed to automatically manage the lifecycle of time series data. Timestream for Live Analytics offers two data stores—an in-memory store and a cost-effective magnetic store. It also supports configuring table-level policies to automatically transfer data across stores. Incoming high throughput data writes land in the memory store where data is optimized for writes, as well as reads performed around current time for powering dashboard and alerting type queries. When the main time frame for writes, alerting, and dashboarding needs has passed, allowing the data to automatically flow from the memory store to the magnetic store to optimize cost. Timestream for Live Analytics allows setting a data retention policy on the memory store for this purpose. Data writes for late-arriving data are directly written into the magnetic store.

Once the data is available in the magnetic store (because of expiration of the memory store retention period or because of direct writes into the magnetic store), it is reorganized into a format that is highly optimized for large volume data reads. The magnetic store also has a data retention policy that may be configured if there is a time threshold where the data outlives its usefulness. When the data exceeds the time range defined for the magnetic store retention policy, it is automatically removed. Therefore, with Timestream for Live Analytics, other than some configuration, the data lifecycle management occurs seamlessly behind the scenes.

Query architecture

Timestream for Live Analytics queries are expressed in a SQL grammar that has extensions for time series-specific support (time series-specific data types and functions), so the learning curve is easy for developers already familiar with SQL. Queries are then processed by an adaptive, distributed query engine that uses metadata from the tile tracking and indexing service to seamlessly access and combine data across data stores at the time the query is issued. This makes for an experience that resonates well with customers as it collapses many of the Rube Goldberg complexities into a simple and familiar database abstraction.

Queries are run by a dedicated fleet of workers where the number of workers enlisted to run a given query is determined by query complexity and data size. Performance for complex queries over large datasets is achieved through massive parallelism, both on the query runtime fleet and the storage fleets of the system. The ability to analyze massive amounts of data quickly and efficiently is one of the greatest strengths of Timestream for Live Analytics. A single query that runs over terabytes or even petabytes of data might have thousands of machines working on it all at the same time.

Cellular architecture

To ensure that Timestream for Live Analytics can offer virtually infinite scale for your applications, while simultaneously ensuring 99.99% availability, the system is also designed using a cellular architecture. Rather than scaling the system as a whole, Timestream for Live Analytics segments into multiple smaller copies of itself, referred to as cells. This allows cells to be tested at full scale, and prevents a system problem in one cell from affecting activity in any other cells in a given region. While Timestream for Live Analytics is designed to support multiple cells per region, consider the following fictitious scenario, in which there are 2 cells in a region.

In the scenario depicted above, the data ingestion and query requests are first processed by the discovery endpoint for data ingestion and query, respectively. Then, the discovery endpoint identifies the cell containing the customer data, and directs the request to the appropriate ingestion or query endpoint for that cell. When using the SDKs, these endpoint management tasks are transparently handled for you.


When using VPC endpoints with Timestream for Live Analytics or directly accessing REST API operations for Timestream for Live Analytics, you will need to interact directly with the cellular endpoints. For guidance on how to do so, see VPC Endpoints for instructions on how to set up VPC endpoints, and Endpoint Discovery Pattern for instructions on direct invocation of the REST API operations.