Servers and clusters

Algolia’s classic infrastructure runs directly on the hardware without virtualization. This is sometimes called bare metal.

When you create a new Algolia application with the Build, Grow, and Premium plans, it’ll be hosted on classic infrastructure. For more information, see Pricing.

Clusters

Your application is hosted on a cluster of three identical servers. This ensures a high availability: if one or two servers go down, your users can still search, while indexing operations are paused until all three servers are online again. For more information see Availability over consistency.

Having three copies of your data also makes data recovery easier if there’s a hardware failure.

This redundancy guarantees an SLA of 99.99% availability.

Resiliency

To increase resiliency of your clusters, Algolia implements geographical and network redundancies:

• To prevent single events, such as fires or floods, from bringing down the entire cluster, Algolia separates the servers into different data centers.

• To prevent single points of network failures, no server within the same cluster uses the same internet service provider.

Consensus

All three servers in a cluster have the same data. To prevent inconsistencies, Algolia uses the Raft algorithm to ensure that all servers in a cluster update simultaneously. This ensures that each server always has the same data without service interruption.

Scaling beyond clusters

By default, your Algolia application shares a cluster with other applications. You can request dedicated infrastructure as an add on. For more information, see Pricing.

You can add more servers to your cluster. This adds capacity to handle requests, and reduces network latency if you add servers in a region closer to your users. For more information, see Distributed Search Network.

For guidelines about working with a large amount of data, see Large indexing jobs.

What happens when a server goes down

Servers may fail to respond for a number of reasons:

• Temporary network issues
• High server load
• Hardware failures

If a server goes down, Algolia handles search and indexing requests differently.

Search requests

If one or two servers go down, the others will still handle search requests without interruption. For search, consistency is less important. As long as one server is able to handle the request, your users still get search results.

Indexing requests

While one server might be unreachable to the other servers within the cluster, it might still receive requests from the outside. In that case, the other two servers continue handling requests, finding consensus among themselves. On unreachable servers, indexing jobs are queued. If the third server becomes reachable again, it can synchronize with the other two.

Availability over consistency

Hosted services need to find a balance between:

• Availability guarantees constant access to your data without service outage.
• Consistency guarantees the same data everywhere at the same time, for example, all users get the same search results simultaneously.

For more information, see CAP theorem and eventual consistency.

Algolia chooses availability over consistency.

Because when users search, they should always get results. This is more important than small differences between searches. Under normal load, synchronizing data between three servers takes seconds or less, so that you only infrequently experience data discrepancies.

That’s why all three servers in a cluster respond to search requests, so that at least one server is always available. If Algolia were to delay search requests until all three servers have the same data (consistency), it might cause large delays, leading to an unresponsive search.

Location of Algolia infrastructure (classic)

When you create new Algolia applications with classic infrastructure, you can choose a region where your new application will be hosted. Some regions aren’t available for all plans. For more details, see Pricing.