This topic describes a highly available multi-site setup and the behavior to expect. It outlines the requirements of the high availability architecture and describes the benefits and tradeoffs.
When the data is changed in one {project_name} instance, that data is updated in the database, and an invalidation message is sent to the other site using the `work` cache.
| Multiple {project_name} instances run in each site. If one instance fails some incoming requests might receive an error message or are delayed for some seconds.
| Multiple {jdgserver_name} instances run in each site. If one instance fails, it takes a few seconds for the other nodes to notice the change. Entities are stored in at least two {jdgserver_name} nodes, so a single node failure does not lead to data loss.
| If the {jdgserver_name} cluster fails in one of the sites, {project_name} will not be able to communicate with the external {jdgserver_name} on that site, and the {project_name} service will be unavailable.
The loadbalancer will detect the situation as `/lb-check` returns an error, and will direct all traffic to the other site.
^1^ Recovery point objective, assuming all parts of the setup were healthy at the time this occurred. +
^2^ Recovery time objective. +
^3^ Manual operations needed to restore the degraded setup.
The statement "`No data loss`" depends on the setup not being degraded from previous failures, which includes completing any pending manual operations to resynchronize the state between the sites.
Manual operations that re-synchronize the {jdgserver_name} state between the sites will issue a full state transfer which will put a stress on the system.
Two sites restriction::
This setup is tested and supported only with two sites.
Each additional site increases overall latency as it is necessary for data to be synchronously written to each site.
Furthermore, the probability of network failures, and therefore downtime, also increases. Therefore, we do not support more than two sites as we believe it would lead to a deployment with inferior stability and performance.
A synchronously replicated database ensures that data written in one site is always available in the other site after site failures and no data is lost.
It also ensures that the next request will not return stale data, independent on which site it is served.
A synchronously replicated {jdgserver_name} ensures that cached data in one site are always available on the other site after a site failure and no data is lost.
For synchronous database replication and synchronous {jdgserver_name} replication, a low latency is necessary as each request can have potentially multiple interactions between the sites when data is updated which would amplify the latency.
Is a synchronous cluster less stable than an asynchronous cluster?::
An asynchronous setup would handle network failures between the sites gracefully, while the synchronous setup would delay requests and will throw errors to the caller where the asynchronous setup would have deferred the writes to {jdgserver_name} or the database on the other site.
However, as the two sites would never be fully up-to-date, this setup could lead to data loss during failures.
// TODO storing sessions in Infinispan is experimental. Add this bullet point back when we support it
//* Lost logouts, meaning sessions are logged in one site although they are logged out in the other site at the point of failure when using an asynchronous {jdgserver_name} replication of sessions.
* Lost changes leading to users being able to log in with an old password because database changes are not replicated to the other site at the point of failure when using an asynchronous database.
* Invalid caches leading to users being able to log in with an old password because invalidating caches are not propagated at the point of failure to the other site when using an asynchronous {jdgserver_name} replication.
Therefore, tradeoffs exist between high availability and consistency. The focus of this topic is to prioritize consistency over availability with {project_name}.
