Scaling Upsource services in the cluster
You should consider scaling Upsource services if you experience server performance issues because of increasing number of projects or users, or you want your distributed system to be fault-tolerant 24/7.
The following services can be scaled:
Scaling frontend
To scale frontend, run:
./cluster.sh -H tcp://<swarm.master.host>:<swarm.master.port> scale frontend=NScaling PSI-agent
To scale PSI agent, run:
./cluster.sh -H tcp://<swarm.master.host>:<swarm.master.port> scale psi-agent=NScaling analyzer
The default setup provides for a single analyzer that covers all Upsource projects. Let's say you have ten projects, nine of which are small to medium-scale projects that see several commits a day, while the last one is a huge project with hundreds of commits every day. Overall indexing performance might be improved by moving that resource-intensive project to a separate analyzer, relieving the original analyzer from stress.
To scale analyzer, run:
./cluster.sh -H tcp://<swarm.master.host>:<swarm.master.port> scale analyzer=NNew analyzer nodes will be detected automatically, and newly added projects will be assigned to the least-loaded node.
Inaccessible nodes will be removed automatically as well, and the projects assigned to them will be distributed to other (active) nodes.
Services that cannot be scaled
- cassandra
- opscenter
- haproxy
- PSI broker
These are singleton services and therefore cannot be scaled. Scaling of file-clustering is also not available, as even a single service can support very high loads of over 100 projects.