You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When performing an ELB deployment for very large areas (e.g., NYSDOT), the time to load a graph can take quite some time. Currently, the "builder" EC2 instance (i.e., the first one that is started up to build the graph) must complete its graph loading before other instances can start up. This adds substantial time to the deployment. Instead, it may be a good idea to kick off the other servers as soon as the builder instance has uploaded the Graph.obj file to S3. This way, all of the servers can essentially start loading the graph in parallel.
If I were to be a devil's advocate, I would say that we actually don't want to kick off the other instances until we know that the Graph.obj can successfully be loaded by the first OTP instance. However, I think once the graph builds successfully, we can rest assured that it will be loaded with little issue. Even if it were to fail, the other instances would fail just the same -- so I'm not sure if anything is gained by waiting for the first instance to load.
The text was updated successfully, but these errors were encountered:
When performing an ELB deployment for very large areas (e.g., NYSDOT), the time to load a graph can take quite some time. Currently, the "builder" EC2 instance (i.e., the first one that is started up to build the graph) must complete its graph loading before other instances can start up. This adds substantial time to the deployment. Instead, it may be a good idea to kick off the other servers as soon as the builder instance has uploaded the
Graph.obj
file to S3. This way, all of the servers can essentially start loading the graph in parallel.If I were to be a devil's advocate, I would say that we actually don't want to kick off the other instances until we know that the
Graph.obj
can successfully be loaded by the first OTP instance. However, I think once the graph builds successfully, we can rest assured that it will be loaded with little issue. Even if it were to fail, the other instances would fail just the same -- so I'm not sure if anything is gained by waiting for the first instance to load.The text was updated successfully, but these errors were encountered: