Agraph notes

There are several important things to note about Agraph deployments.

Split pipeline

In a Parallel Forge Agraph deployment it is necessary to split the data prior to executing the parallel Forge. The Deployment Template uses Forge to accomplish this. It first runs a single Forge on all the data with a rollover element specified in the pipeline which tells Forge to split the data into a number of pieces. Each one of those individual pieces of data is then used as incoming data for its corresponding Parallel Forge client. In some cases some other means of splitting the data may be used or the data may come from its source pre-split. In these cases the split Forge step can be removed and replaced with the alternate splitting method.

Multiple Agidxs

The default Agraph deployment has a single Agidx specified. This is because all of the Dgidx processes are set to run on the same machine. In a more complex deployment you may have Dgidx processes running on a number of machines. This would require you to define an Agidx for each machine that has one or more Dgidx processes running to it. Each Agidx process would run in serial and would specify the previous Agidx processes' output as part of its input.

Agraph and Dgraph restart groups

Each Agraph and Dgraph can specify its own restart group. In most cases, an Agraph and all of its children Dgraphs should specify the same restart group. There is currently no technical restriction requiring this but it is recommended to ensure standard expected graph restart behavior.