Miriam tried to run a risk computation on the new cluster and trashed the system: the hazard getters were taking more than 128Gb of RAM and swapping a lot. Similar problems happened in the past for the scenario damage calculator: the solution was to change the data structure (the gmf table) to avoid having to aggregate several rows at the same location. Here we are trying to see if the same solution can be applied to the event_based calculator. Here the data structure is much more complex. The plan is to introduce a new table gmf_agg to store the aggregated values, so that the risk calculator does not need to perform the aggregation again (and hundreds of times, depending on the number of the workers). The gmf_agg table is populated once for all in the post_processing phase of the hazard calculator.
See https:/