The CDP preparation scratch area. The 'value' field should be accumulated in such a way that it always contains the value that will eventually be the CDP. This means that for averages, we accumulate 1/n * the PDP, and renormalise when we have unknown PDPs. For the other types it's much easier
RRA - RRD archive This is an archive that holds consolidated data points (CDPs) belonging to a single consolidation function. They are stored in rings buffers, each one related to a single different data-source. It defines the type of consolidation that happens (average, max, min or last), the number of primary data points (PDPs) that go to make a CDP, and the number of CDPs to store.
To better visualize how the datapoints are stored:
│ Datasources ┃ ┃ ┃ └─────────────────┨ Memory ┃ cputime ┃ Consolidators ┃ ┃ ┃ ━━━━━━━━━━━━━━━━━━╋━━━━━━━━━━━━━━━━━╋━━━━━━━━━━━━━━━━━┫ Average ┃ Fring of CDPs ┃ Fring of CDPs ┃ ← RRA ━━━━━━━━━━━━━━━━━━╋━━━━━━━━━━━━━━━━━╋━━━━━━━━━━━━━━━━━┫ Max ┃ Fring of CDPs ┃ Fring of CDPs ┃ ← RRA ━━━━━━━━━━━━━━━━━━┻━━━━━━━━━━━━━━━━━┻━━━━━━━━━━━━━━━━━┛
Add in a new DS into a pre-existing RRD. Preserves data of all the other archives and fills the new one full of NaNs. Note that this doesn't fill in the CDP values correctly at the moment!
Find the RRA with a particular CF that contains a particular start time, and also has a minimum pdp_cnt. If it can't find an appropriate one, either return the RRA with the correct CF that has the most ancient data, or raise No_RRA_Available if there's not archive with the correct CF. Assumes the RRAs are stored in increasing time-length