I have a bunch of time-series indices (one per week) with a lot of deleted documents (because older data gets aggregated):
According to this I should not force-merge them when I expect there to be writes (more aggregation) later.
Apparently the alternative is reindexing.
What is the most sensible strategy to do this regularly? Because there seems to be no way to rename indexes, I can only come up with somewhat convoluted solutions:
a) Have an alias (e.g.
histograms-2017-33) for every index. When initially creating the actual index, name it for example
histograms-2017-33-v1. When reindexing, reindex to
histograms-2017-33-v2 and then point the alias to the new version.
b) Reindex to
histograms-2017-33-temp and then delete
histograms-2017-33 and reindex again back to
c) Something entirely different?