We are upgrading from ElasticSearch 5.x to 6.x
Currently we have 8 language specific indexes with 4 mapping types in each. Approximately 30K documents, which are not evenly distributed between the 8 indexes (certain indexes are exponentially more popular for content).
As a result of the breaking changes:
and a review of the following guidance piece:
It seems that of my three options
- 32 indexes of which many will only have less than a hundred documents
- 4 type indexes with mixed languages
- 8 language indexes with a custom type specifier and a bloated mapping
Option 3 is the most viable because of the following reasons - option 1 results in paying a high sharding cost for what are sparsely populated indexes, option 2 searching/analyzing a mixed language index is still discouraged much like it was back in 2.x and though Option 3 results in a bloated mapping, which might be a pain from a maintainability point of view, but has no performance pitfalls.
Is my reasoning, conclusions sound? Or have I misunderstood the documentation provided? Thanks a ton.