We have a use case for wide columnar data, used for mostly performance analytics. There are many types of events that share same columns, but mostly total unique columns are 20x of a typical events columns. My use case is filtering by some boolean logic and aggregating them. I use Elasticsearch with a mapping that does not tokenize text fields, (all strings are keyword type) and it works very well. I can add nodes as I want and adjust shards with ease with spot instances and make it cost effective.
I have not found a way/tool to replace this. Many of the tools fail at dynamic data with cardinality. Wanted to use Clickhouse like this, adding columns as they are discovered but it did not go well, but it has been a while. Also adding replicas is not as easy as Elasticsearch.
Does anyone have a similar use case implemented with Clickhouse when data is not known before hand and unfolds as time goes by?
I have not found a way/tool to replace this. Many of the tools fail at dynamic data with cardinality. Wanted to use Clickhouse like this, adding columns as they are discovered but it did not go well, but it has been a while. Also adding replicas is not as easy as Elasticsearch.
Does anyone have a similar use case implemented with Clickhouse when data is not known before hand and unfolds as time goes by?