Discover a ways to split HAF image into 2 ones: one containing Hived+SQL serializer and second holdinga Postresql instance + preinstalled HFM extension
Above setup would allow to use HAF on 2 smaller machines instead of one bigger. Another benefit is natural stop of Hived-based part and still accessing postgres instance for some maintenance reasons/data analysis etc.
Problems to analyze/solve:
- we need to support single image deployment still due to simplicity like also testing codebase requirements
- image sanity must be preserved: whole HAF is built using 2 main parts: Hived+sql_serializer and very tightly coupled hfm extension and query governor. Both HFM and querygovernor parts are deployed directly into PostgreSQL installation directories, so in new scheme - to separate image
- still support automatic upgrade when only some runtime code is changing and there is no schema changes