Hi, I’d like to use PG as an analytics engine on multiple separately created read-only datasets. The datasets, in my case, are independently created by bringing up a local PG instance, populating a table, and shutting down the PG instance. I do have control over both the creation and the query processes.
After some research, I realize this is not a standard use-case, so I’m open to development work in the internals as required. I’d appreciate some advice/guidance on possible approaches. 1. For dealing with independently created tables, I’m planning to have them organized in tablespaces during creation, let’s say one per table. In the PG instance where querying is performed, I can use symlinks to the relevant tablespaces from the data directory, and modify system catalogs appropriately for PG to be able to work with those tables. Is this a viable approach? Are there better alternatives? Any pitfalls I should expect? 1. From my research so far, it seems PG wouldn’t work out-of-the-box if those tablespaces are read-only. (Even assuming various services like vacuum, etc. are all turned off.) In some documentation/slides, it’s mentioned that even in case of only the select queries PG might write some metadata into the table’s pages (hints, or something like that.) Is this correct? Or is there a way to prevent all writes to the tables without modifying PG code? For my education, where can I find some info on those “hints”? (Would also appreciate pointers to code.) If it’s not possible to work with readonly tables without code changes, what would be the possible approaches for development? Would appreciate any advice/guidance/recommendations. Thanks in advance!
