• Review existing indexes on the tally_snapshots table.
• Create a new partial index that can help with read and doesn't affect write process or update existing indexes.
• Check if runInTransaction=false helps since CREATE INDEX CONCURRENTLY cannot run inside a transaction and having this avoids locking issues on the 198M row table. https://docs.liquibase.com/reference-guide/changelog-attributes/runintransaction 
• Ensure the index aligns with current query patterns and filtering conditions. Not sure if adding snapshot_date in index will be helpful.
• Include migration and rollback support for index changes.
• Since is_primary defaults to false and index is WHERE true, the index starts empty (should be fast even with 198M)

Example:

 <changeSet id="" author="" runInTransaction="false">
    <sql>
      CREATE INDEX CONCURRENTLY IF NOT EXISTS idx_tally_snapshots_primary
      ON tally_snapshots (org_id, product_id, granularity)
      WHERE is_primary = true;
    </sql>
    <rollback>
      DROP INDEX IF EXISTS idx_tally_snapshots_primary;
    </rollback>
  </changeSet> 

We can also try the same locally, by directly doing below for testing purpose

 CREATE INDEX CONCURRENTLY IF NOT EXISTS idx_tally_snapshots_primary
  ON tally_snapshots (org_id, product_id, granularity)
  WHERE is_primary_record = true;

AC:

• Index is added to the table
• No functional impact to existing queries or data.