Cache node's `tree_id` by node's PK

[bjester]

Desired behavior

To get the maximum benefit of table partitioning, we'll need to ensure our DB queries for the content node table use the tree_id as a filter so the query planner can prune the partitions it will access while executing the query.

Current behavior

If we don't scope queries to a tree, we often use the node's PK, which won't necessarily get the benefit of partitioning.

Deliverables

• A cache using Redis hashes should be keyed by the node's PK and map to the node's tree ID
• The cache should query the DB if it doesn't contain the information, and populate the cache
• The cache should be updated when the node moves to a new tree (like to the trash tree)
• Any query to the ContentNode table should utilize the cache and filter by tree_id

Suggestions

• Use Redis hashes, and for context why, see an old Instagram engineering blog post about the performance implications

Value add

• We can ensure that queries are structured optimally for the most benefit from our DB partitioning

Possible tradeoffs

• There are some caveats to PKs on partitioned tables which could complicate the whole strategy but perhaps the cache strategy could be leveraged for getting around them.