I'm glad that you asked: https://gitlab.com/soapbox-pub/eclipse/-/snippets/4782163

The most important one is this:

Table "public.bsky_blocks"

Column | Type | Collation | Nullable | Default

--------+---------------+-----------+----------+---------

cid | character(59) | | not null |

bytes | bytea | | not null |

They store everything as CBOR-encoded binary values, content-addressable by their IPFS (actually IPLD) CID.

However, they still want to be able to _query_ this data...

So they end up storing everything in bytes AND in plaintext in the database.

Not only that, they also store binary _diffs_ of every single user action, making everything take up 2x more space.