Lower memory overhead of offline cache database

[kkaefer]

Apart from improving SQLite handling in https://github.com/mapbox/mapbox-gl-native/issues/6334, we can also change the schema of our caching database to be more space-efficient.

• Move URL template to its own table (e.g. tileset) to avoid duplicating it for every tile
• Use more efficient ETag encoding: Many CloudFront/S3 responses use an MD5 hash wrapped in quotes. Instead of storing this as ASCII/UTF-8, we could special case this to go from 34 bytes to 16 bytes (micro-optimization, not sure this is worth it)
• Separate layers in a vector tile into their own database entries: This would allow us to load on the layers we need, and avoid spending time on I/O + zlib inflation for unused data. This could also have implications on what page size we use (currently default 4K) to avoid wasting too much space in overflow pages.
• Change epoch baseline for timestamp fields: We currently use the standard UNIX epoch, which means that most of the numbers we store require 5 bytes for storing them in SQLite's varint format. We could move up the epoch to save a few bytes for each of the three timestamp fields (expires, modified, accessed)
• Compress bitfields: compressed, must_revalidate to one column
• Remove id field for tiles (although that probably doesn't make sense since SQLite automatically creates a rowid table for every table that is shared with the INTEGER NOT NULL PRIMARY KEY AUTOINCREMENT.
• Use LZFSE as the compression format

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[stale[bot]]

This issue has been automatically detected as stale because it has not had recent activity and will be archived. Thank you for your contributions.

[stale[bot]]

This issue has been automatically detected as stale because it has not had recent activity and will be archived. Thank you for your contributions.