add cache buster hash to snapshot names

[sudeepdino008]

reasons to do it:
#17368 (comment)
#17368 (comment)
#17344 (comment)

[sudeepdino008]

@anacrolix assigning to you since i saw some PR about this

[anacrolix]

I started on this, but progress is early days, so let me know if anyone else has ideas.

[anacrolix]

@taratorio per our chat to make you aware of this

[sudeepdino008]

I started on this, but progress is early days, so let me know if anyone else has ideas.

so in general our snapshot name generation and parsing is a bit of mess:

• it's all over the place; for blocks snapshots, it's some global function to parse file names
• for state files, name generators are members of domain/history etc. making it difficult to use in tests; that's why db/states tests have duplicated name gen logic all over the place
• then we added more complex logic with versions which can be wildcard; and we had things like FindFilesWithVerdionMask etc all over.
• now we adding cache buster - 2 wildcards; things are about to get more complicated.
• index names go hand in hand with data file names

I had something called snap_schema which i created during forkables. I even use it in some db/state tests and publishable command - it does make things easier and general to work with.

However when i tried to get claude work on pushing snap_schema in more places, I realized that snap_schema was at a wrong level. It is created per entity-- so account history has a different schema and storage domain has a different schema. Then we can make a collection of schema, and call it e3schema.
But it doesn't suit us because very often we don't know what type of file we're parsing. Specially in downloader, we just get filename and we need to parse and give a snapinfo. The problem was chicken and egg - to know which snap schema to use, i need the type, for which i need to parse, for which i need snap schema...

I aborted the attempt. Now that I think back, maybe the problem could be solved by having a more intelligent collector, which knows which snap schema to route to.

EDIT:
I think you can probably look at snap_schema and see what concerns it addresses. In particular I remember that it wrapped schema for both the data file and all index files, allowed for version wildcard, knew about which folder data and index files are (e.g. domain kvi are in domain/ folder, but history indexes are in accessors/ ).
Maybe those things would be useful properties to have in whatever design you're think about too.

[VBulikov]

anacrolix added Imp2 as Importance

[anacrolix]

My run suggested most of the parsing leaves an ideal place to put the atom, that for file name parsing works easily.

For writing it's more involved, but also should involve screening at the webseed upload, and potentially not even using the traditional file name on webseeds and providing names only in the chain.toml (content addressed on the remote).

[anacrolix]

#19150 was the start of my attempt.