Thanks Steven.
I understand of course that there are limitations to what Tumblr will allow its developer community to know about its underlying workings and have never
expected the full picture (sigh). The API then is a partial revealing - those hooks into the innards Tumblr allows the developer community to use. As to my question about
how Tumblr thinks of its own product, from a kind of owner/designer/inventor/proprietor point of view, that's probably way out of line, but I didn't mean it as a way of
prying into secrets for some competitive reason, I'm just curious about everything. So I will [attempt to] cease prying into everything except the API which I think I need
now to begin investigating.
The scope of the problem of handling all the data involved in such a product as Tumblr I have actually been thinking about from a DB point of view.
I have speculated, e.g., that you MUST have a database engine, that in that lingo, you have a table of users and a table of unique posts (regardless of how often that post occurs).
So your user table probably points to a xref table which links the users to their
list of individual posts This is the table with all the volatility, and the HUGE one,
in terms of number of rows, that starts, for a post, when the post first enters the 'system'
-undoubtedly thru submission - I can't think of any other input - submission from outside (altho I have an inkling there's
more than one way to do this provided by TUmblr)
and has a huge number of rows but each row is not that large - the unique key of the user,
the unique key of the post, probably the date of the user's posting of the post (by reblogging or submission), not sure what else. Maybe more info for initial submissions.
Do the submitters go in the list of users, e.g.? I'd want to index that guy, the user/post xref, by user and by post id's. And I forgot tags. They'd be in the posts
table, not the xref table, at first glance.
Anyway we're talking, for the xref table, if it were one table, probably billions of rows. There would no need to have just one xref table though - the 'virtual' xref table
could be split up into hundreds of individual tables. I've never been involved w/ a database that large - tables w/ 100M rows max.
Anyway this is all speculation which is a lot of fun to write and do, but probably a lot less to read, especially given my tendency to logorrhea.
If you managed to get this far thx!
Thanks again!
Regards,
Ken