Unlogged tables, persistent kind

Unlogged tables are a good new feature.

I noticed Bruce had mentioned they were the equivalent of NoSQL, which
I don't really accept. I guess it depends upon whether you mean NoSQL
for caches (e.g. memcached) or NoSQL for useful datastores (e.g.
Mongo). It seems worth discussin now before we get too far into the
marketing hype around Beta.

If you don't log changes to tables you have two choices if we crash
1) truncate the table and any indexes
2) rebuild any indexes damaged by the crash

Currently, we do (1). That certainly has its place but most data
stores don't do this if they crash, since it would lead to data loss.
Not just a couple of rows either - serious, major data loss if you put
the wrong kind of data in it. We even delete data that has been safely
on disk for weeks, months, which IMHO some people could easily get
confused about.

In the future, I would like to work on (2), which preserves as much
data as possible, while recognising indexes may be damaged. I don't
really have any name for this, since the current naming seems to
assume there is only one kind of unlogged table.

My implementation path for that would be to add a crash_number onto
pg_control and pg_index. Any index marked as "unlogged, persistent"
would only be usable if it's crash number is the same as current
system crash number.

REINDEX would update the index crash number to current value. That
also allows us to imagine a "repair index" command in the future as
well.

Heap blocks would be zeroed if they were found to be damaged, following a crash.

--
 Simon Riggs                   http://www.2ndQuadrant.com/
 PostgreSQL Development, 24x7 Support, Training & Services