Обсуждение: number of tables limited over time (not simultaneous)?

We've settled upon a method for gathering raw statistics from widely
scattered data centers of creating one sequence per-event, per minute.

Each process (some lapp, some shell, some python, some perl etc)  can
call a shell script which calls ssh->psql to execute a nextval('event')
sequence.  Periodically (every 2-10 minutes, depending on other factors)
Another process picks up the value and inserts it into a permanent home.

We're only talking a few 7-10k calls per minute, but going to this from
a query that does an update has saved a *huge* amount of overhead.

If I needed to a periodic dump and restore would only take a minute.
This data is highly transient.  More frequently than biweekly or so
would be annoying though.

Aside from security concerns, did we miss something?  Should I be
worried we're going through ~60,000 sequences per day?

TIA,
dave

dave crane wrote:
> We've settled upon a method for gathering raw statistics from widely
> scattered data centers of creating one sequence per-event, per minute.

> Aside from security concerns, did we miss something?  Should I be
> worried we're going through ~60,000 sequences per day?

Well, you'll want to make sure you're vacuuming the system tables
regularly. And you might want to make sure your code can retry creating
the sequences once OIDs wrap around.

Not sure I'd have thought of doing things this way - I'd probably have
totalled figures outside the database and then summarised them once a
minute.

--
   Richard Huxton
   Archonet Ltd