On 10/20/2014 11:02 PM, jesper@krogh.cc wrote:
>>> >>I do suspect the majority is from 30 concurrent processes updating an
>>> >>506GB GIN index, but it would be nice to confirm that. There is also a
>>> >>message-queue in the DB with a fairly high turnaround.
>> >
>> >A 506GB GIN index? Uh, interesting :). What's it used for? Trigrams?
> It is for full-text-search, but it is being updated entirely regulary,
> ~100M records. A dump/restore cycle typically reduces the size to 30-40%
> of current size.
Try 9.4 beta. The on-disk format of GIN indexes was rewritten in 9.4,
making them a lot smaller. That might help with WAL volume too. Or not,
but I'd love to hear what the impact is, in a real-life database :-).
- Heikki