> On Dec 10, 2016, at 6:25 AM, Tom DalPozzo <t.dalpozzo@gmail.com> wrote: > > Hi, > you're right, VACUUM FULL recovered the space, completely. > So, at this point I'm worried about my needs. > I cannot issue vacuum full as I read it locks the table. > In my DB, I (would) need to have a table with one bigint id field+ 10 bytea fields, 100 bytes long each (more or less, not fixed). > 5/10000 rows maximum, but let's say 5000. > As traffic I can suppose 10000 updates per row per day (spread over groups of hours; each update involving two of those fields, randomly. > Also rows are chosen randomly (in my test I used a block of 2000 just to try one possibility). > So, it's a total of 50 millions updates per day, hence (50millions * 100 bytes *2 fields updated) 10Gbytes net per day. > I'm afraid it's not possible, according to my results. > Reagrds > Pupillo >
Are each of the updates visible to a user or read/analyzed by another activity? If not you can do most of the update in memory and flush a snapshot periodically to the database.
