Yes - severe bloat on this toast table is quite true. I'm trying to get things righted.
A bit more info:
* autovacuum was on, runtime was weeks...
* maintenace_work_mem is higher than 1G
* there is a huge amount of mutation & updates
* ballpark is 500M rows
* after 'looping' on the vacuum of this toast, i see:
relname | n_live_tup | n_dead_tup----------------+------------+------------pg_toast_17283 | 173145066 | 2461775286
Main question now: is there actually new work getting done when it repeats? ie. am i making progress?
thanks in advance!
-gleeco
On Tue, Jan 10, 2017 at 5:04 AM, Tom Lane <tgl@xxxxxxxxxxxxx> wrote:
gleeco <gleeco@xxxxxxxxx> writes:
> hello - i've a long-running manual vacuum on a big size table. It's many
> TB db that has some toast. What is odd is that it keeps looping back on
> the same toast / index, eg.
It's not looping ... it's filling up your maintenance_work_mem (which is
evidently set to 1GB) with dead TIDs and then pausing to do a cycle of
index cleaning. You have evidently let that table get, um, severely
bloated. Was autovacuum off?
regards, tom lane
-gleeco
