Hi,
On 2016-03-16 14:06:35 +0300, Dmitriy Sarafannikov wrote:
> Sorry, autovacuum not involved in this.
> I took several xids from this files and found them in WAL with pg_xlogdump:
> rmgr: Heap len (rec/tot): 67/ 99, tx: 2918822231, lsn:
> 4983/D8FF8388, prev 4983/D8FF8338, bkp: 0000, desc: insert: rel
> 1663/16428/133587958; tid 5737/114
> rmgr: Btree len (rec/tot): 34/ 66, tx: 2918822231, lsn:
> 4983/D8FF83F0, prev 4983/D8FF8388, bkp: 0000, desc: insert: rel
> 1663/16428/133587964; tid 6560/2
> rmgr: Btree len (rec/tot): 50/ 82, tx: 2918822231, lsn:
> 4983/D8FF8438, prev 4983/D8FF83F0, bkp: 0000, desc: insert: rel
> 1663/16428/133926198; tid 27418/14
> rmgr: Btree len (rec/tot): 50/ 82, tx: 2918822231, lsn:
> 4983/D8FF8490, prev 4983/D8FF8438, bkp: 0000, desc: insert: rel
> 1663/16428/148286684; tid 14048/2
>
> i noticed that they are points to table, that filled by trigger functions
> with upsert.
Hm, interesting. I wonder, could you try this with a source checkout of
9.4? There's some fixes in the upcoming minor release that might be
related. If not, I'll look into it.
- Andres