Nominatim update suddenly very slow

Hi,
We've installed Nominatim with world data on our Kubernetes cluster. It has worked fine for several weeks, but since few days, the update is very very slow (from 4 hours to about 22 hours for the daily update).

There are no visible errors, but it seems to be stucked on the following requests at the beginning and end of ranks 26 and 30:

2012915 | nominatim | | 2019-11-04 09:09:20.100902+00 | active | select geometry_sector,count() from placex where rank_search = $1 and indexed_status > 0 group by geometry_sector order by geometry_sector
2049039 | nominatim | | 2019-11-04 09:09:20.100902+00 | active | select geometry_sector,count() from placex where rank_search = $1 and indexed_status > 0 group by geometry_sector order by geometry_sector
2049040 | nominatim | | 2019-11-04 09:09:20.100902+00 | active | select geometry_sector,count(*) from placex where rank_search = $1 and indexed_status > 0 group by geometry_sector order by geometry_sector

We currently use Nominatim 3.3.0, postgresql 11.2 (on a separate machine) and postgis 2.5.2
Update command : update.php --import-osmosis --osm2pgsql-cache 5000

Postgre:
Database is 700GB on SSD
cpu: 4*2.3Ghz
memory: 14GB

Updater:
cpu: 2*2.3Ghz
memory: 8GB

Postgre custom parameters:
- "shared_buffers=5GB"
- "effective_cache_size=9GB"
- "maintenance_work_mem=128MB"
- "work_mem=32MB"
- "checkpoint_completion_target=0.9"
- "random_page_cost=1.1"
- "effective_io_concurrency=500"

Thanks for any idea about why it became suddenly so slow!