Geoserver WFS-T performance degraded after upgrading to 2.20

Question

After upgrading from GeoServer 2.16.1 to 2.20.4 we noticed that large WFS-T requests are getting stuck.

We have a JS OpenLayers client that sends inserts/updates/deletes via WFS-T in chunks of 50 features (so we don't exceed max request size), which get applied to a single table on a PostGIS database. Both GeoServer and PostGIS are running in docker containers (kartoza/geoserver:2.20.4 and postgis/postgis:12-3.0-alpine).
Since updating, transactions with 50 features get stuck. Changing the chunksize to 20 'solves' the issue, but is much slower.

On version 2.16.1, the first chunk takes around 2s, and each following chunk ~0.5s.
On version 2.20.4, the first chunk takes 3.5s, and each following chunk ~2s.

Combined with changing chunk size to 20, saving changes now takes 5-10 times longer.

Were any settings added in recent GeoServer releases that might cause this? Or are there other settings we can play with to improve performance?

Answer 1

May be related to this change: https://osgeo-org.atlassian.net/browse/GEOS-10265

I'm sure the change improved a lot large transactions with tends of thousands of insertions (from tens of minutes down to seconds).

At the datastore configuration, do you have:

• Prepared statements enabled (important)
• A large-ish batch size (e.g. 100 or more?) Both are important for sustained insertion performance.