Check builds for repo.osgeo.org for 502 errors, it seems like it has improved
Migrating away from H2, PostGIS testcontainers
Thanks for moving away from H2, it is a mess:
Broke compatibility between versions, no update path for files on disk … like at all
(thus no longer suitable for our use as an easy on disk local database…)
H2 was removed from core, but is still used in a few extensions, okay a lot of extensions and tests, …
How to address use in Tests
Move to GeoPackage where possible
Move to HSQL where possible
Aside: There is a new H2Store written, but it conflicts with the old H2 package names, so we would have to remove all the others first, so this is not an option until “the end”… which would result in double work.
However SQL view tests, spatial joins tests, … these do not work!
So test with PostgreSQL and TestContainers
These are slower to start, 1-2 seconds, …
… but would it be skipped if Docker is not available?
But we could isolate it in a profile like OnlineTests?
However OnlineTests only enable if config is available, so you can choose what to test locally
Could make profile on by default, so you can disable if you have docker around for another use, …
Test containers has a hidden file you can provide to ask test containers to never unload, so you do not pay startup time each time you build locally
Suggestion: use a docker system variable that controls whether we run those tests or skip them, on by default (and also check if docker/podman are around)
Do not clean up PostgreSQL container, so it can be reused, across modules, …
OSGeo repo migration
Did this go ahead? Yes … it went ahead after our release.
Runs REST requests, checks results, runs OWS requests to follow up and check that the results are working. Basically, like CITE tests but for GeoServer specific functionality
REST API is affected again (sigh) dropping the ability to match a content type with an extension (example featuresTypes.xml)
So we need to role our own for this functionality, Andrea has a prototype but is not confident it is the best approach
The expectation is that this may take a few days, but there may be more surprises in the 25 modules, …
failures start on restconfig, and gwc-rest, …
assume remaining 25 modules are going to fail on rest interaction (like importer-rest,…)
So this is on the table for now, an option to allocate someone, either within the GS3 budget, or with an extension to the activity. Expect this to be a developer rather than a full sprint …
Unknowns:
GeoServer Security 7 is not looked at yet
(a mix of spring security 6 + spring 7 seems to work)
GeoServer Cloud is in scope for GS3, and this this would need to update Spring Boot 4
Step 1: use an AI agent to compare the current with doc and amend
this would be a solid update
Step 2: use an AI agent to add annotations
caution: automatic generation does not just rely on annotations, it also requires the structures, we do not use basic jackson on object beans (which is what swagger expects)
We use xstream and converters, so the objects do not clearly map to swagger structure
Yes it went out, thanks Gabe and others for the release…
Two security vulnerabilities announced at the same time
historical (ie old) that in some cases have been fixed for some time … like 2.25.x era
Build server stability:
we ran out of disk space during foss4g code sprint
Andrea asked it to clear assembly zip files after each build (for each branch being built), …
Most builds currently default to cleanup after 30 days, which is fine for normal operation but can accumulate when running a lot in a short period of time.
There is an option to add a “max number of artifacts to keep” we could enable to avoid this in the future
FOSS4G Auckland
relaxed event, 400 people
smoothly organized
Jody made way too many workshops and presentations (4 workshops and a lot of presentations)
GT/GS intro ones went fine
BOM issues found during the GT workshops
Intellij changed enough that one cannot follow easily (new class now takes full.package.name.with.Class rather than two fields)
GS3 developer one more difficult, quickstart/setup for GS are hard to get into
GN5 harder, too much changing
Presentations went fine, vector tiles a bit more difficult than others, but it was important for GS to be in the room for both cloud native formats and OGCAPI tracks
Interviewed by a podcast, mapscaping, on security with kindness presentation
OSGeo lacking volunteers, need to switch to funding