On Mon, 3 May 2021 at 10:09, Andrea Aime <andrea.a...@geo-solutions.it> wrote:
> Hi, > I was looking at GitHub action checks, that often fail on network > transfers related to Maven repo. > One thing that caught my eye is that I keep on see messages like: > > Cache not found for input keys: > Windows-maven-74f70cf8a20d59d956161298093365c04c4b0a04756953eb29b4484f007f9432, > Windows-maven-. > Cache not found for input keys: > oracle-Linux-maven-5a57ae570a1ce6d4a0c33eee5d829ff7cc2577cc9c15722a328d339e4c6fa082, > oracle-Linux-maven- > > Thing is, these builds that I pulled from did not change the pom files, so > the hash should not have changed... and a cache > should have been found. There are builds that have a cache hit, but seem > to be few. > > Looking at the documentation for the cache action, there is a limit of 5GB > per repository: > https://github.com/actions/cache#cache-limits > > Looking at the workflows, maybe we are using too many cache keys, and > getting close to the cache limits: > > > cat *.yml | grep key | grep runner | sort | uniq > key: ${{ runner.os }}-maven-${{ hashFiles('**/pom.xml') }} > key: geopkg-${{ runner.os }}-maven-${{ hashFiles('**/pom.xml') }} > key: integration-${{ runner.os }}-maven-${{ > hashFiles('**/pom.xml') }} > key: mssql-${{ runner.os }}-maven-${{ hashFiles('**/pom.xml') }} > key: oracle-${{ runner.os }}-maven-${{ hashFiles('**/pom.xml') }} > key: postgis-${{ runner.os }}-maven-${{ hashFiles('**/pom.xml') }} > > That's 6 unique keys, plus osx and windows being two different operating > systems, that's 8 separate caches... per branch! > Yes, because different branches lead to different pom hash, at least due > to the different version names in the pom files. > Those 5GB should thus be split into 24 different caches, leaving 200MB for > each. > > Any objection to removing all the prefixes from the database QA builds? > That should reduce the cache count significantly. > It's true that this could cause a cache to start small, due to a database > build not hitting all modules, but in a few rounds > of builds it should grow to handle everything (it's not like we change the > pom files every other day). > That sounds fine to me > > Another angle to look at this would be, do we really need per OS caches? > Again, I don't think we need OS specific > caches, do we? > I can't think of any reason to have an OS specific cache. Ian
_______________________________________________ GeoTools-Devel mailing list GeoTools-Devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/geotools-devel
