# Packager Workflow Healthcare Demo <!-- Put the link to this slide here so people can follow --> slide: https://hackmd.io/@MxhO_4YgQMSPxLlfV4RZKw/H1hO8BAmP#/ --- ## Who are we? - Stephen Coady - Adam Saleh - Will Woods --- ## Thesis The Packager Workflow for RPMS as is, is fit for purpose, but more improvements will require significant investments to gathering metrics about our interconnected systems. --- # State of Packager Workflow - State of gating - Functional view - Performance view - Problems in general --- ## Gating - only *68* packages in src.fedoraproject.org - only 53 distinct package names end up with `greenwave.decision.update` - only 7% NVRs --- ## View from CI messages * 294708 tests completed succesfully * 16506 with runtime error (probably test-case failing) * 23766 with infrastructure error 8% error-rate --- ## View from monitor-gating 728 completed runs with monitor-gating single-build * 281 failed test runs total * 2 failures only were during ‘Building the package’ * 220 either had problems with seeing result in results db, or saw the correct result, which means the testing did complete successfuly. 8% error-rate, if we don't consider gating 38% error-rate in general, but large number of these might be problems with the test --- ## Performance view * we saw spikes between the time a NVR has been queued and the time it has been running * hard to get exact numbers * we don't have run-ids * single NVR can go through test multiple times --- ## Problems in general The most glaring issue we uncovered is that geting answers to fairly simple questions is hard. * How long does it take to build a given package, on average? * Is the build/update system running slower than normal right now? * How many koji builders are available for each arch? * How many builds / tests / etc are waiting for systems to be available? (And how long have they been waiting?) * How many packages have been succesfully built? --- ## Problems in general * How long does it take to get an update * built * tested * pushed?