Story #13048
Updated by Tom Clegg almost 7 years ago
Functionally, source:services/crunch-run is doing a reasonable job. However, the way it's implemented makes it difficult to make some of the changes we want. Relevant issues * #10181 save logs to keep periodically while a container is running (not just after it exits & saves staged outputs) * #13005 timestamps are sometimes wrong/confusing because of throttle behavior * source:services/crunch-run and source:sdk/go/crunchrunner should drop their custom manifest-writing code, now that we have generalized write support in #12483 * The implementation is more complicated / harder to follow than it should be, given the low complexity of the problem it's solving Proposed improvements * Refactor the various functional aspects (add timestamps, throttle, write to apiserver) into modular parts that communicate through simple interfaces like io.Writer. * Use io.MultiWriter from stdlib, instead of custom routing built into the processing modules. * Use @(*arvados.Collection)FileSystem()@ (*arvados.Collection)FileSystem() to open/write log files (and staged outputs? → delete @upload*.go@) upload*.go) * Drop the pretense of splitting long lines (apparently this isn't needed; MaxLogLine seems to have been disconnected 2 years ago in commit:b719ef57055ba2fd06c7a1377cc0d47ee5df935e)