Open infotroph opened 7 years ago
@mdietze I think you discussed this with me in Gitter shortly after I opened it, but now I can't grep up the conversation.
My memory is that we decided this is at least partly expected -- download.Geostreams
is a remote call here and in general we don't always want a remote logger.severe
to kill the workflow -- but I forget whether you thought this example was a bug. Close or fix?
I think there's two separate bugs here to be fixed:
1) Both remote executions throw the same error -- that the requested data is before the start date. It seems like we should be catching that before calling the remote function (unless, in this case, the Input start date is coming from the remote source not from the database).
2) It's possible that some combination of convert.input or remote.execute.R is failing to detect that the remote workflow is throwing an error. Whether we decide that a given workflow should stop on severe or not, we need to at least make sure that remote error is detected and passed back to the part of the workflow that handles the "stop.on.error" flag. That flag should also be handled consistently -- if stop.on.error is FALSE then the workflow also needs to be tolerant to the 3rd error (the inability to open the file, which is what finally kills the workflow). Furthermore, it would be good to note that in the settings so that failure can be leveraged downstream -- there's no point in calling start.model.runs for a site or ensemble member whose input processing failed.
I think the function download, should fail, and produce a zero byte file, which should trigger the main code to throw another sever to fail the call.
Thing I don't understand is that the call should not be remote, but should just call the function since the requested host, and current host both are the same.
This issue is stale because it has been open 365 days with no activity.
I believe there are some other issues related to this.
My hunch is that the symptom in the title was fixed by #2489. Whether severe ought to have been thrown in the first place is probably still worth considering as part of a larger met robustness effort.
This issue is stale because it has been open 365 days with no activity.
I'm not sure if this is actually a bug or working as intended. In the run below, I'm trying to download raw data to extend an existing input. The initial failure is pure user error -- I'm requesting data that doesn't exist -- but I was expecting the whole conversion workflow to exit after the
logger.severe
in the download stage. Instead, met2CF goes ahead and the workflow only stops at the secondlogger.severe
, when met2CF discovers that it's attempting to convert a file that wasn't downloaded.workflow output: