[pbs-devel] [PATCH proxmox-backup v2 1/2] rest-server/worker-task: replace newlines with '\n' in task result

[Thomas Lamprecht]

On 11.10.21 15:57, Dominik Csapak wrote:
> On 10/11/21 15:37, Thomas Lamprecht wrote:
>>>
>>
>> I do not mean the root cause that logged a newline at the end, we cannot avoid the future,
>> but the one that crates the TaskError enum, and if I'm not completely sleep deprived, that
>> is create_state.
> 
> sorry, i misunderstood you, yes, the TaskState::Error gets created in 'create_state', and yes, it makes sense to do it already there...
> 
>>
>>> but wanted to have a generic method to not have a wrong task state even
>>> when it happens again somewhere
>>>
>>> imho, percent encoding(%0A) is not nicer in this context here than \n
>>
>> percent encoding is an actual format than can be rendered again nicely, just randomly
>> mapping the \n control code (0x0A) to 0x5C 0x6E, without escaping existing (literal) \n
>> just isn't a used format, at least FWICR, that's my point in provoking that question.
>>
> 
> while we could decode it again, existing tasks did not get encoded, so we would potentially decode some strings that shouldn't be decoded...



> 
> we could also simply remove the newlines at all, i don't think we would

no, same loss of information as replacing it with a literal \n...

> benefit from having multiline error messages in tasks..
> since most of the time we want to show a single line per task

We want to have the error, if it's longer and over a few lines then
that's fine too.

> 
> or what about trying to parse the task log backwards
> until we find a correct date+taskerror or at least a correct date?
> (in the first case we can assume that the newlines are part of
> the error, in the second the state would still be 'unknown' but
> with a correct date)

brittle to do in general, one cannot guarantee that the error doesn't also contains
what we'd match, may not be likely but if we want to improve on the current solution
I'd rather go for an actual improvement over an hack..

> 
> if you don't like either approach, i'd send a
> version with percent-encoded/decoded, it's imho more important
> that we get the correct task state on task parsing than how we
> solve it exactly

I'd like to have an actual solution without loss of information, and with our current
approach I see only a two-way encoding doing that, I don't care much for what one we
use, but good would be as human readable as possible as this is a issue that happens
rather rarely.

IMO having log + result intertwined into the log isn't ideal either, having a "$UPID.result"
(or the like) file with the information already present in a structured format would
simplify things a lot..