Re: RFC: error codes on exit

From: brian m. carlson <sandals@crustytoothpaste.net>
Date: May 21, 2021, 01:20 UTC
Message-ID: <YKcK2elWqBQ+IVDt@camp.crustytoothpaste.net>
In-Reply-To: <YKWggLGDhTOY+lcy@google.com>

On 2021-05-19 at 23:34:24, Jonathan Nieder wrote:

Show 32 quoted lines

> Hi,
> 
> (Danger, jrn is wading into error handling again...)
> 
> At $DAYJOB we are setting up some alerting for some bot fleets and
> developer workstations, using trace2 as the data source.  Having
> trace2 has been great --- combined with gradual weekly rollouts of
> "next", it helps us to understand quickly when a change is creating a
> regression for users, which hopefully improves the quality of Git for
> everyone.
> 
> One kind of signal we haven't been able to make good use of is error
> rates.  The problem is that a die() call can be an indication of
> 
>  a. the user asked to do something that isn't sensible, and we kindly
>     rebuked the user
> 
>  b. we contacted a server, and the server was not happy with our
>     request
> 
>  c. the local Git repository is corrupt
> 
>  d. we ran out of resources (e.g., disk space)
> 
>  e. we encountered an internal error in handling the user's
>     legitimate request
> 
> and these different cases do not all motivate the same response.
> (E.g., if (c) affects just a single bot but produces a high error rate
> from that bot, we shouldn't be alarmed; if (d) is happening on a bot,
> then we should look into giving it more disk; if (e) is increasing
> significantly during a rollout then we should roll back quickly.)

In general, I'm in favor of adding some sort of error code here. Even though I don't normally use trace2, I think there's a lot of benefit to having a standardized set of error codes, and this seems like as good a place as any to introduce them.

A future iteration of this might look like us returning a negative error code from a function instead of -1 for us to signal to the caller that a particular error case occurred. We need not implement that now, of course, but I bring it up in case we want to accommodate that in our design now for future us.

I do agree with Peff that this may not necessarily provide all of the insight you want, since it can be hard to distinguish why the error occurred. For example, in Git LFS, we sometimes will pass objects we don't have to git rev-list (with --missing) and that's completely expected, whereas a missing object with git fsck would generally be cause for alarm. Provided you're comfortable with some ambiguity, I think this would be a nice improvement.

-- 
brian m. carlson (he/him or they/them)
Houston, Texas, US