Re: [Tools-discuss] How do we diagnose DOI errors?

[Carsten Bormann <[e-mail]>]

> On 2020-10-19, at 08:19, Martin Thomson <[e-mail]> wrote:
> 
> 
> On Mon, Oct 19, 2020, at 16:39, Carsten Bormann wrote:
>> On 2020-10-19, at 03:58, Martin Thomson <[e-mail]> wrote:
>>> 
>>> I've come to rely on this mechanism for citing documents, but it isn't good if it isn't reliable.  Some better information about how this operates would be great.
>> 
>> As far as I understand, the DOI bibxml is based on the doilit tool that 
>> is part of kramdown-rfc2629 (which in turn is using 
>> https://dx.doi.org/).  A 24-h cache helps keeping the load down but the 
>> data reasonably fresh.  As far as I can see, all kinds of errors are 
>> mapped to 404.
>> 
>> One tweak that can help coping with doi.org outages might be keeping 
>> the cache indefinitely.  Freshness could still be achieved by *trying* 
>> to get a refresh after 24 h, but keeping the cached value in place if 
>> that fails for some reason.
> 
> That seems like a reasonable enhancement.  These outages are far to frequent for my liking.  This is hardly the first occurrence.

Indeed, the reason kramdown-rfc doesn’t do the processing entirely on its own was to take advantage of the cache at ietf.org to improve availability.

> You can tune the timeout right down at the same time; these rarely change and anyone who uses these often (through CI, for instance) will ensure that they are kept fresh enough.  (stale-while-revalidate...)

Exactly.

> You might need to consider whether or not the cache needs an occasional flush to avoid unbounded accumulation of old entries.

I don’t think so; the number of DOIs in existence is sufficiently limited (unless we try to do negative caching, but I see no point in that).

Grüße, Carsten