[squid-dev] Proposal: Helper response: concatenated values with custom delimiter

[Alex Rousskov]

On 2026-04-28 04:37, Andrey K wrote:

> On my part, I see two reasons for implementing support for the 
> key=v1,v2,v3 format:
> 
> 1. As Amos pointed out in the previous email, using this format will 
> reduce the volume of traffic sent from the helper, thereby reducing the 
> processing overhead.

Yes, this optimization has been recognized as such since the initial 
code changes were posted on GitHub. To avoid a misunderstanding, I am 
not disputing this optimization existence. I was answering a different 
question: Do we need this optimization? My current (short) answer is 
"no". My earlier response has the details.


> However, the impact will be marginal; for 300 groups, the savings would 
> be roughlylen("group=") * 300 = 1800 bytes.
> It would be far more efficient to filter groups during PAC record 
> parsing. This filtering should be performed as early as possible to 
> prevent further processing of unused groups. I intend to spin this 
> suggestion off into a separate discussion thread.

Yes, and we have already discussed related filtering ideas on GitHub. If 
helper response size reduction and Squid work/memory reduction are the 
goals, then early filtering is a better answer when dealing with 
hundreds of groups that are of no interest to Squid.


> 2. Annotation of client connections. Currently, only one tag is 
> supported for client connection annotation: clt_conn_tag, and the 
> current handler for this tag does not support multiple occurrences of 
> the same tag in the helper response.

I disagree that connection annotation is (or should be viewed as) 
related to supporting key=list format. It is a separate/orthogonal issue.


> Therefore, I suggest we close this proposal for now

I support this suggestion.


Thank you,

Alex.

> вс, 26 апр. 2026 г. в 06:26, Amos Jeffries <squid3 at treenet.co.nz 
> <mailto:squid3 at treenet.co.nz>>:
> 
>     On 24/04/2026 01:42, Alex Rousskov wrote:
>      > On 2026-04-22 07:24, Andrey K wrote:
>      >> Hello, Amos and Alex,
>      >>
>      >> Thank you for the discussion and the proposed formats.
>      >>
>      >> Since we already support the
>      >>      key=v1 key=v2
>      >> format, and the
>      >>      key=v1,v2
>      >> format is currently undocumented, I think we should answer two
>      >> questions first:
>      >>
>      >> 1. Do we really need to support list-values? It seems Alex
>     thinks we
>      >> might not need them at all.
>      >
>      > If you have to ask, then the answer is "no" or "not yet": We need
>     a very
>      > compelling use case to add a "list-values" optimization (when we
>     already
>      > support lists of values). Since you are not sure, and you were
>     the only
>      > one making/wanting related changes (AFAICT), we evidently lack
>     such a
>      > case today.
>      >
>      > Rule of thumb: Avoid complex optimizations until their expense
>     can be
>      > justified by a use case (that can also guide their implementation).
>      >
> 
>     Remember that the earlier work by Andrey did bring up a reasonable
>     rational for doing it anyway.
> 
>     Namely that supporting A,B,C lists improves the bandwidth cost of
>     transmission between helper and Squid.
> 
>     Several of the Kerberos use-cases are sending many group names out of
>     the helper.  So this specific (accept comma-delimited value list) is
>     not
>     a theoretical optimization.
> 
> 
>     Cheers
>     Amos
> 
>     _______________________________________________
>     squid-dev mailing list
>     squid-dev at lists.squid-cache.org <mailto:squid-dev at lists.squid-cache.org>
>     https://lists.squid-cache.org/listinfo/squid-dev
>     <https://lists.squid-cache.org/listinfo/squid-dev>
> 
> 
> _______________________________________________
> squid-dev mailing list
> squid-dev at lists.squid-cache.org
> https://lists.squid-cache.org/listinfo/squid-dev