Shawn,
How about if we consider the questions that we want to be able to answer 
using this data?

- How many people have installed package such and such?
- How many people have updated package such and such to a new version?
-- How many of those were a specific request to update that package?
-- How many were part of an overall image-update?
- How many people have looked at information about a package?

To answer these questions, it really isn't necessary to send the command 
that triggered the operation, a transaction id.  Also, if a client is 
installing hundreds of packages, then there will be hundreds of requests 
for manifests (with a little bit of information on each one). 

I'd like to suggest that we return to Stephen's original proposal of 
just sending the intent information:

install/null
install/[previous_version]
info
contents
image-update/[previous_version]

In the header of the manifest request. This allows us to answer the questions 
above.

Tom




Shawn Walker wrote:
> Greetings,
>
> While looking into how to implement bug 2022 (client should provide 
> operational intent to server), one question that came up was just how 
> much information we could get away with sending in the header.
>
> Specifically, I've read multiple warnings posted on various pages that 
> say that the size of the HTTP header should be as small as possible for 
> various reasons.
>
> Right now, I'm thinking of sending the history start_state to the server 
> as that will tell us what the client is installing for the first time or 
> what they're upgrading from and upgrading to.  I could also send along 
> the command that triggered it, the name of the operation (as perceived 
> by the client) and eventually a transaction id (that could be used to 
> help link error log entries to client tracebacks).
>
> For the cli, I'm not expecting the user to request more than one or two 
> packages (explicitly); however, for the gui, I could easily see hundreds 
> or thousands of packages being explicitly requested by the user at a 
> single time.
>
> With this in mind, this seems like quite a bit of information to pass 
> along in the header.  As such, I've been considering a separate POST 
> operation before the client fetches a manifest (regardless of cache 
> status), begins a filelist operation, etc. that provides this 
> information instead of trying to squeeze it into the headers.
>
> Thoughts?
>
>   

_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to