[Spice-devel] [PATCH spice-common 0/4] RFC: add structured logging and log category

[Christophe de Dinechin]

> On 13 Jun 2017, at 10:42, Marc-André Lureau <marcandre.lureau at redhat.com> wrote:
> 
> Hi
> 
> ----- Original Message -----
>> On Tue, Jun 13, 2017 at 09:45:20AM +0200, Christophe de Dinechin wrote:
>>>> 5. The spice_log_init deals with hard-coded variables in a
>>>> hard-coded way. I purposely had a function taking a string, and
>>>> called either with env vars or command line options. I was thinking
>>>> long-term about other inputs, notably the client being able to send
>>>> debug options to the server.
>>>> 
>>>> 
>>>> Whoo, client sending debug options to the server? That is quite strange.
>>> 
>>> What I would like is to be able to enable collection of debug
>>> information (ideally on a per-session basis) without having to restart
>>> the server. Is that strange?
> 
> "enabling a collection of debug information" at run-time could be done through a server API, exposed as a QMP command (or other interface for other servers).

This is another interface that we could expose with the approach I suggest, yes.

> I don't think the client & SPICE protocol should have this admin/debug feature.

Just because it’s a good idea to add it also as a QMP command does not mean it’s a bad idea to make it convenient to get server debug for a given session. It’s really different use cases.

I would use QMP if I’m really debugging the server.

I would use a per-session debug if I’m interested in something related to the interaction between server and client. In that case, I would ideally use something like mm time in the logs instead of standard time, to make sure I can correlate server and client logs.


> 
>> Let's first see if we manage to add useful debug logs to both the client
>> and the server, then we can look into adding this :)

I agree. I was explaining why I had made a choice to have a dedicated function taking a string argument with the options, rather than directly reading the environment variable.

>> In the mean time, this will probably be possible to do through a gdb script.

One that I can run client side?

BTW, having a function also means I could do something like:

- attach gdb to the server
- Run ‘p spice_set_trace(“ssl:marshal_read”)’

and easily activate specific traces on an already-running server.


Christophe