On Wed, Feb 14, 2018 at 10:56:08AM -0800, William Tu wrote:
> Usually ofproto/trace is used to debug the flow translation error.
> When translation error such as recursion too deep or too many resubmit,
> the issue might happen momentary; flows causing the recursion expire
> when users try to debug it.  This patch enables the ofproto trace
> automatically when recursion is too deep or too many resubmit, by
> invoking the translation again, and log the ofproto trace as warnings.
> Since the log will be huge, rate limit to one per minute.
> VMWare-BZ: #2054659
> Signed-off-by: William Tu <u9012...@gmail.com>
> Tested-by: Greg Rose <gvrose8...@gmail.com>
> Reviewed-by: Greg Rose <gvrose8...@gmail.com>

Thanks for working on this!

Some of the data passed to ofproto_trace() is also passed to the
xlate_actions() call, indirectly.  Did you check whether that data is
possibly modified by xlate_actions()?  If it is, then we might have to
reconsider this approach, because flow data, etc. is very large and I
don't think that we can afford to always make a copy of it in advance on
the chance that the original might be needed for tracing later.

I think that VLOG_WARN is a very high log level for this data.  I would
tend to use DBG.


dev mailing list

Reply via email to