----- Original Message -----
> On 06/23/2016 01:56 AM, Nathan Scott wrote:
> > ----- Original Message -----
> >>> 651 ditto
> >> The only clue I have here so far is that 651 only fails for me with a non-
> >> secure-sockets enabled build. Still investigating though.
> >>
> > Somehow, this is fallout from commit 1b74b0f5e2bc - pmproxy connections
> > [...]
> I'm not getting this failure with pcp built either way (--with or
> --without-secure sockets).
>
> The "Bad version sting" messages are expected. If n connections are
> attempted then n-1 of them will be abandoned and there should be n-1
> such messages. It was one of the concerns that was raised when this
> implementation was suggested.
>
Ah yes, of course. We should handle this like pmcd (i.e. quietly, with
a debug diagnostic - will do).
> AFAIK, pmproxy uses the same code as pmcd to connect. Also, all of the
Sort of - there's certainly different code paths travelled in libpcp on
connect to pmproxy, although alot is also shared yes.
I'll keep digging into it. For me, it's 100% reproducible without secure
sockets - very odd! I imagine it is also going to be dependent on how
local interfaces are configured, so that's a potential area of difference
that could be affecting reproducibility there.
Ken, is it always non-secure-sockets builds that fail with your QA hosts?
(pmconfig -L secure_sockets)
cheers.
--
Nathan
|