Asad Sajjad Ahmed
Asad Sajjad Ahmed
> As commenting on unchanged lies is not possible: In addition to the issues pointed out, if this _was_ the right approach, an `n_conn` decrement is missing for the `bo->htc...
> As of [edfeb73](https://github.com/varnishcache/varnish-cache/commit/edfeb73ebbfbea7db487105a0e7ae12f6706ced8), b68 fails for me locally. Interesting. > > looks like this was because the "blackhole" connection does not time out as intended, but rather fails immediately:...
> I do not think that we will be able to find some magic IP which times out. Maybe we need a facility in varnishtest which listens on a socket,...
> > In order to make the VTCP_Connect to fail, the backend needs to either not be in the LISTEN state or we need to initiate the handshake with a...
> also with uds? This is not how the connect syscall works (at least not on Linux). The connect syscall associate the socket to an address, and gives you back...
> Besides specific comments, I am not sure if adding global counters makes much sense. If users see these increase, won't they immediately ask "yeah, but which backend?". So unless...
> > The global counter is a quick way to tell if looking into backends make sense. > > I am not entirely convinced that the double accounting pays off,...
> So I _think_ this is OK, but I would like to ask @asadsa92 to make life easier for reviewers. Could you please Thanks for your 2nd review, we are...
@nigoroll Thanks once again for your review, this should be ready now. To view the full diff compared to the previous version: ```shell $ git range-diff 8a219c644..a3fdba2c9 720b939c2.. ```