[squid-users] Spliced domains tunnel connect is very slow

[Ben Goz]

By the help of God.

I'll clarify more when I'm looking at chrome's Internet Tools network
traffic of one of the users that several
Requests are timed out and at the same time I see in Squid's access log a
trace that according to your explanations
was opened for more that 60 secs for the timed out url at client side. This
issue is observed only for certain urls while the other urls are working as
expected.

I asked the question very generally but I think I should say that the
question is regarding TPROXY setup with ssl-bump
And eventually an icap service. The weird thing here is that for many other
URLs this setup works fine.
So currently I don't have any idea how to work it out.

If more information about my setup is needed please let me know.

Thanks,
Ben



‫בתאריך יום ו׳, 20 באוק׳ 2023 ב-6:27 מאת ‪Amos Jeffries‬‏ <‪
squid3 at treenet.co.nz‬‏>:‬

> On 19/10/23 01:21, Ben Goz wrote:
> > By the help of God.
> >
> > Hi,
> > I saw in my access log a traces that shows that spliced URLs tunneling
> > is very slowly:
>
> Please clarify what you mean by "slow" ?
>
>   How have you determined speed ?
>   What speed are you expecting / would you call non-slow ?
>
>
>
> FYI, Several things to be aware of:
>
>   1) CONNECT tunnel is not a simple thing with a constant "speed" of
> transfer. It represents and entire set of tunneled messages (or other
> opaque data) over indefinite timespan. Each of those messages has its
> own "speed" of transfer, with possible empty periods of 0 bytes
> transferred between.
>
>   2) the SSL-Bump procedure may pause a CONNECT tunnel during TLS
> handshake and/or validation process to asynchronously fetch missing
> certificate details, and/or validate other data with ACLs, etc.
>    Each of these subsidiary transactions may add indefinite effects on
> timing of the 'bumped' CONNECT tunnel.
>
>   3) modern networking systems utilize "Happy Eyeballs" algorithms
> wherein they may open multiple TCP connections to various (or same)
> services in parallel and only utilize the fastest to connect. This can
> result in CONNECT tunnel being initiated and unused - either closed
> immediately or left open waiting activity for long periods.
>
>
> So, as you should be able to see the log snippet shows some details
> about tunnels duration of use, you cannot tell "speed" from these logs.
>
>
> For example:
> >
> > 18/Oct/2023:15:18:50 +0300 240841 192.168.3.98 TCP_TUNNEL/200 6225
> > CONNECT beacons2.gvt2.com:443 <http://beacons2.gvt2.com:443> -
> > HIER_DIRECT/172.217.0.67 <http://172.217.0.67> - beacons2.gvt2.com
> > <http://beacons2.gvt2.com> - splice -
>
>
> Tunnel was _open_ for 240 seconds. 6225 bytes transferred.
>
> Those bytes may have been transferred in the first 1 milliseconds of the
> tunnel being open. Then Squid leaving it open waiting for further uses
> which never came.
>
> ... "slow" at 1.4 GB/sec.
>
>   ... or it could have been "slow" at 10 bytes/sec the whole time. One
> cannot tell.
>
>
> > 18/Oct/2023:15:18:50 +0300    680 192.168.3.173 TCP_TUNNEL/500 4977
> > CONNECT mobile.events.data.microsoft.com:443
> > <http://mobile.events.data.microsoft.com:443> - HIER_DIRECT/13.89.178.26
> > <http://13.89.178.26> - mobile.events.data.microsoft.com
> > <http://mobile.events.data.microsoft.com> - splice -
>
> This tunnel was never open at all. It was *rejected*.
> "speed" in that case was 3 KB/sec.
>
>
> > 18/Oct/2023:15:18:51 +0300 127307 192.168.3.97 TCP_TUNNEL/500 3101
> > CONNECT array612.prod.do.dsp.mp.microsoft.com:443
> > <http://array612.prod.do.dsp.mp.microsoft.com:443> -
> > HIER_DIRECT/20.54.24.148 <http://20.54.24.148> -
> > array612.prod.do.dsp.mp.microsoft.com
> > <http://array612.prod.do.dsp.mp.microsoft.com> - splice -
>
> Tunnel was _open_ for 127 seconds. 3101 bytes transferred.
>
> Those bytes may have been transferred in the first 1 milliseconds of the
> tunnel being open. Then Squid leaving it open waiting for further uses
> which never came.
>
>   ... "slow" at 376 MB/sec.
>
>   ... or it could have been "slow" at 25 bytes/sec the whole time. One
> cannot tell.
>
>
>
> > This is my squid configurations:
> >
> > acl NoSSLInterceptRegexp_always ssl::server_name --client-requested
> >   "/usr/local/squid/etc/splice.list"
>
> > acl alwaysBump ssl::server_name --client-requested
> > storage.googleapis.com <http://storage.googleapis.com>
> > youtubei.googleapis.com <http://youtubei.googleapis.com> www.eset.com
> > <http://www.eset.com> eset.com <http://eset.com>
> > safebrowsing.googleapis.com <http://safebrowsing.googleapis.com>
> > play.google.com <http://play.google.com>
> > on_unsupported_protocol tunnel
> > acl DiscoverSNIHost at_step SslBump1
> > ssl_bump peek DiscoverSNIHost
> > ssl_bump bump alwaysBump   -  Used to bumd certain subdomains before the
> > whole domain is bumped.
> > ssl_bump splice NoSSLInterceptRegexp_always
> > ssl_bump stare all
> >
> >
> >
> > Other CONNECT requests are served noramly.
> > Is this issue could be a root cause for the generally slow internet?
> >
> > Thanks,
> > Ben
> >
> > _______________________________________________
> > squid-users mailing list
> > squid-users at lists.squid-cache.org
> > https://lists.squid-cache.org/listinfo/squid-users
> _______________________________________________
> squid-users mailing list
> squid-users at lists.squid-cache.org
> https://lists.squid-cache.org/listinfo/squid-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squid-cache.org/pipermail/squid-users/attachments/20231021/c4e57e33/attachment.htm>