On 2024-01-09 05:56, Stephen Borrill wrote:
On 09/01/2024 09:51, Stephen Borrill wrote:
On 09/01/2024 03:41, Alex Rousskov wrote:
On 2024-01-08 08:31, Stephen Borrill wrote:
I'm trying to determine why squid 6.x (seen with 6.5) connected via
IPv4-only periodically fails to connect to the destination and then
requires a restart to fix it (reload is not sufficient).
The problem appears to be that a host that has one address each of
IPv4 and IPv6 occasionally has its IPv4 address go missing as a
destination. On closer inspection, this appears to happen when the
IPv6 address (not the IPv4) address is marked as bad.
ipcache.cc(990) have: [2001:4860:4802:32::78]:443 at 0 in 216.239.38.120 #1/2-0
Thank you for sharing more debugging info!
The above log line is self-contradictory AFAICT: It says that the cache
has both IPv6-looking and IPv4-looking address at the same cache
position (0) and, judging by the corresponding code, those two IP
addresses are equal. This is not possible (for those specific IP address
values). The subsequent Squid behavior can be explained by this
(unexplained) conflict.
I assume you are running official Squid v6.5 code.
I can suggest the following two steps for going forward:
1. Upgrade to the latest Squid v6 in hope that the problem goes away.
2. If the problem is still there, patch the latest Squid v6 to add more
debugging in hope to explain what is going on. This may take a few
iterations, and it will take me some time to produce the necessary
debugging patch.
HTH,
Alex.
Note that there have been many connections to
clientservices.googleapis.com prior to this where markAsBad was not
called, even though IPv6 connectivity was never available.
No markAsBad() is probably normal if Squid did not try to establish
an IPv6 connection or did not wait long enough to know the result of
that attempt. However, that does not explain why Squid selected an
IPv6 address as the next "good" address right after marking that IPv6
address as bad (at "restoreGoodness" line) when there was another
good IP address available. It is as if Squid stored two identical
IPv6 addresses (and not IPv4 ones), but that should not happen either.
This is tangentially related to this thread too:
https://lists.squid-cache.org/pipermail/squid-users/2023-November/026266.html
Once only the IPv6 address is being used, then it returns 503 for that
host and thus can quickly get marked as dead by a downstream squid
meaning it does not get used at all (and if it's the only peer all
access stops).
_______________________________________________
squid-users mailing list
squid-users@xxxxxxxxxxxxxxxxxxxxx
https://lists.squid-cache.org/listinfo/squid-users
_______________________________________________
squid-users mailing list
squid-users@xxxxxxxxxxxxxxxxxxxxx
https://lists.squid-cache.org/listinfo/squid-users
