Good catch! This didn't fix the issue by itself, but it did remind me that proxy.config.http.cache.when_to_revalidate was set to 2 (Always Stale).
I guess on our non-chained proxies everything worked because of our rules in cache.config? I appreciate the advice, Jeremy. Thank you! On Mon, Jan 8, 2018 at 8:29 AM, Jeremy Payne <[email protected]> wrote: > Can you modify the below to reflect non-zero values ? > > ############################################################################## > # Heuristic cache expiration. Docs: > # https://docs.trafficserver.apache.org/records.config#heuristic-expiration > ############################################################################## > # Setting all of these to zero in order for everything to be considered stale > # by ATS. If we want to change this in the future, this is where to do it. > CONFIG proxy.config.http.cache.heuristic_min_lifetime INT 0 > CONFIG proxy.config.http.cache.heuristic_max_lifetime INT 0 > CONFIG proxy.config.http.cache.heuristic_lm_factor FLOAT 0.000000 > ############################################################################## > > On Mon, Jan 8, 2018 at 10:24 AM, Patrick O'Brien > <[email protected]> wrote: >> Hey Jeremy, >> >> Here's the full config from the child node: >> https://gist.github.com/poblahblahblah/a1abdad3d64ebdfba003c75e50c9fe00 >> >> Thanks for looking! >> >> On Sun, Jan 7, 2018 at 8:13 PM, Jeremy Payne <[email protected]> wrote: >>> Can you post the complete records.config from your child cache ? >>> If I had to guess, it looks like heurstics or fuzzytimer may be set too low >>> ??? >>> But if you provide a company sanitized version of your records.config >>> that should give us a better understanding of your setup. >>> >>> >>> >>> >>> >>> >>> On Wed, Jan 3, 2018 at 6:10 PM, Patrick O'Brien >>> <[email protected]> wrote: >>>> Hello, >>>> >>>> I am testing out hierarchical caching in ATS 7.1.1. Everything seems >>>> to work, although it looks like the child node is always needing to >>>> revalidate objects fetched from the parent node. Here are the headers >>>> on an example GET: >>>> https://gist.github.com/poblahblahblah/928aaa88c24ef3c05366433a514f1a28 >>>> >>>> The parent has the object stored in RAM and doesn't require >>>> validation, but the child thinks the response is stale. Here is the >>>> cache.config file on both the parent and the child: >>>> https://gist.github.com/poblahblahblah/36b54ac8b7d75071c1342da6694d3c51 >>>> >>>> The problem with this is that the connection between the child and >>>> parent is pretty flaky and when the parent is down we are met with a >>>> 502: >>>> https://gist.github.com/poblahblahblah/d50657354c815040faae50539fa45836 >>>> >>>> Here is the parent.config on the child: >>>> https://gist.github.com/poblahblahblah/9f1dcb4cf674c21c80a27173bd0ef81b >>>> >>>> Relevant configs from records.config for the parent: >>>> https://gist.github.com/poblahblahblah/e6ac573e9f9cd63088c2e06e6544d801 >>>> Relevant configs from records.config for the child: >>>> https://gist.github.com/poblahblahblah/5a6dc02a88b24cafca0e8c70d353b92e >>>> >>>> Any guidance on this would be much appreciated. >>>> >>>> -patrick
