unexpected high amount of Cold lookups

[ezko]

Hi,
i'm getting a pretty high number of
proxy.process.http.transaction_counts.miss_cold which I don't expect.
All my cached objects are set with an explicit Cache-Control max-age , so
when they expire , if a new copy is retrieved from the origin , it has to be
new.
How can i debug this issue ?
thanks in advance.
EZ



--
Sent from: http://apache-traffic-server.24303.n7.nabble.com/

[Miles Libbey]

Post by ezko
Hi,
i'm getting a pretty high number of
proxy.process.http.transaction_counts.miss_cold which I don't expect.
All my cached objects are set with an explicit Cache-Control max-age , so
when they expire , if a new copy is retrieved from the origin , it has to be
new.
How can i debug this issue ?

Some questions I would be asking myself:
- Is my cache big enough to support my working set size?
proxy.process.cache.wrap_count tells you how often the cache has
wrapped since the last restart.
- When I look at the requested urls and their cache status, Do I get
any hints as to why there are misses? (For instance, perhaps there are
random query strings in the requests making ATS think they are novel
objects).
- Do my objects Vary on reasonable headers (eg, *not* user-agent,
cookie, ...)? Perhaps there are too many variants?
- Are there POSTs or other http verbs that would invalidate the cached object?

miles

[ezko]

thanks for the suggestion miles.

i checked the cache as you suggested. I think its OK. There are no
unexpected query strings , or other noticeable object variations.
proxy.process.cache.wrap_count seems low , and I don't see any non-GET
verbs.

I suspect there is something wrong with the way Cache-control max-age header
is set.
If for example I set Cache-control max-age to 4 seconds , when ATS caches
the object , how does it decide when the object is expired ? I assumed it
will simply expire the object after 4 seconds. Is that true ? , or are there
any other affecting factors ?

looking into the diag.log - can I spot ATS making a COLD decision ?

thanks in advance.





--
Sent from: http://apache-traffic-server.24303.n7.nabble.com/

[Miles Libbey]

Post by ezko
thanks for the suggestion miles.
i checked the cache as you suggested. I think its OK. There are no
unexpected query strings , or other noticeable object variations.
proxy.process.cache.wrap_count seems low , and I don't see any non-GET
verbs.
I suspect there is something wrong with the way Cache-control max-age header
is set.
If for example I set Cache-control max-age to 4 seconds , when ATS caches
the object , how does it decide when the object is expired ? I assumed it
will simply expire the object after 4 seconds. Is that true ? , or are there
any other affecting factors ?

The main other factor I can think of is around Vary:. ATS will store
(up to 5 by default) variants. So, when the request comes in, ATS
picks the variant that best matches the request. If the object varies
on user-agent (of which there are zillions), there is likely not going
to be a match.

Post by ezko
looking into the diag.log - can I spot ATS making a COLD decision ?

I'm not sure. We use cacheStatus=%<chm> in our logs
https://docs.trafficserver.apache.org/en/7.1.x/admin-guide/logging/formatting.en.html

Post by ezko
thanks in advance.
--
Sent from: http://apache-traffic-server.24303.n7.nabble.com/

[ezko]

thanks miles,
i actually found in the cache 2 variants (different Accept-Encoding:)

On Tue, Dec 12, 2017 at 2:46 AM, ezko &lt;

Post by ezko
thanks for the suggestion miles.
i checked the cache as you suggested. I think its OK. There are no
unexpected query strings , or other noticeable object variations.
proxy.process.cache.wrap_count seems low , and I don't see any non-GET
verbs.
I suspect there is something wrong with the way Cache-control max-age header
is set.
If for example I set Cache-control max-age to 4 seconds , when ATS caches
the object , how does it decide when the object is expired ? I assumed it
will simply expire the object after 4 seconds. Is that true ? , or are there
any other affecting factors ?

The main other factor I can think of is around Vary:. ATS will store
(up to 5 by default) variants. So, when the request comes in, ATS
picks the variant that best matches the request. If the object varies
on user-agent (of which there are zillions), there is likely not going
to be a match.

Post by ezko
looking into the diag.log - can I spot ATS making a COLD decision ?

I'm not sure. We use cacheStatus=%
<chm>
in our logs
https://docs.trafficserver.apache.org/en/7.1.x/admin-guide/logging/formatting.en.html

Post by ezko
thanks in advance.
--
Sent from: http://apache-traffic-server.24303.n7.nabble.com/

--
Sent from: http://apache-traffic-server.24303.n7.nabble.com/