Replies: 7 comments 5 replies
-
Seems broken? Although, a prefetch |
Beta Was this translation helpful? Give feedback.
-
Agree it seems broken. "1" is a fine value for AMOUNT, it should basically
force it to prefetch almost all the time, I think. IIRC, the code basically
does this:
- For each incoming query, if the last time the query was seen is less than
DURATION ago, bump a counter.
- As we get close to the TTL if the counter is greater than AMOUNT,
prefetch.
So, AMOUNT of 1 basically saying if we saw 2 queries inside of DURATION, we
will prefetch it.
…On Fri, May 19, 2023 at 7:45 AM chrisohaver ***@***.***> wrote:
Seems broken? Although, a prefetch AMOUNT of 1 is kind of meaningless per
the README description, since there is nothing "between" a single query. I
don't know how the code handles a value of 1. Maybe it does something
awkward in this case?
—
Reply to this email directly, view it on GitHub
<#6103 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ACIHRM5EVL5WKMY6PZ25ZXLXG6BRDANCNFSM6AAAAAAYHZOTHQ>
.
You are receiving this because you are subscribed to this thread.Message
ID: ***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
We should check the code and not take my faulty memory as fact…
…On Fri, May 19, 2023 at 8:44 AM chrisohaver ***@***.***> wrote:
... the robot in me want to just add a +1 to AMOUNT ... "Popular means
AMOUNT+1 queries have been seen with no gaps of DURATION or more between
them."
—
Reply to this email directly, view it on GitHub
<#6103 (reply in thread)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ACIHRMZXTQIJCX4HSYU627LXG6IMDANCNFSM6AAAAAAYHZOTHQ>
.
You are receiving this because you commented.Message ID:
***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
Hello everyone, I played around with the prefetch values before, I spent the whole day, and I did it again now for ~ 1h. If you guys need more tests or debug, please let me know. I'm glad to help or give more information. OH, more info:
thx again! |
Beta Was this translation helpful? Give feedback.
-
Ok. I didn’t take offense. Not sure what I said to imply I was offended. (FWIW, Im also not offended by you thinking I was offended :) ) |
Beta Was this translation helpful? Give feedback.
-
Any news? It's still broken. |
Beta Was this translation helpful? Give feedback.
-
Actually looks like a bug was already opened and then closed after a maintainer confirmed that prefetch is working as expected. #6111 |
Beta Was this translation helpful? Give feedback.
-
Hello,
First, thx for your work on coreDNS!
Here's my question. I'm trying to avoid long queries towards a IBM service that sits on another country.
I was trying to understand prefetch, but I feel dumb reading it's documentation. I'm trying to avoid slow queries, and, by reading the documentation I believe cache { prefetch } is the way to go.. I'm open for other suggestions. ^^
I'm using minikube on windows 10. CoreDNS version is 1.9.3, I tested as well 1.10.1
16GB mem + 4 cores and 8 processors.
My config is the following:
I know the config is being hit because if I set "cache 30", the TTL changes to 30 when I use dig, as well the log format.
Here's the command and the logs:
I'm running a dig every 5s, so the prefetch would be triggered. right?
I gave more memory and cpu to coreDNS, but I have the same behavior.
I tried as well a specific config for the host, trying different approachs for the prefetch, like:
The CPU usage is below 20% and memory usabe below 50%.(yeah.. docker on windows XD)
I was expecting to have always a good query time using the local cache while prefetching when needed.
Am I'm not understanding how this feature works?
Do you guys have any other suggestion?
oh, the whole dig response:
May be important[redacted]:
thx!!
Beta Was this translation helpful? Give feedback.
All reactions