Mirrors/coredns
1
0
Fork 0
mirror of https://github.com/coredns/coredns.git synced 2026-08-27 15:17:06 -04:00
Code Issues Packages Projects Releases Wiki Activity

plugin/timeouts: add maxtcpqueries option to bound queries per TCP/TLS connection (#8376)

Browse Source
This commit is contained in:
Pujitha Paladugu
2026-08-03 16:16:43 -07:00
committed by GitHub
parent 335c8b9a33
commit 4438f6d708
10 changed files with 244 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
plugin/timeouts/README.md
View File

@@ -2,7 +2,7 @@
## Name
*timeouts* - allows you to configure the supported server read, write and idle timeouts for the TCP, TLS, DoH and DoQ servers.
*timeouts* - allows you to configure the supported server read, write and idle timeouts for the TCP, TLS, DoH and DoQ servers, and the maximum number of queries served on a single TCP or TLS connection.
## Description
@@ -15,7 +15,8 @@ over HTTPS. Allowing a longer idle timeout helps performance and reduces issues
with such routers.
The *timeouts* "plugin" allows you to configure CoreDNS server read, write and
idle timeouts.
idle timeouts, and the maximum number of queries CoreDNS will serve on a single
TCP or TLS connection before closing it.
## Syntax
@@ -24,13 +25,24 @@ timeouts {
read DURATION
write DURATION
idle DURATION
maxtcpqueries MAXIMUM
}
~~~
For any timeouts that are not provided, default values are used which may vary
depending on the server type. At least one timeout must be specified otherwise
depending on the server type. At least one option must be specified otherwise
the entire timeouts block should be omitted.
* `maxtcpqueries` sets the maximum number of queries served on a single TCP or
TLS connection before CoreDNS closes it. **MAXIMUM** must be a positive
integer, or `-1` to allow an unlimited number of queries per connection
(the default). Long-lived connections that serve an unlimited number of
queries can cause uneven load distribution across CoreDNS replicas that sit
behind a connection-based load balancer, since new queries keep reusing the
same connection instead of establishing a new one. Setting a bound, e.g.
`maxtcpqueries 128`, forces clients to periodically reconnect, which allows
the load balancer to redistribute load.
The configured timeouts apply where the selected server transport supports
them. TCP, TLS and DoH servers use the read, write and idle timeouts. DoQ
servers use the read timeout to bound receiving a query on an opened QUIC
@@ -94,3 +106,16 @@ configured. The timeouts are only applied to the TCP side of the server.
forward . /etc/resolv.conf
}
~~~
Start a standard TCP/UDP server that closes a TCP connection after it has
served 128 queries, to help spread load evenly across replicas sitting behind
a connection-based load balancer.
~~~
. {
timeouts {
maxtcpqueries 128
}
forward . /etc/resolv.conf
}
~~~