Add a note re: GH-1667 and redistributing client RPC requests
This commit is contained in:
parent
4a56b5e50e
commit
c08b3f3484
10
CHANGELOG.md
10
CHANGELOG.md
|
@ -2,6 +2,16 @@
|
|||
|
||||
IMPROVEMENTS:
|
||||
|
||||
* Consul agents will now periodically reconnect to available Consul servers
|
||||
in order to redistribute their RPC query load. Consul clients will, by
|
||||
default, attempt to establish a new connection every 120s to 180s, however
|
||||
the rate at which agents begin to query new servers is proportional to the
|
||||
size of the Consul cluster (servers should never receive more than 64 new
|
||||
connections per second per Consul server as a result of rebalancing).
|
||||
Clusters in stable environments who use `allow_stale` should see a more
|
||||
even distribution of query load across all of their Consul
|
||||
servers. [GH-1667]
|
||||
|
||||
BUG FIXES:
|
||||
|
||||
* Updated the internal web ui (`-ui` option) to latest released build, fixing
|
||||
|
|
Loading…
Reference in New Issue