1
0
mirror of https://github.com/taigrr/nats.docs synced 2025-01-18 04:03:23 -08:00

Add lame duck mode documentation.

Signed-off-by: Colin Sullivan <colin@synadia.com>
This commit is contained in:
Colin Sullivan 2021-02-06 11:15:16 -07:00
parent 21588b40d7
commit 6e33b1cc75
3 changed files with 30 additions and 0 deletions

View File

@ -102,6 +102,7 @@
* [Upgrading a Cluster](nats-server/nats_admin/upgrading_cluster.md)
* [Slow Consumers](nats-server/nats_admin/slow_consumers.md)
* [Signals](nats-server/nats_admin/signals.md)
* [Lame Duck Mode](nats-server/nats_admin/lame_duck_mode.md)
* [NATS and Docker](nats-server/nats_docker/README.md)
* [Tutorial](nats-server/nats_docker/nats-docker-tutorial.md)
* [Docker Swarm](nats-server/nats_docker/docker_swarm.md)

View File

@ -8,4 +8,5 @@ Managing a NATS server is simple, typical lifecycle operations include:
* Monitoring the server via:
* The monitoring [endpoint](../configuration/monitoring.md) and tools like [nats-top](../../nats-tools/nats_top/)
* By subscribing to [system events](../configuration/sys_accounts/)
* Gracefully shut down a server with [Lame Duck Mode](lame_duck_mode.md)

View File

@ -0,0 +1,28 @@
# Lame Duck Mode
In production we recommend that a server is shut down with lame duck mode
as a graceful way to slowly evict clients. With large deployments this
mitigates the "thundering herd" situation that will place CPU pressure on
servers as TLS enabled clients reconnect.
## Server
Lame duck mode is initiated by signaling the server:
```text
nats-server --signal ldm
```
After entering lame duck mode, the server will stop accepting new connections,
wait for a 10 second grace period, then begin to evict clients over a period of time
configurable by the [lame_duck_duration](https://docs.nats.io/nats-server/configuration#runtime-configuration)
configuration option. This period defaults to 2 minutes.
## Clients
When entering lame duck mode, the server will send a message to clients. Some
maintainer supported clients will invoke an optional callback indicating that
a server is entering lame duck mode. This is used for cases where an application
can benefit from preparing for the short outage between the time it is evicted and
automatically reconnected to another server.