raboof/pekko
1
0
Fork 0
Code Issues Pull requests Projects Releases 6 Packages Wiki Activity Actions 10

=clu #18577 Option to stop cluster client after no receptionist contact timeout

Browse source
This commit is contained in:
Johan Andrén 2016-01-07 19:54:55 +01:00
parent 673174a64a
commit 19c7017482
6 changed files with 227 additions and 26 deletions
Download patch file Download diff file Expand all files Collapse all files

22
akka-docs/rst/java/cluster-client.rst
View file

@ -146,3 +146,25 @@ a parameter to the ``ClusterClient.props`` factory method, i.e. each client can
with different settings if needed.
.. includecode:: ../../../akka-cluster-tools/src/main/resources/reference.conf#cluster-client-config
Failure handling
----------------
When the cluster client is started it must be provided with a list of initial contacts which are cluster
nodes where receptionists are running. It will then repeatedly (with an interval configurable
by ``establishing-get-contacts-interval``) try to contact those until it gets in contact with one of them.
While running, the list of contacts are continuously updated with data from the receptionists (again, with an
interval configurable with ``refresh-contacts-interval``), so that if there are more receptionists in the cluster
than the initial contacts provided to the client the client will learn about them.
While the client is running it will detect failures in its connection to the receptionist by heartbeats
if more than a configurable amount of heartbeats are missed the client will try to reconnect to its known
set of contacts to find a receptionist it can access.
When the cluster cannot be reached at all
-----------------------------------------
It is possible to make the cluster client stop entirely if it cannot find a receptionist it can talk to
within a configurable interval. This is configured with the ``reconnect-timeout``, which defaults to ``off``.
This can be useful when initial contacts are provided from some kind of service registry, cluster node addresses
are entirely dynamic and the entire cluster might shut down or crash, be restarted on new addresses. Since the
client will be stopped in that case a monitoring actor can watch it and upon ``Terminate`` a new set of initial
contacts can be fetched and a new cluster client started.