Hi,
Often, problems with establishing a communication connection can be
solved with the following configuration:
1)You may have multiple network interfaces and the wrong one could be
used. Solved by changing the SPI communication timeouts.:
<property name="communicationSpi">
<bean
class="org.apache.ignite.spi.communication.tcp.TcpCommunicationSpi">
...
<property name="connectTimeout" value="5000"/>
<property name="maxConnectTimeout" value="10000"/>
...
</bean>
</property>
Otherwise, you can wait more than 10 minutes when trying to create a
connection (due to the ExponentialBackoffTimeoutStrategy strategy).
2)Some operations in the cluster require communication with clients
through communication. In case you have communication problems, but you
can still access through the discovery SPI, such operations may hang. To
avoid it please set the following property:
https://ignite.apache.org/releases/latest/javadoc/org/apache/ignite/IgniteSystemProperties.html#IGNITE_ENABLE_FORCIBLE_NODE_KILL
If these recommendations do not help, then yes, as Ilya said, we require
a loudspeaker on your part.
BR,
Andrei
10/30/2020 2:20 PM, Ilya Kasnacheev пишет:
Hello!
Do you have a reproducer for this behaviour that I could run and see
it failing?
Regards,
--
Ilya Kasnacheev
вт, 27 окт. 2020 г. в 22:02, VeenaMithare <v.mith...@cmcmarkets.com
<mailto:v.mith...@cmcmarkets.com>>:
Hi Ilya, The node communication issue is because one of the node
is being restarted - and not due to network failure . The original
issue is as below : Our setup : Servers - 3 node cluster Reader
clients : wait for an update on an entry of a cache ( around 20 of
them ) Writer Client : 1 If one of the reader client restarts
while the writer is writing into the entry of the cache , the
server attempts to send the update to the failed client's local
listener . It keeps attempting to communicate with the failed
client ( client's continous query local listener ? ) till it
timesout as per
connTimeoutStrategy=ExponentialBackoffTimeoutStrategy . ( Please
find the snippet of the exception below. The complete log is
attached as an attachment ) This delays the completion of the
transaction that was started by the writer client. Is there any
way the writer client could complete the transaction without
getting impacted by the reader client restarts ? regards, Veena.
------------------------------------------------------------------------
Sent from the Apache Ignite Users mailing list archive
<http://apache-ignite-users.70518.x6.nabble.com/> at Nabble.com.