[
https://issues.apache.org/jira/browse/NIFI-2615?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15754618#comment-15754618
]
Andrew Psaltis commented on NIFI-2615:
--------------------------------------
Comments from users mailing list:
http://apache-nifi-users-list.2361937.n4.nabble.com/Re-Processors-on-the-fly-for-many-sensor-devices-tp47p466.html
When working with Nio in blocking mode (as we're currently doing) we will
need to ensure that the processor is cable of
- reconnecting properly and handle read timeouts / server disconnects / ....
properly
- cleaned up all tcp connections that it manages
This will require additional plumbing IMHO (I don't know to what extend NiFi
supports base classes for this), but I found that during testing that
- Sometimes TCP socket communication hangs (on read operations) when
connection is flaky
- TCP sockets are not always properly closed, even after processor shutdown
(It requires a NiFi restart to close them).
> Add support for GetTCP processor
> --------------------------------
>
> Key: NIFI-2615
> URL: https://issues.apache.org/jira/browse/NIFI-2615
> Project: Apache NiFi
> Issue Type: New Feature
> Components: Core Framework
> Affects Versions: 1.0.0, 0.7.0, 0.6.1
> Reporter: Andrew Psaltis
> Assignee: Andrew Psaltis
>
> This processor will allow NiFi to connect to a host via TCP, thus acting as
> the client and consume data. This should provide the following properties:
> * Endpoint - this should accept a list of addresses in the format of
> <Address>:<Port> - if a user wants to be able to track the ingestion rate per
> address then you would want to have one address in this list. However, there
> are times when multiple endpoints represent a logical entity and the
> aggregate ingestion rate is representative of it.
> * Failover Endpoint - An endpoint to fall over to if the list of Endpoints is
> exhausted and a connection cannot be made to them or it is disconnected and
> cannot reconnect.
> * Receive Buffer Size -- The size of the TCP receive buffer to use. This does
> not related to the size of content in the resulting flow file.
> * Keep Alive -- This enables TCP keep Alive
> * Connection Timeout -- How long to wait when trying to establish a connection
> * Batch Size - The number of messages to put into a Flow File. This will be
> the max number of messages, as there may be cases where the number of
> messages received over the wire waiting to be emitted as FF content may be
> less then the desired batch.
> This processor should also support the following:
> 1. If a connection to end endpoint is broken, it should be logged and
> reconnections to it should be made. Potentially an exponential backoff
> strategy will be used. The strategy if there is more than one should be
> documented and potentially exposed as an Attribute.
> 2. When there are multiple instances of this processor in a flow and NiFi is
> setup in a cluster, this processor needs to ensure that received messages are
> not dual processed. For example if this processor is configured to point to
> the endpoint (172.31.32.212:10000) and the data flow is running on more than
> one node then only one node should be processing data. In essence they should
> form a group and have similar semantics as a Kafka consumer group does.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
