[
https://issues.apache.org/jira/browse/IGNITE-419?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14355232#comment-14355232
]
Ivan Veselovsky edited comment on IGNITE-419 at 3/11/15 9:28 AM:
-----------------------------------------------------------------
IGFS communication protocol configured on server only. Client knows host and
port number, but does not know the protocol of the communication. It tries each
protocol (embedded, shmem, local TCP, remote TCP, in this order) until first
success.
The warnings appear because TCP protocol is configured to communicate with IGFS
(default), and on Unix systems the client first tries to connect using shmem
and fails. After that it tries again with TCP and succeeds.
Adding the following property to Ignite config solves the problem:
{code}
<property>
<name>[email protected]_local_shmem</name>
<value>true</value>
</property>
{code}
Two warnings appear because due to disabled filesystem instance caching upon
filesystem closing the filesystem instance is created again, what, in turn,
causes one more handshake (see
org.apache.ignite.hadoop.fs.v1.IgniteHadoopFileSystem#close) .
Currently (for the Sprint-2 release) we just disabled the handshake warnings on
server side.
IGNITE-462 created to clarify those issues.
was (Author: iveselovskiy):
IGFS communication protocol configured on server only. Client knows host and
port number, but does not know the protocol of the communication. It tries each
protocol (embedded, shmem, local TCP, remote TCP, in this order) until first
success.
The warnings appear because TCP protocol is configured to communicate with IGFS
(default), and on Unix systems the client first tries to connect using shmem
and fails. After that it tries again with TCP and succeeds.
Adding the following property to Ignite config solves the problem:
{code}
<property>
<name>[email protected]_local_shmem</name>
<value>true</value>
</property>
{code}
Two warnings appear because due to disabled filesystem instance caching upon
filesystem closing the filesystem instance is created again, what, in turn,
causes one more handshake (see
org.apache.ignite.hadoop.fs.v1.IgniteHadoopFileSystem#close) .
Currently (for the Sprint-2 release) we just disabled the handshake warnings on
server side.
> Each IGFS operation causes two "Handshake failed" warnings to appear
> --------------------------------------------------------------------
>
> Key: IGNITE-419
> URL: https://issues.apache.org/jira/browse/IGNITE-419
> Project: Ignite
> Issue Type: Bug
> Components: hadoop
> Affects Versions: sprint-1
> Reporter: Ivan Veselovsky
> Assignee: Vladimir Ozerov
> Priority: Minor
> Fix For: sprint-2
>
> Attachments: IGNITE-419--a.patch
>
>
> Pre-condition: Hadoop client configured to work with igfs:// (does not matter
> this is pure IGFS or a layer upon HDFS).
> In Hadoop client do any fs-related operation, e.g.
> ./hadoop fs -touchz /foo
> Observe in Ignite console the following 2 lines:
> [21:07:10,947][WARN ][igfs-client-worker-69-#192%null%][GridWorker] Handshake
> failed.
> [21:07:11,658][WARN ][igfs-client-worker-71-#194%null%][GridWorker] Handshake
> failed.
> However, in most cases no functionality appears to be broken.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
