[ https://issues.apache.org/jira/browse/IGNITE-9858?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16714767#comment-16714767 ]
Pavel Pereslegin commented on IGNITE-9858: ------------------------------------------ [~DmitriyGovorukhin] thank you, I checked the approach you suggested, and as a result I refactored {{TcpDiscoveryMulticastIpFinder}} to resolve (and collect) local addresses on demand, otherwise it's possible that a client node cannot find cluster from another subnet (I can explain in more detail how to reproduce this case). Also, {{TcpDiscoverySharedFsIpFinder}} uses injected ignite instance resource to obtain working directory. If we replace this code with IgniteUtils.defaultWorkDirectory(), we may lose backward compatibility - so I decided to keep this code unchanged. Please take a look at these changes. > [Test Failed] SystemCacheNotConfiguredTest#test flaky fails on TC (timeout). > ---------------------------------------------------------------------------- > > Key: IGNITE-9858 > URL: https://issues.apache.org/jira/browse/IGNITE-9858 > Project: Ignite > Issue Type: Bug > Affects Versions: 2.6 > Reporter: Pavel Pereslegin > Assignee: Pavel Pereslegin > Priority: Major > Labels: MakeTeamcityGreenAgain > Fix For: 2.8 > > > SystemCacheNotConfiguredTest hangs sometimes on TeamCity (timeout). > Example of such failures on master branch: > [https://ci.ignite.apache.org/project.html?projectId=IgniteTests24Java8&buildTypeId=&tab=testDetails&testNameId=-2762467041583095183&order=TEST_STATUS_DESC&itemsCount=50&branch_IgniteTests24Java8=%3Cdefault%3E] > When we using ip finder in shared mode each node should register self address > (except clients, obviously). > Check that the node is a client uses installed (via DI > @IgniteInstanceResource) Ignite instance (see > {{TcpDiscoveryIpFinderAdapter#initializeLocalAddresses}}). > So when client and server starts simultaneously, the following scenario is > possible - Ignite server injected at first, then the Ignite client injected > when the SPI is initialized ({{spiStart}}) both nodes assume that the local > node is a client and don't register local address. > {noformat} > [2018-10-11 18:03:49,794][WARN > ][tcp-client-disco-msg-worker-#57%client%][TcpDiscoverySpi] IP finder > returned empty addresses list. Please check IP finder configuration. Will > retry every 2000 ms. Change 'reconnectDelay' to configure the frequency of > retries.{noformat} -- This message was sent by Atlassian JIRA (v7.6.3#76005)