OpenDaylight JIRA

OpenDaylight JIRA https://jira.opendaylight.org This file is an XML representation of an issue en-us

8.20.10

820010

22-06-2022

[NETVIRT-1110] wrong bind and unbind leads to stale data https://jira.opendaylight.org/browse/NETVIRT-1110 netvirt NETVIRT-1110

wrong bind and unbind leads to stale data

Bug Medium Resolved Done Faseela K Sam Hague Wed, 21 Feb 2018 03:42:29 +0000 Tue, 2 Oct 2018 19:44:01 +0000 Tue, 2 Oct 2018 19:44:01 +0000 Nitrogen Carbon Oxygen Fluorine General 0 3 A lot of stale service bindings are seen in netvirt CSIT runs with respect to ACL service, This needs to be fixed as this is a potential issue which can lead to stale flows in OVS, causing even bigger troubles in longer run. Assigning this to Som, for debugging from ACL perspective. <a href="https://git.opendaylight.org/gerrit/#/c/68425/" class="external-link" target="_blank" rel="nofollow noopener">https://git.opendaylight.org/gerrit/#/c/68425/</a> ACL related fix is merged <a href="https://git.opendaylight.org/gerrit/#/c/68425/" class="external-link" target="_blank" rel="nofollow noopener">https://git.opendaylight.org/gerrit/#/c/68425/</a>. Assigning back to Faseela for interface manager fix. <a href="https://git.opendaylight.org/gerrit/#/c/68559/" class="external-link" target="_blank" rel="nofollow noopener">https://git.opendaylight.org/gerrit/#/c/68559/</a> Trying the fix, this will work only if there are no services which does back to back unbind and bind. Will run CSIT to confirm the same. Else interface-manager trying to remove the parent, will result in ConflictingModificationExceptions. <a href="https://jira.opendaylight.org/secure/ViewProfile.jspa?name=k.faseela" class="user-hover" rel="k.faseela">k.faseela</a> is this still an issue? <a href="https://jira.opendaylight.org/secure/ViewProfile.jspa?name=shague" class="user-hover" rel="shague">shague</a> : As per the discussions we had in mail with <a href="https://jira.opendaylight.org/secure/ViewProfile.jspa?name=thapar" class="user-hover" rel="thapar">thapar</a> and others, the final interfacemanager side fix cannot be tried, as that can lead to multiple other problems. We cannot really decide when to remove the dangling parent node in service-binding DS, as there is no deterministic way of identifying when all services are unbound. <a href="https://jira.opendaylight.org/secure/ViewProfile.jspa?name=thapar" class="user-hover" rel="thapar">thapar</a> : After the ACL fix, there is no more functionality impact left with the bug, and <a href="https://jira.opendaylight.org/secure/ViewProfile.jspa?name=thapar" class="user-hover" rel="thapar">thapar</a>'s suggestion was that we can use the stale clean-up script to address these pending entries. <a href="https://jira.opendaylight.org/secure/ViewProfile.jspa?name=k.faseela" class="user-hover" rel="k.faseela">k.faseela</a> are you saying there will always be stale entries? If so what are the stale data exactly - flows, model data etc? And the only way to clean it up is to have a script running that cleans it up? That doesn't sound like a fix. service-binding DS is a DS shared across 7 applications in netvirt. ELAN, ACL, DHCP etc writes to this DS. it has a format like { interface-name -> list(services). } . Initially we had issues that some of the services in the list were not cleaned up. Som has fixed that. After that, he was pointing that, once all services are unbound properly, the parent node just lies there. ie, {interfaceName->[empty-list]}. Since interface-manager is not the owner of the writes to this DS, it is difficult for genius to clean up this DS when all services finish unbinding. For eg : in VM migration usecases, applications unbind and bind back to back. So, how do we decide when to remove this parent dangling node, and when not to?- Development Rank 0|i03akn: