简体繁体 English

如何在 SolrCloud 上强制领导者？

[英]How to force a leader on SolrCloud?

原文 2018-08-28 19:53:50 6 2 solr/ apache-zookeeper/ solrcloud

I have a 5-node SolrCloud (Solr 7.0) with an external 3-node Zookeeper ensemble.我有一个带有外部 3 节点 Zookeeper 集成的 5 节点 SolrCloud (Solr 7.0)。 There is one collection called "production" that is sharded to 5 shards with a replication factor of 5. See the screenshot below:有一个名为“production”的集合被分片为 5 个分片，复制因子为 5。请参见下面的屏幕截图：

shard5 was struggling to elect a new leader for a long time and other cores were complaining with the following error: shard5 很长一段时间都在努力选举新的领导者，其他核心抱怨以下错误：

azsolr1 solr: 2018-08-28 19:32:43.575 ERROR (qtp1124317168-9304) [c:production s:shard2 r:core_node9 x:production_shard2_replica_n4] oashRequestHandlerBase org.apache.solr.common.SolrException: No registered leader was found after waiting for 4000ms , collection: production slice: shard5 azsolr1 solr：2018-08-28 19:32:43.575 错误（qtp1124317168-9304）[c:production s:shard2 r:core_node9 x:production_shard2_replica_n4]等待 4000 毫秒，集合：生产切片：shard5

After restarting all nodes one by one (I even restarted the zookeeper nodes), I had no luck in electing the only active replica (azsolr1) as the leader.一一重启所有节点后（我什至重启了zookeeper节点），我没有运气选择唯一的活动副本（azsolr1）作为领导者。 I then unloaded the 4 replicas with the 'down' state using the CoreAdmin API UNLOAD command which caused the replicas to disappear completely.然后我使用CoreAdmin API UNLOAD命令卸载了 4 个处于“关闭”状态的副本，这导致副本完全消失。

With that setup, trying to force the leader of the shard using the Collection API FORCELEADER does nothing.使用该设置，尝试使用Collection API FORCELEADER强制分片的领导者什么也不做。 I also tried this before unloading the cores.在卸载内核之前，我也试过这个。

Here is the current status:这是目前的状态：

Why can't Solr just elect the only active replica for shard 5 as the leader?为什么 Solr 不能只选举分片 5 的唯一活动副本作为领导者？ Isn't this obvious, especially after forcing the leader on the shard?这不是很明显，尤其是在强制领导者进入分片之后？

Assuming the leader was elected successfully somehow, do I recreate the replicas that I deleted using the Collection API ADDREPLICA ?假设以某种方式成功选举了领导者，我是否使用Collection API ADDREPLICA重新创建我删除的副本？ In this case, should I reuse the same instanceDir and dataDir of the deleted replicas?在这种情况下，我应该重用已删除副本的相同instanceDir和dataDir吗？ Or I just let it replicate from scratch?或者我只是让它从头开始复制？

2 个解决方案

Restarting azsolr1 which was hosting the only replica for shard5 forced the election of the leader.重新启动托管azsolr1唯一副本的shard5强制选举领导者。 Sounds crazy, but that was it.听起来很疯狂，但就是这样。 After doing that, I added the other 4 replicas using the ADDREPLICA command.这样做之后，我使用ADDREPLICA命令添加了其他 4 个副本。

I had the same problem.我有同样的问题。

one collection with 3 replicas (solr1 --> was a leader before, solr2, solr3).一个包含 3 个副本的集合（solr1 --> 之前是领导者，solr2，solr3）。 one of the shards has no leader!其中一个碎片没有领导者！ and I did these steps :我做了这些步骤：

1 - stop solr2 and solr3 1 - 停止solr2和solr3

2- call FORCE LEADER API ( http://xx.xx.xxx.xx:8983/solr/admin/collections?action=FORCELEADER&collection=your_collection_name&shard=shard1 ) 2- 调用 FORCE LEADER API ( http://xx.xx.xxx.xx:8983/solr/admin/collections?action=FORCELEADER&collection=your_collection_name&shard=shard1 )

3 - after a few minutes solr1 elected as a leader 3 - after a few minutes solr1 elected as a leader