简体繁体 English

分布式模式 Infinispan 的节点之间是否可能存在数据延迟？

[英]Is it possible to have a data delay between nodes in distributed mode Infinispan?

原文 2020-07-07 10:15:53 1 1 caching/ distributed-computing/ infinispan/ distributed-caching/ infinispan-9

With Infinispan v9, my app uses a distributed async cache mode where number of cluster-wide data replicas is 3.使用 Infinispan v9，我的应用程序使用分布式异步缓存模式，其中集群范围的数据副本数为 3。

Let's say data owner/replica of cache entry 123 is in Nodes A, B, C .假设缓存条目 123的数据所有者/副本位于节点 A、B、C中。

With this, I have two questions (assuming they were not caused by network issues)有了这个，我有两个问题（假设它们不是由网络问题引起的）

If Node C enquires for entry 123 , is it possible that Node C will find the entry missing due to replication delay?如果节点 C查询条目 123 ，节点 C 是否会发现由于复制延迟而丢失的条目？
If Node D enquires for entry 123 , is it possible that Node D will find the entry missing due to some sort of lag?如果节点 D查询条目 123 ，节点 D 是否会发现由于某种滞后而丢失的条目？

1 个解决方案

Assuming node A as primary owner of entry 123 (each key has 1 primary owner and N-1 backup owners, where N is the number of replicas)假设节点A是条目 123的主要所有者（每个密钥有 1 个主要所有者和N-1备份所有者，其中N是副本数）

Yes, node C can read stale values due to replication lag.是的，由于复制滞后，节点C可以读取过时的值。
If you have some lag, yes it can happen.如果你有一些滞后，是的，它可能会发生。 When a read happens in a non-owner node (like D ), it first tries to fetch the value from the primary owner (node A ).当在非所有者节点（如D ）中发生读取时，它首先尝试从主所有者（节点A ）获取值。 If A doesn't reply in time, then it goes to the next owner (node B ) and so one.如果A没有及时回复，那么它会转到下一个所有者（节点B ）等等。

Now, if A replies in time, it won't read the stale the value.现在，如果A及时回复，它就不会读取过时的值。 Otherwise, it may read the stale value from node B or C否则，它可能会从节点B或C读取过时的值

You have this information (and more) in the documentation .您在文档中有这些信息（以及更多信息）。