Elasticsearch cannot downgrade a node from version [7.x.x] to version [7.x.x]

cannot downgrade a node from version [7.x.x] to version [7.x.x]
Elasticsearch降级实践

背景和问题分析

首先官方是不支持降级的，想要降级必须从快照中恢复，这是最保险的，前提也比较多，成本较高。
但是你做了很小的版本升级，过后又非常想回退到原来的版本，怎么办呢？

注意：我的Elasticsearch是单机模式，集群模式自行研究吧。

以我的操作环境为例：从7.13.4回退到7.11.2。
直接启动7.11.2，报错如下：

cannot downgrade a node from version [7.13.4] to version [7.11.2]

报错原因是启动时会检查，当前服务版本和集群内部状态的版本，源码如下

VSCode中打开该仓库：https://github1s.com/elastic/elasticsearch/blob/HEAD/server/src/main/java/org/elasticsearch/env/NodeMetadata.java

1. 最简单方案

版本做了很小的升级，你也很清楚，降级不会影响业务和程序。
最简单的方法，就是把集群内部状态的版本改下。

集群内部状态是基于lucene的文件系统，修改集群状态，代码如下：

public class Application {
    static final String NODE_VERSION_KEY = "node_version";

    //7130499 -> 7110299
    public static void main(String[] args) {
        int targetNodeVersion = 7110299;
        String _state = "/data/elk/elasticsearch-7.11.2/data/nodes/0/_state/";

        Path path = Paths.get(_state);
        try (DirectoryReader reader = DirectoryReader.open(new SimpleFSDirectory(path))) {
            final Map<String, String> userData = reader.getIndexCommit().getUserData();
            System.out.println(userData.toString());
            try (IndexWriter indexWriter =
                         createIndexWriter(new SimpleFSDirectory(path), true)) {
                final Map<String, String> commitData = new HashMap<>(userData);
                commitData.put(NODE_VERSION_KEY, Integer.toString(targetNodeVersion));
                indexWriter.setLiveCommitData(commitData.entrySet());
                indexWriter.commit();
            }
        } catch (Exception e) {
            e.printStackTrace();
        }
    }

    private static IndexWriter createIndexWriter(Directory directory, boolean openExisting) throws IOException {
        final IndexWriterConfig indexWriterConfig = new IndexWriterConfig(new KeywordAnalyzer());
        // start empty since we re-write the whole cluster state to ensure it is all using the same format version
        indexWriterConfig.setOpenMode(openExisting ? IndexWriterConfig.OpenMode.APPEND : IndexWriterConfig.OpenMode.CREATE);
        // only commit when specifically instructed, we must not write any intermediate states
        indexWriterConfig.setCommitOnClose(false);
        // most of the data goes into stored fields which are not buffered, so we only really need a tiny buffer
        indexWriterConfig.setRAMBufferSizeMB(1.0);
        // merge on the write thread (e.g. while flushing)
        indexWriterConfig.setMergeScheduler(new SerialMergeScheduler());

        return new IndexWriter(directory, indexWriterConfig);
    }
}

2. 复制数据仓库

如果你的数据很少，比如索引、账号、Kibana索引等。重新建设很快，那我建议还是重新建设吧。
如果重建过程非常耗时，成本比较高，并且你有另外一套相同版本的环境，那就好办了。
打包另一个环境的path.data目录，复制到当前环境中，然后重新设置账号信息即可。

# 1. 创建一个管理员账号
bin/elasticsearch-users useradd admin -p 123456 -r superuser

# 2. 重置elastic密码
curl --user admin:123456 -XPUT "http://localhost:9200/_security/user/elastic/_password?pretty" -H 'Content-Type: application/json' -d '{"password" : "123456"}'