I'm encountering the same issue.

[wangzhiliang99]

My environment is Flink 2.0.0 + flink-connector-elasticsearch8:4.0.0-2.0.

I tried placing the Elasticsearch-related dependency JARs into the flink/lib/ directory, which resolved the initial problem but introduced a new issue during recovery:
2025-09-29 14:10:07,036 WARN org.apache.flink.runtime.taskmanager.Task [] - Source: kafka source -> Map -> Filter -> Sink: Writer (1/3)#144 (15db1e631d1b496a28f441db3787ddcc_cbc357ccb763df2852fee8c4fc7d55f2_0_144) switched from RUNNING to FAILED with failure cause: java.lang.IncompatibleClassChangeError: Class co.elastic.clients.elasticsearch.core.bulk.IndexOperation does not implement the requested interface co.elastic.clients.elasticsearch.core.bulk.BulkOperationVariant at co.elastic.clients.elasticsearch.core.bulk.BulkOperation.<init>(BulkOperation.java:103) ~[blob_p-873649d343acc55522a2748b6694e6c3343e6ab7-747a9878f1de5fca64c3bb34670856f7:?] at org.apache.flink.connector.elasticsearch.sink.Elasticsearch8AsyncWriter.submitRequestEntries(Elasticsearch8AsyncWriter.java:125) ~[blob_p-873649d343acc55522a2748b6694e6c3343e6ab7-747a9878f1de5fca64c3bb34670856f7:?] at org.apache.flink.connector.base.sink.writer.AsyncSinkWriter.flush(AsyncSinkWriter.java:339) ~[flink-connector-files-2.0.0.jar:2.0.0] ... at java.base/java.lang.Thread.run(Thread.java:842) [?:?]
Do you have any solutions for this now?

[mayorandrew]

Hi @wangzhiliang99, thanks for reaching out!

Unfortunately, I don't have a solution neither for the original problem, nor for the new problem you mention. I have raised a bug ticket in Flink Jira: https://issues.apache.org/jira/browse/FLINK-38054, however I haven't heard any response from anyone so far. Please vote for it there

[mayorandrew]

Hi @wangzhiliang99,

Seems like a small change in the flink-connector-elasticsearch8 source code fixes the problem. I have opened a pull request: apache/flink-connector-elasticsearch#136

Not sure if this is the right solution, I will be waiting for more feedback from Apache Flink team.

Until the PR is merged and published, you can checkout the branch, publish it to a local or private maven registry, and configure your Flink job project to install a forked package. Note that the PR builds on top of version 3.1.0 for compatibility with Flink 1.20. For newer versions of Flink feel free to merge newer updates from main before publishing.