When Nuxeo is configured to use managed aliases for the elastic index.
During a full re-indexing, a new write index is created, the search index alias point to the existing index, once the re-indexing is completed the search index alias is updated to point to the new write index..
But starting a new node while the re-indexing is still running results in updating the search index alias too early.
This happens with legacy re-indexing using the WorkManager or with the Bulk indexing action
See org.nuxeo.elasticsearch.core.ElasticSearchAdminImpl#initializeIndexes which is called during service start, which is calling initIndex with sync=true.