After HSEARCH-2764 In Progress (and maybe also before, but probably to a lesser extent), when Elasticsearch request processing threads are interrupted, the requests are considered as failed, and their content is printed out as part of the stack trace. It would be acceptable (-ish) if it was only the currently executing request, but it seems pending requests also get printed out, resulting in huge stack traces (we're talking about dozens of megabytes) when under heavy load, for instance when mass indexing.