How to decease bulk queue size?

I find some of nodes were running into max bulk queue. But after I stop all my rsyslog/logstash process, the bulk queue size didn't decease. Some other nodes would even increase its bulk queue size!

I try to get hot_threads of such node, got only two lucene segment merge thread. (old indices)

Are you sure you are looking at the number of entries in the queue or are
you looking at the number of threads spun up to handle bulk? Have a look
for more on these queues.