Jemter uses 100 threads to access the nginx homepage and executes reload nginx every two seconds（nginx worker auto = 8）. soon nginx or vpp will crash. I found vcl and vls share 
var __vcl_worker_index, when old nginx exit and new nginx fork,  the index of vcl and vls wokers in pool is not match, so i solve it (add lock bettwen vcom&vcl exit and child_fork ). now testting continue..., but vpp will infinite loop in session_wrk_handle_evts_main_rpc, so i solve it again (using "_lprev(he, evt_list) != _lnext(he, evt_list)" instead of "!clib_llist_is_empty (wrk->event_elts, evt_list, he)" in function session_mq_connect_handler), be continue! Oops, nginx crash again in vcl_send_session_accepted_reply (because session->vpp_evt_q = 0) called by vppcom_session_unbind..., sometimes vpp carshed in app_worker_get, now i giveup and report the bug. 
 
the patch to fix bug：
v2306.patch
 
^{session_wrk_handle_evts_main_rpc test code:}

clib_llist_test.c

^{some backtrace：}