If requests come in fast enough on a single connection, typically

when running synthetic benchmarks, we need to reset the worker threads
workspace between requests, or we will eventually run out.

This is an embarrasingly old bug.

Fixes: #808



git-svn-id: http://www.varnish-cache.org/svn/trunk/varnish-cache@5491 d4fa192b-c00b-0410-8231-f00ffab90ce4

bin/varnishd/cache_center.c

@@ -291,6 +291,7 @@ cnt_done(struct sess *sp)
 		WRK_SumStat(sp->wrk);
 	/* Reset the workspace to the session-watermark */
 	WS_Reset(sp->ws, sp->ws_ses);
+	WS_Reset(sp->wrk->ws, NULL);
 
 	i = HTC_Reinit(sp->htc);
 	if (i == 1) {