Introduce the "busyobj" structure which is only valid for busy objects.

Use it to hold the derived vary string for the request (if any) and
skip busy objects with non-matching Vary during hash lookup.

Idea by:	sky

bin/varnishd/cache.h

@@ -97,6 +97,7 @@ struct director;
 struct object;
 struct objhead;
 struct objcore;
+struct busyobj;
 struct storage;
 struct workreq;
 struct vrt_backend;
@@ -281,6 +282,7 @@ struct worker {
 	struct objhead		*nobjhead;
 	struct objcore		*nobjcore;
 	struct waitinglist	*nwaitinglist;
+	struct busyobj		*nbusyobj;
 	void			*nhashpriv;
 	struct dstat		stats;
 
@@ -416,6 +418,7 @@ struct objcore {
 	void			*priv;
 	unsigned		priv2;
 	struct objhead		*objhead;
+	struct busyobj		*busyobj;
 	double			timer_when;
 	unsigned		flags;
 #define OC_F_BUSY		(1<<1)
@@ -470,7 +473,16 @@ oc_getlru(const struct objcore *oc)
 	return (oc->methods->getlru(oc));
 }
 
+/* Busy Object structure ---------------------------------------------*/
+
+struct busyobj {
+	unsigned		magic;
+#define BUSYOBJ_MAGIC		0x23b95567
+	uint8_t			*vary;
+};
+
 /* Object structure --------------------------------------------------*/
+
 VTAILQ_HEAD(storagehead, storage);
 
 struct object {

bin/varnishd/cache_hash.c

@@ -111,6 +111,11 @@ HSH_Prealloc(const struct sess *sp)
 	}
 	CHECK_OBJ_NOTNULL(w->nwaitinglist, WAITINGLIST_MAGIC);
 
+	if (w->nbusyobj == NULL) {
+		ALLOC_OBJ(w->nbusyobj, BUSYOBJ_MAGIC);
+		XXXAN(w->nbusyobj);
+	}
+
 	if (hash->prep != NULL)
 		hash->prep(sp);
 }
@@ -342,8 +347,15 @@ HSH_Lookup(struct sess *sp, struct objhead **poh)
 		assert(oc->objhead == oh);
 
 		if (oc->flags & OC_F_BUSY) {
-			if (!sp->hash_ignore_busy)
-				busy_oc = oc;
+			CHECK_OBJ_NOTNULL(oc->busyobj, BUSYOBJ_MAGIC);
+			if (sp->hash_ignore_busy)
+				continue;
+
+			if (oc->busyobj->vary != NULL &&
+			    !VRY_Match(sp, oc->busyobj->vary))
+				continue;
+
+			busy_oc = oc;
 			continue;
 		}
 
@@ -445,6 +457,10 @@ HSH_Lookup(struct sess *sp, struct objhead **poh)
 	AN(oc->flags & OC_F_BUSY);
 	oc->refcnt = 1;
 
+	w->nbusyobj->vary = sp->vary_b;
+	oc->busyobj = w->nbusyobj;
+	w->nbusyobj = NULL;
+
 	/*
 	 * Busy objects go on the tail, so they will not trip up searches.
 	 * HSH_Unbusy() will move them to the front.
@@ -609,6 +625,9 @@ HSH_Unbusy(const struct sess *sp)
 	VTAILQ_REMOVE(&oh->objcs, oc, list);
 	VTAILQ_INSERT_HEAD(&oh->objcs, oc, list);
 	oc->flags &= ~OC_F_BUSY;
+	AZ(sp->wrk->nbusyobj);
+	sp->wrk->nbusyobj = oc->busyobj;
+	oc->busyobj = NULL;
 	hsh_rush(oh);
 	AN(oc->ban);
 	Lck_Unlock(&oh->mtx);

bin/varnishtest/tests/c00043.vtc

+varnishtest "predictive vary"
+
+
+server s1 {
+	rxreq
+	txresp -hdr "Vary: foo" -bodylen 1
+	rxreq
+	sema r2 sync 2
+	sema r1 sync 2
+	txresp -hdr "Vary: foo" -bodylen 2
+} -start
+
+server s2 {
+	rxreq
+	txresp -hdr "Vary: foo" -bodylen 3
+} -start
+
+varnish v1 -vcl+backend {
+	sub vcl_recv {
+		if (req.http.bar) {
+			set req.backend = s2;
+		}
+	}
+} -start
+
+client c1 {
+	txreq -hdr "Foo: vary1"
+	rxresp
+	expect resp.bodylen == 1
+	txreq -hdr "Foo: vary2"
+	rxresp
+	expect resp.bodylen == 2
+} -start
+
+client c2 {
+	sema r2 sync 2
+	txreq -hdr "Foo: vary3" -hdr "bar: yes"
+	rxresp
+	sema r1 sync 2
+	expect resp.bodylen == 3
+} -start
+
+client c1 -wait
+client c2 -wait