Log using SLT_Notice on condfetch streaming delay

Log a notice message when delaying a conditional fetch and the stale
template object is still streaming.

Conflicts:
    bin/varnishd/cache/cache_fetch.c
    doc/sphinx/reference/vsl.rst

bin/varnishd/cache/cache_fetch.c

@@ -706,6 +706,34 @@ vbf_stp_condfetch(struct worker *wrk, struct busyobj *bo)
 
 	CHECK_OBJ_NOTNULL(wrk, WORKER_MAGIC);
 	CHECK_OBJ_NOTNULL(bo, BUSYOBJ_MAGIC);
+	CHECK_OBJ_NOTNULL(bo->fetch_objcore, OBJCORE_MAGIC);
+	CHECK_OBJ_NOTNULL(bo->stale_oc, OBJCORE_MAGIC);
+
+	stale_boc = HSH_RefBoc(bo->stale_oc);
+	CHECK_OBJ_ORNULL(stale_boc, BOC_MAGIC);
+	if (stale_boc) {
+		/* Wait for the stale object to become fully fetched, so
+		 * that we can catch fetch errors, before we unbusy the
+		 * new object. This serves two purposes. First it helps
+		 * with request coalesching, and stops long chains of
+		 * IMS-updated short-TTL objects all streaming from a
+		 * single slow body fetch. Second it makes sure that all
+		 * the object attributes are complete when we copy them
+		 * (this would be an issue for ie OA_GZIPBITS). */
+		VSLb(bo->vsl, SLT_Notice,
+		    "[core] Conditional fetch wait for streaming object");
+		ObjWaitState(bo->stale_oc, BOS_FINISHED);
+		stale_state = stale_boc->state;
+		HSH_DerefBoc(bo->wrk, bo->stale_oc);
+		stale_boc = NULL;
+		if (stale_state != BOS_FINISHED) {
+			(void)VFP_Error(bo->vfc, "Template object failed");
+			vbf_cleanup(bo);
+			wrk->stats->fetch_failed++;
+			return (F_STP_FAIL);
+		}
+	}
+	AZ(bo->stale_oc->flags & OC_F_FAILED);
 
 	AZ(vbf_beresp2obj(bo));
 

doc/sphinx/reference/vsl.rst

@@ -102,6 +102,26 @@ Error
 	Backend request failed to vcl_backend_error.
 
 
+NOTICE MESSAGES
+===============
+
+Notice messages contain informational messages about the handling of a
+request. These can be exceptional circumstances encountered that causes
+deviation from the normal handling. The messages are prefixed with [core]
+for core Varnish generated messages, or with [<VMOD name>] for VMOD
+generated messages. The [core] messages are described in detail below, see
+the individual VMOD documentation for messages from VMODs.
+
+[core] Conditional fetch wait for streaming object
+	The backend answered 304 Not Modified on a conditional fetch using
+	an object that has not yet been fully fetched as the stale
+	template object. This can only happen when the TTL of the object
+	is less than the time it takes to fetch it. The fetch is halted
+	until the stale object is fully fetched, upon which the new object
+	is created as normal. While waiting, any grace time on the stale
+	object will be in effect.
+
+
 HISTORY
 =======