test/adq: Simplify connection counting strategy

We can use the nvmf_get_stats RPC to directly count the number of
connections on each poll group.

Also, we cannot rely on knowing the exact algorithm that the driver will
use to assign connections to NAPI values. We can only know that the
number of unique NAPI values is AT MOST equal to the number of traffic
classes.

Fixes #2577

Change-Id: I6758fdcda752ff5372b1bc59984257b804ad0ea7
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/13833
Reviewed-by: wanghailiang <hailiangx.e.wang@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>
Reviewed-by: Aleksey Marchuk <alexeymar@nvidia.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>

test/nvmf/target/perf_adq.sh

@@ -53,44 +53,28 @@ function adq_reload_driver() {
 	sleep 5
 }
 
-function get_nvmf_poll_groups() {
-	"$rpc_py" thread_get_pollers | jq -r '.threads[] | .active_pollers[] |
-	select(.name == "nvmf_poll_group_poll").busy_count'
-}
-
-function num_busy_count() {
-	get_pollers_busy_count0=($(get_nvmf_poll_groups))
-	sleep 2
-	get_pollers_busy_count1=($(get_nvmf_poll_groups))
-	local num=0
-	for i in "${!get_pollers_busy_count0[@]}"; do
-		increment=$((get_pollers_busy_count1[i] - get_pollers_busy_count0[i]))
-		if ((increment > 0)); then
-			((++num))
-		fi
-	done
-	echo $num
-}
-
 # Clear the previous configuration that may have an impact.
 # At present, ADQ configuration is only applicable to the ice driver.
 adq_reload_driver
 
-# Testcase 1 and Testcase 2 show the SPDK interacting with ADQ.
+# We are going to run the test twice, once with ADQ enabled and once with it disabled.
-# The number of continuously increasing nvmf_poll_group_poll's busy_count, we define it as "num_busy_count".
+# The nvmf target is given 4 cores and ADQ creates queues in one traffic class. We then run
-# When ADQ enabled, num_busy_count will be equal to the number of tc1 queues of traffic classes.
+# perf with 4 cores (i.e. 4 connections) and examine how the connections are allocated to the nvmf target's
-# When ADQ disabled, num_busy_count will be equal to the smaller value of initiator connections and target cores.
+# poll groups.
-# Testcase 1: Testing 2 traffic classes and 2 tc1 queues without ADQ
+
+# When ADQ is disabled, we expect 1 connection on each of the 4 poll groups.
 nvmftestinit
 adq_start_nvmf_target 0 0xF
 sleep 2
-$perf -q 64 -o 4096 -w randread -t 10 -c 0x70 \
+$perf -q 64 -o 4096 -w randread -t 10 -c 0xF0 \
 	-r "trtype:${TEST_TRANSPORT} adrfam:IPv4 traddr:${NVMF_FIRST_TARGET_IP} trsvcid:${NVMF_PORT} \
-subnqn:nqn.2016-06.io.spdk:cnode1" &
+	subnqn:nqn.2016-06.io.spdk:cnode1" &
 perfpid=$!
 sleep 2
-if [[ $(num_busy_count) -ne 3 ]]; then
+
-	echo "ERROR: num_busy_count != cores of initiators! Testcase 1 failed."
+count=$("$rpc_py" nvmf_get_stats | jq -r '.poll_groups[] | select(.current_io_qpairs == 1) | length' | wc -l)
+if [[ "$count" -ne 4 ]]; then
+	echo "ERROR: With ADQ disabled, connections were not evenly distributed amongst poll groups!"
 	exit 1
 fi
 wait $perfpid
@@ -98,24 +82,24 @@ nvmftestfini
 
 adq_reload_driver
 
-# Testcase 2: Testing 2 traffic classes and 2 tc1 queues with ADQ
+# When ADQ is enabled, we expect the connections to reside on AT MOST two poll groups.
 nvmftestinit
 sleep 2
 adq_configure_driver
 adq_start_nvmf_target 1 0xF
 sleep 2
-# The number of I/O connections from initiator is the core count * qpairs per ns, so here its 12.
+$perf -q 64 -o 4096 -w randread -t 10 -c 0xF0 \
-# ADQ on target side will work if 12 connections are matched to two out of four cores on the target.
-$perf -q 64 -o 4096 -w randread -t 15 -P 4 -c 0x70 \
 	-r "trtype:${TEST_TRANSPORT} adrfam:IPv4 traddr:${NVMF_FIRST_TARGET_IP} trsvcid:${NVMF_PORT} \
-subnqn:nqn.2016-06.io.spdk:cnode1" &
+	subnqn:nqn.2016-06.io.spdk:cnode1" &
 perfpid=$!
-sleep 3
+sleep 2
 
-if [[ $(num_busy_count) -ne 2 ]]; then
+count=$("$rpc_py" nvmf_get_stats | jq -r '.poll_groups[] | select(.current_io_qpairs == 0) | length' | wc -l)
-	echo "ERROR: num_busy_count != tc1 queues of traffic classes! Testcase 2 failed."
+if [[ "$count" -lt 2 ]]; then
+	echo "ERROR: With ADQ enabled, did not find 0 connections on 2 of the poll groups!"
 	exit 1
 fi
+
 wait $perfpid
 nvmftestfini