[ovs-dev] [PATCH v6] Detailed packet drop statistics per dpdk and vhostuser ports

Sriram Vatala sriram.v at altencalsoftlabs.com
Wed Aug 28 11:51:48 UTC 2019


Hi All,

The patch " netdev-dpdk: Refactor vhost custom stats for extensibility" by
Ilya is merged to master. So I will rebase and send updated patch to avoid
any git conflicts.

Thanks & Regards,
Sriram.

-----Original Message-----
From: Sriram Vatala <sriram.v at altencalsoftlabs.com> 
Sent: 26 August 2019 18:29
To: ovs-dev at openvswitch.org; i.maximets at samsung.com
Cc: blp at ovn.org; ian.stokes at intel.com; Sriram Vatala
<sriram.v at altencalsoftlabs.com>
Subject: [PATCH v6] Detailed packet drop statistics per dpdk and vhostuser
ports

OVS may be unable to transmit packets for multiple reasons and today there
is a single counter to track packets dropped due to any of those reasons.
The most common reason is that a VM is unable to read packets fast enough
causing the vhostuser port transmit queue on the OVS side to become full.
This manifests as a problem with VNFs not receiving all packets. Having a
separate drop counter to track packets dropped because the transmit queue is
full will clearly indicate that the problem is on the VM side and not in
OVS. Similarly maintaining separate counters for all possible drops helps in
indicating sensible cause for packet drops.

This patch adds custom stats counters to track packets dropped at port level
and these counters are displayed along with other stats in "ovs-vsctl get
interface <iface> statistics"
command. The detailed stats will be available for both dpdk and vhostuser
ports.

Signed-off-by: Sriram Vatala <sriram.v at altencalsoftlabs.com>
---
 lib/netdev-dpdk.c                                  | 120
++++++++++++++++++---
 utilities/bugtool/automake.mk                      |   3 +-
 utilities/bugtool/ovs-bugtool-get-iface-stats      |  25 +++++
 .../bugtool/plugins/network-status/openvswitch.xml |   1 +
 vswitchd/vswitch.xml                               |  24 +++++
 5 files changed, 157 insertions(+), 16 deletions(-)  create mode 100755
utilities/bugtool/ovs-bugtool-get-iface-stats

diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c index 4805783..6685f32
100644
--- a/lib/netdev-dpdk.c
+++ b/lib/netdev-dpdk.c
@@ -447,8 +447,14 @@ struct netdev_dpdk {
 
     PADDED_MEMBERS(CACHE_LINE_SIZE,
         struct netdev_stats stats;
-        /* Custom stat for retries when unable to transmit. */
+        /* Counters for Custom device stats */
+        /* No. of retries when unable to transmit. */
         uint64_t tx_retries;
+        /* Pkts left untransmitted in Tx buffers. Probably Tx Que is full
*/
+        uint64_t tx_failure_drops;
+        uint64_t tx_mtu_exceeded_drops;
+        uint64_t tx_qos_drops;
+        uint64_t rx_qos_drops;
         /* Protects stats */
         rte_spinlock_t stats_lock;
         /* 4 pad bytes here. */
@@ -2205,6 +2211,7 @@ netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq,
     struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev);
     uint16_t nb_rx = 0;
     uint16_t dropped = 0;
+    uint16_t qos_drops = 0;
     int qid = rxq->queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
     int vid = netdev_dpdk_get_vid(dev);
 
@@ -2236,11 +2243,13 @@ netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq,
                                     (struct rte_mbuf **) batch->packets,
                                     nb_rx, true);
         dropped -= nb_rx;
+        qos_drops = dropped;
     }
 
     rte_spinlock_lock(&dev->stats_lock);
     netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets,
                                          nb_rx, dropped);
+    dev->rx_qos_drops += qos_drops;
     rte_spinlock_unlock(&dev->stats_lock);
 
     batch->count = nb_rx;
@@ -2266,6 +2275,7 @@ netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct
dp_packet_batch *batch,
     struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev);
     int nb_rx;
     int dropped = 0;
+    int qos_drops = 0;
 
     if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) {
         return EAGAIN;
@@ -2284,12 +2294,14 @@ netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct
dp_packet_batch *batch,
                                     (struct rte_mbuf **) batch->packets,
                                     nb_rx, true);
         dropped -= nb_rx;
+        qos_drops = dropped;
     }
 
     /* Update stats to reflect dropped packets */
     if (OVS_UNLIKELY(dropped)) {
         rte_spinlock_lock(&dev->stats_lock);
         dev->stats.rx_dropped += dropped;
+        dev->rx_qos_drops += qos_drops;
         rte_spinlock_unlock(&dev->stats_lock);
     }
 
@@ -2373,6 +2385,9 @@ __netdev_dpdk_vhost_send(struct netdev *netdev, int
qid,
     struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts;
     unsigned int total_pkts = cnt;
     unsigned int dropped = 0;
+    unsigned int tx_failure;
+    unsigned int mtu_drops;
+    unsigned int qos_drops;
     int i, retries = 0;
     int max_retries = VHOST_ENQ_RETRY_MIN;
     int vid = netdev_dpdk_get_vid(dev); @@ -2390,9 +2405,12 @@
__netdev_dpdk_vhost_send(struct netdev *netdev, int qid,
     rte_spinlock_lock(&dev->tx_q[qid].tx_lock);
 
     cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt);
+    mtu_drops = total_pkts - cnt;
+    qos_drops = cnt;
     /* Check has QoS has been configured for the netdev */
     cnt = netdev_dpdk_qos_run(dev, cur_pkts, cnt, true);
-    dropped = total_pkts - cnt;
+    qos_drops -= cnt;
+    dropped = qos_drops + mtu_drops;
 
     do {
         int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; @@ -2417,12
+2435,16 @@ __netdev_dpdk_vhost_send(struct netdev *netdev, int qid,
         }
     } while (cnt && (retries++ < max_retries));
 
+    tx_failure = cnt;
     rte_spinlock_unlock(&dev->tx_q[qid].tx_lock);
 
     rte_spinlock_lock(&dev->stats_lock);
     netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts,
                                          cnt + dropped);
     dev->tx_retries += MIN(retries, max_retries);
+    dev->tx_failure_drops += tx_failure;
+    dev->tx_mtu_exceeded_drops += mtu_drops;
+    dev->tx_qos_drops += qos_drops;
     rte_spinlock_unlock(&dev->stats_lock);
 
 out:
@@ -2447,12 +2469,15 @@ dpdk_do_tx_copy(struct netdev *netdev, int qid,
struct dp_packet_batch *batch)
     struct rte_mbuf *pkts[PKT_ARRAY_SIZE];
     uint32_t cnt = batch_cnt;
     uint32_t dropped = 0;
+    uint32_t tx_failure = 0;
+    uint32_t mtu_drops = 0;
+    uint32_t qos_drops = 0;
 
     if (dev->type != DPDK_DEV_VHOST) {
         /* Check if QoS has been configured for this netdev. */
         cnt = netdev_dpdk_qos_run(dev, (struct rte_mbuf **) batch->packets,
                                   batch_cnt, false);
-        dropped += batch_cnt - cnt;
+        qos_drops = batch_cnt - cnt;
     }
 
     uint32_t txcnt = 0;
@@ -2465,13 +2490,13 @@ dpdk_do_tx_copy(struct netdev *netdev, int qid,
struct dp_packet_batch *batch)
             VLOG_WARN_RL(&rl, "Too big size %u max_packet_len %d",
                          size, dev->max_packet_len);
 
-            dropped++;
+            mtu_drops++;
             continue;
         }
 
         pkts[txcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp);
         if (OVS_UNLIKELY(!pkts[txcnt])) {
-            dropped += cnt - i;
+            dropped = cnt - i;
             break;
         }
 
@@ -2488,13 +2513,17 @@ dpdk_do_tx_copy(struct netdev *netdev, int qid,
struct dp_packet_batch *batch)
             __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **)
pkts,
                                      txcnt);
         } else {
-            dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, txcnt);
+            tx_failure = netdev_dpdk_eth_tx_burst(dev, qid, pkts, 
+ txcnt);
         }
     }
 
+    dropped += qos_drops + mtu_drops + tx_failure;
     if (OVS_UNLIKELY(dropped)) {
         rte_spinlock_lock(&dev->stats_lock);
         dev->stats.tx_dropped += dropped;
+        dev->tx_failure_drops += tx_failure;
+        dev->tx_mtu_exceeded_drops += mtu_drops;
+        dev->tx_qos_drops += qos_drops;
         rte_spinlock_unlock(&dev->stats_lock);
     }
 }
@@ -2536,18 +2565,25 @@ netdev_dpdk_send__(struct netdev_dpdk *dev, int qid,
         dp_packet_delete_batch(batch, true);
     } else {
         int tx_cnt, dropped;
+        int tx_failure, mtu_drops, qos_drops;
         int batch_cnt = dp_packet_batch_size(batch);
         struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets;
 
         tx_cnt = netdev_dpdk_filter_packet_len(dev, pkts, batch_cnt);
+        mtu_drops = batch_cnt - tx_cnt;
+        qos_drops = tx_cnt;
         tx_cnt = netdev_dpdk_qos_run(dev, pkts, tx_cnt, true);
-        dropped = batch_cnt - tx_cnt;
+        qos_drops -= tx_cnt;
 
-        dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, tx_cnt);
+        tx_failure = netdev_dpdk_eth_tx_burst(dev, qid, pkts, tx_cnt);
 
+        dropped = tx_failure + mtu_drops + qos_drops;
         if (OVS_UNLIKELY(dropped)) {
             rte_spinlock_lock(&dev->stats_lock);
             dev->stats.tx_dropped += dropped;
+            dev->tx_failure_drops += tx_failure;
+            dev->tx_mtu_exceeded_drops += mtu_drops;
+            dev->tx_qos_drops += qos_drops;
             rte_spinlock_unlock(&dev->stats_lock);
         }
     }
@@ -2816,6 +2852,17 @@ netdev_dpdk_get_custom_stats(const struct netdev
*netdev,
     uint32_t i;
     struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
     int rte_xstats_ret;
+    uint16_t index = 0;
+
+#define DPDK_CSTATS                         \
+    DPDK_CSTAT(tx_failure_drops)            \
+    DPDK_CSTAT(tx_mtu_exceeded_drops)       \
+    DPDK_CSTAT(tx_qos_drops)                \
+    DPDK_CSTAT(rx_qos_drops)
+
+#define DPDK_CSTAT(NAME) +1
+    custom_stats->size = DPDK_CSTATS;
+#undef DPDK_CSTAT
 
     ovs_mutex_lock(&dev->mutex);
 
@@ -2830,9 +2877,10 @@ netdev_dpdk_get_custom_stats(const struct netdev
*netdev,
         if (rte_xstats_ret > 0 &&
             rte_xstats_ret <= dev->rte_xstats_ids_size) {
 
-            custom_stats->size = rte_xstats_ret;
+            index = rte_xstats_ret;
+            custom_stats->size += rte_xstats_ret;
             custom_stats->counters =
-                    (struct netdev_custom_counter *)
xcalloc(rte_xstats_ret,
+                    (struct netdev_custom_counter *) 
+ xcalloc(custom_stats->size,
                             sizeof(struct netdev_custom_counter));
 
             for (i = 0; i < rte_xstats_ret; i++) { @@ -2846,7 +2894,6 @@
netdev_dpdk_get_custom_stats(const struct netdev *netdev,
             VLOG_WARN("Cannot get XSTATS values for port:
"DPDK_PORT_ID_FMT,
                       dev->port_id);
             custom_stats->counters = NULL;
-            custom_stats->size = 0;
             /* Let's clear statistics cache, so it will be
              * reconfigured */
             netdev_dpdk_clear_xstats(dev); @@ -2855,6 +2902,27 @@
netdev_dpdk_get_custom_stats(const struct netdev *netdev,
         free(values);
     }
 
+    if (custom_stats->counters == NULL) {
+        custom_stats->counters =
+            (struct netdev_custom_counter *) xcalloc(custom_stats->size,
+                                  sizeof(struct netdev_custom_counter));
+    }
+
+    rte_spinlock_lock(&dev->stats_lock);
+    i = index;
+#define DPDK_CSTAT(NAME)                                     \
+    ovs_strlcpy(custom_stats->counters[i++].name, #NAME,     \
+                NETDEV_CUSTOM_STATS_NAME_SIZE);
+    DPDK_CSTATS;
+#undef DPDK_CSTAT
+
+    i = index;
+#define DPDK_CSTAT(NAME)                                     \
+    custom_stats->counters[i++].value = dev->NAME;
+    DPDK_CSTATS;
+#undef DPDK_CSTAT
+    rte_spinlock_unlock(&dev->stats_lock);
+
     ovs_mutex_unlock(&dev->mutex);
 
     return 0;
@@ -2865,17 +2933,39 @@ netdev_dpdk_vhost_get_custom_stats(const struct
netdev *netdev,
                                    struct netdev_custom_stats
*custom_stats)  {
     struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
+    int i;
+
+#define VHOST_CSTATS                        \
+    VHOST_CSTAT(tx_retries)                 \
+    VHOST_CSTAT(tx_failure_drops)           \
+    VHOST_CSTAT(tx_mtu_exceeded_drops)      \
+    VHOST_CSTAT(tx_qos_drops)               \
+    VHOST_CSTAT(rx_qos_drops)
+
+#define VHOST_CSTAT(NAME) +1
+    custom_stats->size = VHOST_CSTATS;
+#undef VHOST_CSTAT
 
-    ovs_mutex_lock(&dev->mutex);
 
-    custom_stats->size = VHOST_CUSTOM_STATS_SIZE;
     custom_stats->counters = xcalloc(custom_stats->size,
                                      sizeof *custom_stats->counters);
-    ovs_strlcpy(custom_stats->counters[0].name, VHOST_STAT_TX_RETRIES,
+    i = 0;
+#define VHOST_CSTAT(NAME)                                    \
+    ovs_strlcpy(custom_stats->counters[i++].name, #NAME,     \
                 NETDEV_CUSTOM_STATS_NAME_SIZE);
+    VHOST_CSTATS;
+#undef VHOST_CSTAT
+
+    ovs_mutex_lock(&dev->mutex);
 
     rte_spinlock_lock(&dev->stats_lock);
-    custom_stats->counters[0].value = dev->tx_retries;
+
+    i = 0;
+#define VHOST_CSTAT(NAME)                                    \
+    custom_stats->counters[i++].value = dev->NAME;
+    VHOST_CSTATS;
+#undef VHOST_CSTAT
+
     rte_spinlock_unlock(&dev->stats_lock);
 
     ovs_mutex_unlock(&dev->mutex);
diff --git a/utilities/bugtool/automake.mk b/utilities/bugtool/automake.mk
index 18fa347..9657468 100644
--- a/utilities/bugtool/automake.mk
+++ b/utilities/bugtool/automake.mk
@@ -22,7 +22,8 @@ bugtool_scripts = \
 	utilities/bugtool/ovs-bugtool-ovs-bridge-datapath-type \
 	utilities/bugtool/ovs-bugtool-ovs-vswitchd-threads-affinity \
 	utilities/bugtool/ovs-bugtool-qos-configs \
-	utilities/bugtool/ovs-bugtool-get-dpdk-nic-numa
+	utilities/bugtool/ovs-bugtool-get-dpdk-nic-numa \
+	utilities/bugtool/ovs-bugtool-get-iface-stats
 
 scripts_SCRIPTS += $(bugtool_scripts)
 
diff --git a/utilities/bugtool/ovs-bugtool-get-iface-stats
b/utilities/bugtool/ovs-bugtool-get-iface-stats
new file mode 100755
index 0000000..0fe175e
--- /dev/null
+++ b/utilities/bugtool/ovs-bugtool-get-iface-stats
@@ -0,0 +1,25 @@
+#! /bin/bash
+
+# This library is free software; you can redistribute it and/or # 
+modify it under the terms of version 2.1 of the GNU Lesser General # 
+Public License as published by the Free Software Foundation.
+#
+# This library is distributed in the hope that it will be useful, # but 
+WITHOUT ANY WARRANTY; without even the implied warranty of # 
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU # 
+Lesser General Public License for more details.
+#
+# Copyright (C) 2019 Ericsson AB
+
+for bridge in `ovs-vsctl -- --real list-br` do
+    echo -e "\nBridge : ${bridge}\n"
+    for iface in `ovs-vsctl list-ifaces ${bridge}`
+    do
+        echo -e "iface : ${iface}"
+        ovs-vsctl get interface ${iface} statistics
+        echo -e "\n"
+    done
+    echo -e "iface : ${bridge}"
+    ovs-vsctl get interface ${bridge} statistics done
diff --git a/utilities/bugtool/plugins/network-status/openvswitch.xml
b/utilities/bugtool/plugins/network-status/openvswitch.xml
index d39867c..f8c4ff0 100644
--- a/utilities/bugtool/plugins/network-status/openvswitch.xml
+++ b/utilities/bugtool/plugins/network-status/openvswitch.xml
@@ -34,6 +34,7 @@
     <command label="ovs-appctl-dpctl-dump-flows-netdev" filters="ovs"
repeat="2">ovs-appctl dpctl/dump-flows netdev at ovs-netdev</command>
     <command label="ovs-appctl-dpctl-dump-flows-system" filters="ovs"
repeat="2">ovs-appctl dpctl/dump-flows system at ovs-system</command>
     <command label="ovs-appctl-dpctl-show-s" filters="ovs"
repeat="2">ovs-appctl dpctl/show -s</command>
+    <command label="ovs-vsctl-get-interface-statistics" 
+ filters="ovs">/usr/share/openvswitch/scripts/ovs-bugtool-get-iface-sta
+ ts</command>
     <command label="ovs-ofctl-show"
filters="ovs">/usr/share/openvswitch/scripts/ovs-bugtool-ovs-ofctl-loop-over
-bridges "show"</command>
     <command label="ovs-ofctl-dump-flows" filters="ovs"
repeat="2">/usr/share/openvswitch/scripts/ovs-bugtool-ovs-ofctl-loop-over-br
idges "dump-flows"</command>
     <command label="ovs-ofctl-dump-ports" filters="ovs"
repeat="2">/usr/share/openvswitch/scripts/ovs-bugtool-ovs-ofctl-loop-over-br
idges "dump-ports"</command> diff --git a/vswitchd/vswitch.xml
b/vswitchd/vswitch.xml index 9a743c0..4a7bcf8 100644
--- a/vswitchd/vswitch.xml
+++ b/vswitchd/vswitch.xml
@@ -3486,6 +3486,30 @@ ovs-vsctl add-port br0 p0 -- set Interface p0
type=patch options:peer=p1 \
           the above.
         </column>
       </group>
+      <group title="Statistics: Custom stats">
+        <column name="statistics" key="tx_retries">
+          Total number of transmit retries on a vhost-user or
vhost-user-client
+          interface.
+        </column>
+        <column name="statistics" key="tx_failure_drops">
+          Total number of packets dropped because DPDP transmit API for
+          physical/vhost ports fails to transmit the packets. This happens
+          most likely because the transmit queue is full or has been filled
+          up. There are other reasons as well which are unlikely to happen.
+        </column>
+        <column name="statistics" key="tx_mtu_exceeded_drops">
+          Number of packets dropped due to packet length exceeding the max
+          device MTU.
+        </column>
+        <column name="statistics" key="tx_qos_drops">
+          Total number of packets dropped due to transmission rate
exceeding
+          the configured egress policer rate.
+        </column>
+        <column name="statistics" key="rx_qos_drops">
+          Total number of packets dropped due to reception rate exceeding
+          the configured ingress policer rate.
+        </column>
+      </group>
     </group>
 
     <group title="Ingress Policing">
--
2.7.4


More information about the dev mailing list