Linux Network Troubleshooting
Internal
Overview
This page needs to be reviewed and re-organized.
Organizatorium
- TCP zero windows sent when application receive buffer is empty https://access.redhat.com/solutions/2480661
Packet Capture and Analysis
tcpdump -s 0 -i eno16780032 -w /tmp/$HOSTNAME.pcap
Network Monitoring
On each node, run the monitor.sh script: https://access.redhat.com/articles/1311173. This script will record OS network stats at a set interval and it will allow monitoring changes over time and correlate these changes with packet capture data.
Network Driver Error Messages
grep vmxnet3 sos_commands/kernel/dmesg [ 5.731844] VMware vmxnet3 virtual NIC driver - version 1.1.30.0-k-NAPI [ 5.731858] vmxnet3 0000:0b:00.0: # of Tx queues : 4, # of Rx queues : 4 [ 5.737730] vmxnet3 0000:0b:00.0: irq 72 for MSI/MSI-X [ 5.737786] vmxnet3 0000:0b:00.0: irq 73 for MSI/MSI-X [ 5.737860] vmxnet3 0000:0b:00.0: irq 74 for MSI/MSI-X [ 5.737891] vmxnet3 0000:0b:00.0: irq 75 for MSI/MSI-X [ 5.737916] vmxnet3 0000:0b:00.0: irq 76 for MSI/MSI-X [ 5.738367] vmxnet3 0000:0b:00.0 eth0: NIC Link is Up 10000 Mbps [ 8.186233] vmxnet3 0000:0b:00.0 eno16780032: intr type 3, mode 0, 5 vectors allocated [ 8.187854] vmxnet3 0000:0b:00.0 eno16780032: NIC Link is Up 10000 Mbps
Kernel Network Paramenters
cat etc/sysctl.conf
net.ipv4.ip_forward = 0 net.ipv4.conf.default.rp_filter = 1 net.ipv4.conf.default.accept_source_route = 0 net.ipv4.conf.all.accept_redirects=0 net.ipv4.conf.default.accept_redirects=0 net.ipv4.conf.all.log_martians=1 net.ipv4.conf.default.log_martians=1 net.core.wmem_max = 12582912 net.core.rmem_max = 26214400 net.ipv4.tcp_rmem = 10240 87380 26214400 net.ipv4.tcp_window_scaling = 1 net.ipv4.tcp_timestamps = 1 net.ipv4.tcp_sack = 1 net.ipv4.tcp_no_metrics_save = 1 net.core.netdev_max_backlog = 5000 net.ipv4.conf.all.send_redirects = 0 net.ipv4.conf.default.send_redirects = 0 net.ipv6.conf.all.disable_ipv6 = 1 kernel.sysrq = 0 kernel.core_uses_pid = 1 net.ipv4.tcp_syncookies = 1 kernel.msgmnb = 65536 kernel.msgmax = 65536 kernel.shmmax = 68719476736 kernel.shmall = 4294967296 fs.suid_dumpable = 0
- Working up the stack we see what is effectively zero packet loss at the NIC level in the ethtool -S output for each node. On all nodes, the rate of loss at the NIC is about or less than 1 frame *per day*. This is far from significant.
g61mps-jhq01-2017-Jul-16-16:56:52$ awk '($NF !~ "^0$") {print}' sos_commands/networking/ethtool_-S_eno16780032 | egrep -v "[u,m,b]cast|LRO pkts rx|[LR,TS]O byte(s)?|[LR,TS]O pkts|pkts linearized" NIC statistics:
Tx Queue#: 1 Tx Queue#: 2 Tx Queue#: 3 Rx Queue#: 1 pkts rx OOB: 11 Rx Queue#: 2 Rx Queue#: 3
g61mps-jhq02-2017-Jul-16-16:59:36$ awk '($NF !~ "^0$") {print}' sos_commands/networking/ethtool_-S_eno16780032 | egrep -v "[u,m,b]cast|LRO pkts rx|[LR,TS]O byte(s)?|[LR,TS]O pkts|pkts linearized" NIC statistics:
Tx Queue#: 1 Tx Queue#: 2 Tx Queue#: 3 Rx Queue#: 1 Rx Queue#: 2 Rx Queue#: 3 pkts rx OOB: 52
g61mps-jhq03-2017-Jul-16-17:00:02$ awk '($NF !~ "^0$") {print}' sos_commands/networking/ethtool_-S_eno16780032 | egrep -v "[u,m,b]cast|LRO pkts rx|[LR,TS]O byte(s)?|[LR,TS]O pkts|pkts linearized" NIC statistics:
Tx Queue#: 1 Tx Queue#: 2 Tx Queue#: 3 Rx Queue#: 1 pkts rx OOB: 45 drv dropped rx total: 29 err: 29 Rx Queue#: 2 Rx Queue#: 3
- One anomaly is that each node reports a high number of rx drops for each net device. The System Activity Reporting (SAR) data shows this is occurring at a very constant and continuous rate. This is almost certainly caused by the NIC receiving some frame type which the kernel does not handle. These are unfortunately recorded as netdev drops. The phenomenon is described in the article below:
ifconfig packet drops reported after upgrading to RHEL7 https://access.redhat.com/solutions/2073223
So we can safely ignore this counter. The fact that the drops are not reflected in the vmxnet3 specific counters seen in ethtool -S assures us of this.
g61mps-jhq01-2017-Jul-16-16:56:52$ cat proc/net/dev Inter-| Receive | Transmit
face |bytes packets errs drop fifo frame compressed multicast|bytes packets errs drop fifo colls carrier compressed lo: 727497676 2498032 0 0 0 0 0 0 727497676 2498032 0 0 0 0 0 0
eno16780032: 216193874050 702019404 0 2658277 0 0 0 87265004 195315249141 549883330 0 0 0 0 0 0
- Moving up the stack to the IP and TCP levels we can review the OS protocol handler stats visible via netstat -s. Here I do not see any obvious sign of trouble on any of the nodes.
- IP fragmentation is occurring. This is normal behaviour when an application sends a datagram which exceeds the MTU (1500). There have been a trivial number of failures due to fragment loss. TCP divides data into MSS-sized segments which should not require IP fragmentation so this is likely UDP traffic.
g61mps-jhq01-2017-Jul-16-16:56:52$ cat sos_commands/networking/netstat_-s Ip:
702329554 total packets received 0 forwarded 0 incoming packets discarded 699283912 incoming packets delivered 550941269 requests sent out 16 dropped because of missing route 5 fragments dropped after timeout 4372810 reassemblies required 1336065 packets reassembled ok 7 packet reassembles failed 618990 fragments received ok 1856970 fragments created
....
- The rate of TCP retransmissions on all these nodes is exceedingly low. This is the strongest sign we have that there is no network level issue on the nodes or in the network infrastructure between them. If there were regular problems with packet loss or high latency in the environment for any reason, that would be reflected in a high rate of TCP retransmissions.
g61mps-jhq01-2017-Jul-16-16:56:52$ cat sos_commands/networking/netstat_-s .... Tcp:
52947 active connections openings 22184312 passive connection openings 26635 failed connection attempts 928 connection resets received 897 connections established 610099637 segments received -> 596182347 segments send out -> 287 segments retransmited 0 bad segments received. 8340540 resets sent
....
- The only glimmer of an issue at the TCP level is there has been the odd socket buffer overflow and at least on node 2, a few listen queue overflows. The numbers here are still very small so it is hard to believe that these account for a persistent failure at the cluster level.
g61mps-jhq01-2017-Jul-16-16:56:52$ egrep "pruned|collapsed|overflowed" sos_commands/networking/netstat_-s
542 packets pruned from receive queue because of socket buffer overrun 4 packets pruned from receive queue 1197 packets collapsed in receive queue due to low socket buffer