diff options
Diffstat (limited to 'docs/troubleshooting')
| -rw-r--r-- | docs/troubleshooting/cpuusage.rst | 112 | ||||
| -rw-r--r-- | docs/troubleshooting/index.rst | 13 | ||||
| -rw-r--r-- | docs/troubleshooting/reportingissues/index.rst | 8 | ||||
| -rw-r--r-- | docs/troubleshooting/reportingissues/reportingissues.rst | 216 |
4 files changed, 349 insertions, 0 deletions
diff --git a/docs/troubleshooting/cpuusage.rst b/docs/troubleshooting/cpuusage.rst new file mode 100644 index 00000000000..b9b8942a3dd --- /dev/null +++ b/docs/troubleshooting/cpuusage.rst @@ -0,0 +1,112 @@ +.. _cpuusage: + +************** +CPU Load/Usage +************** + +There are various commands and tools that can help users see FD.io VPP CPU and memory usage at runtime. + +Linux top/htop +============== + +The Linux top and htop are decent tools to look at FD.io VPP cpu and memory usage, but they will only show +preallocated memory and total CPU usage. These commands can be useful to show which cores VPP is running on. + +This is an example of VPP instance that is running on cores 8 and 9. For this output type **top** and then +type **1** when the tool starts. + +.. code-block:: console + + $ top + + top - 11:04:04 up 35 days, 3:16, 5 users, load average: 2.33, 2.23, 2.16 + Tasks: 435 total, 2 running, 432 sleeping, 1 stopped, 0 zombie + %Cpu0 : 1.0 us, 0.7 sy, 0.0 ni, 98.0 id, 0.0 wa, 0.0 hi, 0.3 si, 0.0 st + %Cpu1 : 2.0 us, 0.3 sy, 0.0 ni, 97.7 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st + %Cpu2 : 0.7 us, 1.0 sy, 0.0 ni, 98.3 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st + %Cpu3 : 1.7 us, 0.7 sy, 0.0 ni, 97.7 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st + %Cpu4 : 2.0 us, 0.7 sy, 0.0 ni, 97.4 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st + %Cpu5 : 3.0 us, 0.3 sy, 0.0 ni, 96.7 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st + %Cpu6 : 2.3 us, 0.7 sy, 0.0 ni, 97.0 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st + %Cpu7 : 2.6 us, 0.3 sy, 0.0 ni, 97.0 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st + %Cpu8 : 96.0 us, 0.3 sy, 0.0 ni, 3.6 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st + %Cpu9 :100.0 us, 0.0 sy, 0.0 ni, 0.0 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st + %Cpu10 : 1.0 us, 0.3 sy, 0.0 ni, 98.7 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st + .... + +VPP Memory Usage +================ + +For details on VPP memory usage you can use the **show memory** command + +This is the example VPP memory usage on 2 cores. + +.. code-block:: console + + # vppctl show memory verbose + Thread 0 vpp_main + 22043 objects, 17878k of 20826k used, 2426k free, 2396k reclaimed, 346k overhead, 1048572k capacity + alloc. from small object cache: 22875 hits 39973 attempts (57.23%) replacements 5143 + alloc. from free-list: 44732 attempts, 26017 hits (58.16%), 528461 considered (per-attempt 11.81) + alloc. from vector-expand: 3430 + allocs: 52324 2027.84 clocks/call + frees: 30280 594.38 clocks/call + Thread 1 vpp_wk_0 + 22043 objects, 17878k of 20826k used, 2427k free, 2396k reclaimed, 346k overhead, 1048572k capacity + alloc. from small object cache: 22881 hits 39984 attempts (57.23%) replacements 5148 + alloc. from free-list: 44736 attempts, 26021 hits (58.17%), 528465 considered (per-attempt 11.81) + alloc. from vector-expand: 3430 + allocs: 52335 2027.54 clocks/call + frees: 30291 594.36 clocks/call + +VPP CPU Load +============ + +To find the VPP CPU load or how busy VPP is use the **show runtime** command. + +With at least one interface in polling mode, the VPP CPU utilization is always 100%. + +A good indicator of CPU load is **"average vectors/node"**. A bigger number means VPP +is more busy but also more efficient. The Maximum value is 255 (unless you change VLIB_FRAME_SIZE in code). +It basically means how many packets are processed in batch. + +If VPP is not loaded it will likely poll so fast that it will just get one or few +packets from the rx queue. This is the case shown below on Thread 1. As load goes up vpp +will have more work to do, so it will poll less frequently, and that will result in more +packets waiting in rx queue. More packets will result in more efficient execution of the +code so number of clock cycles / packet will go down. When "average vectors/node" goes up +close to 255, you will likely start observing rx queue tail drops. + +.. code-block:: console + + # vppctl show run + Thread 0 vpp_main (lcore 8) + Time 6152.9, average vectors/node 0.00, last 128 main loops 0.00 per node 0.00 + vector rates in 0.0000e0, out 0.0000e0, drop 0.0000e0, punt 0.0000e0 + Name State Calls Vectors Suspends Clocks Vectors/Call + acl-plugin-fa-cleaner-process event wait 0 0 1 3.66e4 0.00 + admin-up-down-process event wait 0 0 1 2.54e3 0.00 + .... + --------------- + Thread 1 vpp_wk_0 (lcore 9) + Time 6152.9, average vectors/node 1.00, last 128 main loops 0.00 per node 0.00 + vector rates in 1.3073e2, out 1.3073e2, drop 6.5009e-4, punt 0.0000e0 + Name State Calls Vectors Suspends Clocks Vectors/Call + TenGigabitEthernet86/0/0-outpu active 804395 804395 0 6.17e2 1.00 + TenGigabitEthernet86/0/0-tx active 804395 804395 0 7.29e2 1.00 + arp-input active 2 2 0 3.82e4 1.00 + dpdk-input polling 24239296364 804398 0 1.59e7 0.00 + error-drop active 4 4 0 4.65e3 1.00 + ethernet-input active 2 2 0 1.08e4 1.00 + interface-output active 1 1 0 3.78e3 1.00 + ip4-glean active 1 1 0 6.98e4 1.00 + ip4-icmp-echo-request active 804394 804394 0 5.02e2 1.00 + ip4-icmp-input active 804394 804394 0 4.63e2 1.00 + ip4-input-no-checksum active 804394 804394 0 8.51e2 1.00 + ip4-load-balance active 804394 804394 0 5.46e2 1.00 + ip4-local active 804394 804394 0 5.79e2 1.00 + ip4-lookup active 804394 804394 0 5.71e2 1.00 + ip4-rewrite active 804393 804393 0 5.69e2 1.00 + ip6-input active 2 2 0 5.72e3 1.00 + ip6-not-enabled active 2 2 0 1.56e4 1.00 + unix-epoll-input polling 835722 0 0 3.03e-3 0.00 diff --git a/docs/troubleshooting/index.rst b/docs/troubleshooting/index.rst new file mode 100644 index 00000000000..655eb1192e6 --- /dev/null +++ b/docs/troubleshooting/index.rst @@ -0,0 +1,13 @@ +.. _troubleshooting: + +############### +Troubleshooting +############### + +This chapter describes some of the many techniques used to troubleshoot and diagnose +problem with FD.io VPP implementations. + +.. toctree:: + + reportingissues/index.rst + cpuusage diff --git a/docs/troubleshooting/reportingissues/index.rst b/docs/troubleshooting/reportingissues/index.rst new file mode 100644 index 00000000000..4d954ac8746 --- /dev/null +++ b/docs/troubleshooting/reportingissues/index.rst @@ -0,0 +1,8 @@ +.. _reportingissues: + +How to Report an Issue +====================== + +.. toctree:: + + reportingissues diff --git a/docs/troubleshooting/reportingissues/reportingissues.rst b/docs/troubleshooting/reportingissues/reportingissues.rst new file mode 100644 index 00000000000..7437d8ae0cc --- /dev/null +++ b/docs/troubleshooting/reportingissues/reportingissues.rst @@ -0,0 +1,216 @@ +.. _reportingbugs: + +.. toctree:: + +Reporting Bugs +============== + +Although every situation is different, this page describes how to +collect data which will help make efficient use of everyone's time +when dealing with vpp bugs. + +Before you press the Jira button to create a bug report - or email +vpp-dev@lists.fd.io - please ask yourself whether there's enough +information for someone else to understand and possibly to reproduce +the issue given a reasonable amount of effort. **Unicast emails to +maintainers, committers, and the project PTL are strongly discouraged.** + +A good strategy for clear-cut bugs: file a detailed Jira ticket, and +then send a short description of the issue to vpp-dev@lists.fd.io, +perhaps from the Jira ticket description. It's fine to send email to +vpp-dev@lists.fd.io to ask a few questions **before** filing Jira tickets. + +Data to include in bug reports +============================== + +Image version and operating environment +--------------------------------------- + +Please make sure to include the vpp image version and command-line arguments. + +.. code-block:: console + + $ sudo bash + # vppctl show version verbose cmdline + Version: v18.07-rc0~509-gb9124828 + Compiled by: vppuser + Compile host: vppbuild + Compile date: Fri Jul 13 09:05:37 EDT 2018 + Compile location: /scratch/vpp-showversion + Compiler: GCC 7.3.0 + Current PID: 5211 + Command line arguments: + /scratch/vpp-showversion/build-root/install-vpp_debug-native/vpp/bin/vpp + unix + interactive + +With respect to the operating environment: if misbehavior involving a +specific VM / container / bare-metal environment is involved, please +describe the environment in detail: + +* Linux Distro (e.g. Ubuntu 14.04.3 LTS, CentOS-7, etc.) +* NIC type(s) (ixgbe, i40e, enic, etc. etc.), vhost-user, tuntap +* NUMA configuration if applicable + +Please note the CPU architecture (x86_86, aarch64), and hardware platform. + +When practicable, please report issues against released software, or +unmodified master/latest software. + +"Show" command output +--------------------- + +Every situation is different. If the issue involves a sequence of debug CLI command, please enable CLI command logging, and send the sequence involved. Note that the debug CLI is a developer's tool - **no warranty express or implied** - and that we may choose not to fix debug CLI bugs. + +Please include "show error" [error counter] output. It's often helpful to "clear error", send a bit of traffic, then "show error" particularly when running vpp on a noisy networks. + +Please include ip4 / ip6 / mpls FIB contents ("show ip fib", "show ip6 fib", "show mpls fib", "show mpls tunnel"). + +Please include "show hardware", "show interface", and "show interface address" output + +Here is a consolidated set of commands that are generally useful before/after sending traffic. Before sending traffic. + +.. code-block:: console + + vppctl clear hardware + vppctl clear interface + vppctl clear error + vppctl clear run + +Send some traffic and then issue the following commands. + +.. code-block:: console + + vppctl show version verbose + vppctl show hardware + vppctl show hardware address + vppctl show interface + vppctl show run + vppctl show error + +Here are some protocol specific show commands that may also make +sense. Only include those features which have been configured. + +.. code-block:: console + + vppctl show l2fib + vppctl show bridge-domain + + vppctl show ip fib + vppctl show ip arp + + vppctl show ip6 fib + vppctl show ip6 neighbors + + vppctl show mpls fib + vppctl show mpls tunnel + +Network Topology +---------------- + +Please include a crisp description of the network topology, including +L2 / IP / MPLS / segment-routing addressing details. If you expect +folks to reproduce and debug issues, this is a must. + +At or above a certain level of topological complexity, it becomes +problematic to reproduce the original setup. + +Packet Tracer Output +-------------------- + +If you capture packet tracer output which seems relevant, please include it. + +.. code-block:: console + + vppctl trace add dpdk-input 100 # or similar + +send-traffic + +.. code-block:: console + + vppctl show trace + +Capturing post-mortem data +========================== + +It should go without saying, but anyhow: **please put post-mortem data +in obvious, accessible places.** Time wasted trying to acquire +accounts, credentials, and IP addresses simply delays problem +resolution. + +Please remember to add post-mortem data location information to Jira +tickets. + +Syslog Output +------------- + +The vpp signal handler typically writes a certain amount of data in +/var/log/syslog before exiting. Make sure to check for evidence, e.g +via "grep /usr/bin/vpp /var/log/syslog" or similar. + +Binary API Trace +---------------- + +If the issue involves a sequence of control-plane API messages - even +a very long sequence - please enable control-plane API +tracing. Control-plane API post-mortem traces end up in +/tmp/api_post_mortem.<pid>. + +Please remember to put post-mortem binary api traces in accessible +places. + +These API traces are especially helpful in cases where the vpp engine +is throwing traffic on the floor, e.g. for want of a default route or +similar. + +Make sure to leave the default stanza "... api-trace { on } ... " in +the vpp startup configuration file /etc/vpp/startup.conf, or to +include it in the command line arguments passed by orchestration +software. + +Core Files +---------- + +Production systems, as well as long-running pre-production soak-test +systems, **must** arrange to collect core images. There are various +ways to configure core image capture, including e.g. the Ubuntu +"corekeeper" package. In a pinch, the following very basic sequence +will capture usable vpp core files in /tmp/dumps. + +.. code-block:: console + + # mkdir -p /tmp/dumps + # sysctl -w debug.exception-trace=1 + # sysctl -w kernel.core_pattern="/tmp/dumps/%e-%t" + # ulimit -c unlimited + # echo 2 > /proc/sys/fs/suid_dumpable + +Vpp core files often appear enormous. Gzip typically compresses them +to manageable sizes. A multi-GByte corefile often compresses to 10-20 +Mbytes. + +Please remember to put compressed core files in accessible places. + +Make sure to leave the default stanza "... unix { ... full-coredump +... } ... " in the vpp startup configuration file +/etc/vpp/startup.conf, or to include it in the command line arguments +passed by orchestration software. + +Core files from private, modified images are discouraged. If it's +necessary to go that route, please copy the **exact** Debian +packages (or RPMs) corresponding to the core file to the same public +place as the core file. In particular. + +* vpp_<version>_<arch>.deb # the vpp executable +* vpp-dbg_<version>_<arch>.deb # debug symbols +* vpp-dev_<version>_<arch>.deb # development package +* vpp-lib_<version>_<arch>.deb # shared libraries +* vpp-plugins_<version>_<arch>.deb # plugins + +Please include the full commit-ID the Jira ticket. + +If we go through the setup process only to discover that the image and +core files don't match, it will simply delay resolution of the +issue. And it will annoy the heck out of the engineer who just wasted +their time. Exact means **exact**, not "oh, gee, I added a few lines +of debug scaffolding since then..." |