2
0
mirror of https://github.com/openvswitch/ovs synced 2025-08-22 01:51:26 +00:00

documentation: Document ovs-flowviz.

Add a man page for ovs-flowviz as well as a topic page with some more
detailed examples.

Acked-by: Eelco Chaudron <echaudro@redhat.com>
Signed-off-by: Adrian Moreno <amorenoz@redhat.com>
Signed-off-by: Ilya Maximets <i.maximets@ovn.org>
This commit is contained in:
Adrian Moreno 2024-09-25 12:52:14 +02:00 committed by Ilya Maximets
parent 4214bf4b28
commit e532f937f5
9 changed files with 868 additions and 1 deletions

View File

@ -45,7 +45,7 @@ DOC_SOURCE = \
Documentation/topics/fuzzing/ovs-fuzzing-infrastructure.rst \
Documentation/topics/fuzzing/ovs-fuzzers.rst \
Documentation/topics/fuzzing/security-analysis-of-ovs-fuzzers.rst \
Documentation/topics/testing.rst \
Documentation/topics/flow-visualization.rst \
Documentation/topics/integration.rst \
Documentation/topics/language-bindings.rst \
Documentation/topics/networking-namespaces.rst \
@ -55,6 +55,7 @@ DOC_SOURCE = \
Documentation/topics/ovsdb-replication.rst \
Documentation/topics/porting.rst \
Documentation/topics/record-replay.rst \
Documentation/topics/testing.rst \
Documentation/topics/tracing.rst \
Documentation/topics/usdt-probes.rst \
Documentation/topics/userspace-checksum-offloading.rst \
@ -162,6 +163,7 @@ RST_MANPAGES = \
ovs-actions.7.rst \
ovs-appctl.8.rst \
ovs-ctl.8.rst \
ovs-flowviz.8.rst \
ovs-l3ping.8.rst \
ovs-parse-backtrace.8.rst \
ovs-pki.8.rst \

View File

@ -112,6 +112,8 @@ html_static_path = ['_static']
# Define the canonical URL for our domain configured on Read the Docs.
html_baseurl = os.environ.get("READTHEDOCS_CANONICAL_URL", "")
option_emphasise_placeholders = True
# Tell Jinja2 templates the build is running on Read the Docs.
html_context = {}
if os.environ.get("READTHEDOCS", "") == "True":
@ -128,6 +130,8 @@ _man_pages = [
u'utility for configuring running Open vSwitch daemons'),
('ovs-ctl.8',
u'OVS startup helper script'),
('ovs-flowviz.8',
u'utility for visualizing OpenFlow and datapath flows'),
('ovs-l3ping.8',
u'check network deployment for L3 tunneling problems'),
('ovs-parse-backtrace.8',

View File

@ -42,6 +42,7 @@ time:
ovs-actions.7
ovs-appctl.8
ovs-ctl.8
ovs-flowviz.8
ovs-l3ping.8
ovs-pki.8
ovs-sim.1

View File

@ -0,0 +1,540 @@
..
Licensed under the Apache License, Version 2.0 (the "License"); you may
not use this file except in compliance with the License. You may obtain
a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
License for the specific language governing permissions and limitations
under the License.
Convention for heading levels in Open vSwitch documentation:
======= Heading 0 (reserved for the title in a document)
------- Heading 1
~~~~~~~ Heading 2
+++++++ Heading 3
''''''' Heading 4
Avoid deeper levels because they do not render well.
===========
ovs-flowviz
===========
Synopsis
========
``ovs-flowviz``
[``-i`` [*alias*,]\ *file* | ``--input`` [*alias*,]\ *file*]
[``-c`` *file* | ``--config`` *file*]
[``-f`` *filter* | ``--filter`` *filter*]
[``-h`` *filter* | ``--highlight`` *filter*]
[``--style`` *style*]
*flow-type* *format* [*args*...]
``ovs-flowviz --help``
Description
===========
``ovs-flowviz`` helps visualize OpenFlow and datapath flow dumps in different
formats in order to make them more easily understood.
``ovs-flowviz`` reads flows from ``stdin`` or from a *file* specified by the
``--input`` option, filters them, highlights them, and finally outputs
them in one of the predefined *format*\ s.
Options
=======
.. program: ovs-flowviz
.. option:: -h, --help
Print a brief help message to the console.
.. option:: -i [<alias>,]<file>, --input [<alias>,]<file>
File to read flows from. If not provided, ``ovs-flowviz``
will read flows from stdin.
This option can be specified multiple times.
The file path can prepended by an alias that will be shown in the output.
For example: ``--input node1,/path/to/file1 --input node2,/path/to/file2``
.. option:: -c <file>, --config <file>
Style configuration file to use, overriding the default one.
Styles defined in the style configuration file can be selected using
the ``--style`` option.
For more details on the style configuration file, see the
`Style Configuration File`_ section below.
.. option:: -f <filter>, --filter <filter>
Flow filter expression. Only those flows matching the expression will be
shown (although some formats implement filtering differently, see the
`Datapath tree format`_ section below).
The filtering syntax is detailed in `Filtering Syntax`_.
.. option:: -h <filter>, --highlight <filter>
Highlight the flows that match the provided *filter* expression.
The filtering syntax is detailed in `Filtering Syntax`_.
.. option:: --style <style>
Style. The selected *style* must be defined in the style configuration file.
.. option:: flow-type
**openflow** or **datapath**.
.. option:: format
See the `Supported formats`_ section.
Supported formats
=================
``ovs-flowviz`` supports several visualization formats for both OpenFlow and
datapath flows:
.. list-table::
:widths: 20 10 70
:align: center
:header-rows: 1
* - Flow Type
- Format
- Description
* - Both
- console
- Prints the flows in a configurable, colorful style in the console.
* - Both
- json
- Prints the flows in JSON format.
* - Both
- html
- Prints the flows in an HTML list.
* - OpenFlow
- cookie
- Prints the flows in the console sorted by cookie.
* - OpenFlow
- logic
- Prints the logical structure of flows in the console.
* - Datapath
- tree
- Prints the flows as a tree structure arranged by ``recirc_id`` and
``in_port``.
* - Datapath
- graph
- Prints a graphviz graph of the flows arranged by ``recirc_id`` and
``in_port``.
Console format
~~~~~~~~~~~~~~
The ``console`` format works for both OpenFlow and datapath flow types, and
prints flows in the terminal using the style determined by the ``--style``
option.
Arguments:
.. option:: -h, --heat-map
Color of the packet and byte counters to reflect their relative size.
The color gradient goes through the following colors:
blue (coldest, lowest), cyan, green, yellow, red (hottest, highest)
Note filtering is applied before the range is calculated.
JSON format
~~~~~~~~~~~
The ``json`` format works for both OpenFlow and datapath flow types, and prints
flows in JSON format. See the `JSON Syntax`_ section for more details.
HTML format
~~~~~~~~~~~
The ``html`` format works for both OpenFlow and datapath flows, and prints
flows in an HTML table that offers some basic interactivity. OpenFlow flows
are sorted in tables and datapath flows are arranged in flow trees
(see `Datapath tree format`_ for more details).
Styles defined via Style Configuration File and selected via ``--style`` option
also apply to the ``html`` format.
OpenFlow cookie format
~~~~~~~~~~~~~~~~~~~~~~
The OpenFlow ``cookie`` format is similar to the ``console`` format but
instead of arranging the flows by table, it arranges the flows by cookie.
OpenFlow logic format
~~~~~~~~~~~~~~~~~~~~~
The OpenFlow ``logic`` format helps visualize the logic structure of OpenFlow
pipelines by arranging flows into *logical blocks*.
A logical block is a set of flows that have:
* Same ``priority``.
* Match on the same fields (regardless of the match value and mask).
* Execute the same actions (regardless of the actions' arguments,
except for resubmit and output).
* Optionally, the ``cookie`` can be included as part of the logical flow.
Arguments:
.. option:: -s, --show-flows
Show all the flows under each logical block.
.. option:: -d, --ovn-detrace
Use ovn-detrace.py script to extract cookie information (implies '-c').
.. option:: -c, --cookie
Consider the cookie in the logical block.
.. option:: --ovn-detrace-path <path>
Use an alternative path to search for ovn_detrace.py.
.. option:: --ovnnb-db <conn>
OVN NB database connection method (implies '-d').
Default: "unix:/var/run/ovn/ovnnb_db.sock".
.. option:: --ovnsb-db <conn>
OVN SB database connection method (implies '-d').
Default: "unix:/var/run/ovn/ovnsb_db.sock".
.. option:: --o <filter>, --ovn-filter <filter>
Specify the filter to be run on the ovn-detrace information.
Syntax: python regular expression
(See https://docs.python.org/3/library/re.html).
.. option:: -h, --heat-map
Change the color of the packet and byte counters to reflect their relative
size. The color gradient goes through the following colors:
blue (coldest, lowest), cyan, green, yellow, red (hottest, highest)
Note filtering is applied before the range is calculated.
Datapath tree format
~~~~~~~~~~~~~~~~~~~~
The datapath ``tree`` format arranges datapath flows in a hierarchical tree.
The tree is comprised of blocks with the same ``recirc_id`` and ``in_port``.
Within those blocks, flows with the same action are combined. And matches
which are the same are omitted to reduce the visual noise.
When a flow's actions includes the ``recirc()`` action with a specific
``recirc_id``, flows matching on that ``recirc_id`` and the same ``in_port``
are listed below. This is done recursively for all actions.
The result is a hierarchical representation that shows how actions are related
to each other via recirculation. Note that flows with a specific non-zero
``recirc_id`` are listed below each group of flows that have a corresponding
``recirc()`` action. Therefore, the output contains duplicated flows and can be
verbose.
Filtering works in a slightly different way for datapath flow trees.
Unlike other formats where a filter simply removes non-matching flows,
the output of a filtered datapath flow tree will show full sub-trees
that contain at least one flow that satisfies the filter.
The ``html`` format prints this same tree as an interactive HTML table and
the ``graph`` format shows the same tree as a graphviz graph.
Datapath graph format
~~~~~~~~~~~~~~~~~~~~~
The datapath ``graph`` generates a graphviz visual representation of the
same tree-like flow hierarchy that the ``tree`` format prints.
Arguments:
.. option:: -h, --html
Print the graphviz format as an svg image alongside an interactive HTML
table of flows.
JSON Syntax
===========
Printing a single-file OpenFlow or datapath dump without PMD thread blocks in
``json`` format results in a list of JSON objects, each representing a flow.
This list can be found inside one or more levels of JSON dictionaries
if multiple files are processed (filename used as key) or if PMD thread blocks
are found in datapath flows (name of the thread used as key).
Each flow object includes the following keys:
**orig**
Original flow string.
**info**
Object with the flow information such as: cookie, duration, table,
n_packets, n_bytes, etc.
**match**
Object with the flow match.
For each match, the object contains a key-value where the key is the name
of the match as defined in ``ovs-fields(7)`` and ``ovs-ofctl(8)``, and the
value represents the match value. The way each value is represented depends
on its type. See `Value representation`_.
**actions**
List of action objects.
Each action is represented by an JSON object that has one key and one value.
The key corresponds to the action name. The value represents the arguments
of the key. See `Action representation`_.
**ufid**
The UFID (datapath flows only).
Value representation
~~~~~~~~~~~~~~~~~~~~
Values are represented differently depending on their type:
* Flags: The value of flags is ``true``.
* Decimal / Hexadecimal: Represented by their integer value.
If they support masking, represented by a dictionary with two keys:
``value`` contains the field value and ``mask`` contains the mask.
Both are integers.
* Ethernet: Represented by a string: ``{address}[/{mask}]``
* IPv4 / IPv6: Represented by a string ``{address}[/{mask}]``
* Registers: Represented by a dictionary with three keys:
``field``` contains the field value (string), ``start``, and ``end``
represent the first and last bit of the register value.
For example, the register
::
NXM_NX_REG10[0..15]
is represented as
::
{
"field": "NXM_NX_REG10",
"start": 0,
"end": 15
},
Action representation
~~~~~~~~~~~~~~~~~~~~~
Actions are generally represented by an object that has a single key and
value. The key is the action name as defined ``ovs-actions(7)``.
The value of actions that have no arguments (such as ``drop``) is
(boolean) ``true``.
The value of actions that have a list of arguments (e.g:
``resubmit([port],[table],[ct])``) is an object that has the name of the
argument as key. The argument names for each action is defined in
ovs-actions. For example, the action
::
resubmit(,10)
is represented as
::
{
"resubmit": {
"port": "",
"table": 10
}
}
The value of actions that have a key-word list as arguments
(e.g: ``ct([argument])``) is an object whose keys correspond to the keys
defined in ``ovs-actions(7)``. The way values are represented depends
on the type of the argument.
For example, the action
::
ct(table=14,zone=NXM_NX_REG12[0..15],nat)
is represented as
::
{
"ct": {
"table": 14,
"zone": {
"field": "NXM_NX_REG12",
"start": 0,
"end": 15
},
"nat": true
}
}
Style Configuration File
========================
The style configuration file is selected via the ``--config`` option
and has INI syntax. It can define any number of styles to be used by both
``console`` and ``html`` formats. Once defined in the configuration file,
formats are selected using the ``--style`` option.
INI sections are used to define styles, ``[styles.mystyle]`` defines a style
called `mystle`. Within a section styles can be defined as:
::
[FORMAT].[PORTION].[SELECTOR].[ELEMENT] = [VALUE]
**FORMAT**
Either ``console`` or ``html``
**PORTION**
Part of the key-value the style applies to:
``key`` to indicate the key part of a key-value, ``value`` to indicate
the value part of a key-value, ``flag`` to indicate a single flag
or ``delim`` to indicate delimiters such as parentheses, brackets, etc.
**SELECTOR**
Select the key-value the style applies to:
``highlighted`` to indicate highlighted key-values, ``type.<type>``
to indicate certain types such as ``IPAddress`` or ``EthMask`` or
``<keyname>`` to select a particular key name.
**ELEMENT**
Select the style element to modify:
**color** or **underline** (only for ``console`` format).
**VALUE**
Ether a color hex, other color names defined in the rich python
library (https://rich.readthedocs.io/en/stable/appendix/colors.html) or
**true** if the element is ``underline``.
A default configuration file is shipped with ``ovs-flowviz`` and its path is
printed in the ``--help`` output. A detailed description of the syntax
alongside some examples are available there.
Filtering syntax
================
``ovs-flowviz`` provides rich highlighting and filtering. The special command
``ovs-flowviz filter`` dumps the filtering syntax:
::
$ ovs-flowviz filter
Filter Syntax
*************
[! | not ] {key}[[.subkey]...] [OPERATOR] {value})] [LOGICAL OPERATOR] ...
Comparison operators:
= equality
< less than
> more than
~= masking (valid for IP and Ethernet fields)
Logical operators:
!{expr}: NOT
{expr} && {expr}: AND
{expr} || {expr}: OR
Matches and flow metadata:
To compare against a match or info field, use the field directly, e.g:
priority=100
n_bytes>10
Use simple keywords for flags:
tcp and ip_src=192.168.1.1
Actions:
Actions values might be dictionaries, use subkeys to access individual
values, e.g:
output.port=3
Use simple keywords for flags
drop
Examples of valid filters:
nw_addr~=192.168.1.1 && (tcp_dst=80 || tcp_dst=443)
arp=true && !arp_tsa=192.168.1.1
n_bytes>0 && drop=true
Example expressions:
::
n_bytes > 0 and drop
nw_src~=192.168.1.1 or arp.tsa=192.168.1.1
! tcp && output.port=2
Examples
========
Print OpenFlow flows sorted by cookie adding OVN data to each one:
::
$ ovs-flowviz -i flows.txt openflow cookie --ovn-detrace
Print OpenFlow logical structure, showing the flows and heat-map:
::
$ ovs-flowviz -i flows.txt openflow logic --show-flows --heat-map
Display OpenFlow flows in HTML format with "light" style and highlight drops:
::
$ ovs-flowviz -i flows.txt --style "light" --highlight "n_packets > 0 and drop" openflow html > flows.html
Display the datapath flows in an interactive graphviz + HTML view:
::
$ ovs-flowviz -i flows.txt datapath graph --html > flows.html
Display the datapath flow trees that lead to packets being sent to port 10:
::
$ ovs-flowviz -i flows.txt --filter "output.port=10" datapath tree

View File

@ -0,0 +1,313 @@
..
Licensed under the Apache License, Version 2.0 (the "License"); you may
not use this file except in compliance with the License. You may obtain
a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
License for the specific language governing permissions and limitations
under the License.
Convention for heading levels in Open vSwitch documentation:
======= Heading 0 (reserved for the title in a document)
------- Heading 1
~~~~~~~ Heading 2
+++++++ Heading 3
''''''' Heading 4
Avoid deeper levels because they do not render well.
==================================
Visualizing flows with ovs-flowviz
==================================
When troubleshooting networking issues with OVS, it's common to end up looking
at OpenFlow or datapath flow dumps. These dumps tend to be quite dense and
difficult to reason about.
``ovs-flowviz`` is a utility script that helps visualizing OpenFlow and
datapath flows to make it easier to understand what is going on.
The `ovs-flowviz(8)`_ manpage describes its basic usage. In this document a few
of its advanced visualization formats will be expanded.
Installing ovs-flowviz
----------------------
``ovs-flowviz`` is part of the openvswitch python package but its
extra dependencies have to be installed explicitly by running:
::
$ pip install openvswitch[flowviz]
Or, if you are working with the OVS tree:
::
$ cd python && pip install .[flowviz]
Visualizing OpenFlow logical block
----------------------------------
When controllers such as OVN write OpenFlow flows, they typically organize
flows in functional blocks. These blocks can expand to multiple flows that
"look similar", in the sense that they match on the same fields and have
similar actions.
However, looking at a flow dump the number of flows can make it difficult
to perceive this logical functionality that the controller is trying to
implement using OpenFlow.
``ovs-flowviz openflow logic`` visualization can be used to understand an OVN
flow dump a bit better.
On a particular flow dump table 0 contains 23 flows:
::
$ grep -c "table=0" flows.txt
23
Looking at the first few lines, the amount of information can be
overwhelming and difficult our analysis:
::
$ head flows.txt
cookie=0xf76b4b20, duration=765.107s, table=0, n_packets=0, n_bytes=0, priority=180,vlan_tci=0x0000/0x1000 actions=conjunction(100,2/2)
cookie=0xf76b4b20, duration=765.107s, table=0, n_packets=0, n_bytes=0, priority=180,conj_id=100,in_port="patch-br-int-to",vlan_tci=0x0000/0x1000 actions=load:0xa->NXM_NX_REG13[],load:0xc->NXM_NX_REG11[],load:0xb->NXM_NX_REG12[],load:0xb->OXM_OF_METADATA[],load:0x1->NXM_NX_REG14[],mod_dl_src:02:42:ac:12:00:03,resubmit(,8)
cookie=0x0, duration=765.388s, table=0, n_packets=0, n_bytes=0, priority=100,in_port="ovn-6bb3b3-0" actions=move:NXM_NX_TUN_ID[0..23]->OXM_OF_METADATA[0..23],move:NXM_NX_TUN_METADATA0[16..30]->NXM_NX_REG14[0..14],move:NXM_NX_TUN_METADATA0[0..15]->NXM_NX_REG15[0..15],resubmit(,40)
cookie=0x0, duration=765.388s, table=0, n_packets=0, n_bytes=0, priority=100,in_port="ovn-a6ff98-0" actions=move:NXM_NX_TUN_ID[0..23]->OXM_OF_METADATA[0..23],move:NXM_NX_TUN_METADATA0[16..30]->NXM_NX_REG14[0..14],move:NXM_NX_TUN_METADATA0[0..15]->NXM_NX_REG15[0..15],resubmit(,40)
cookie=0xf2ca6195, duration=765.107s, table=0, n_packets=6, n_bytes=636, priority=100,in_port="ovn-k8s-mp0" actions=load:0x1->NXM_NX_REG13[],load:0x2->NXM_NX_REG11[],load:0x7->NXM_NX_REG12[],load:0x4->OXM_OF_METADATA[],load:0x2->NXM_NX_REG14[],resubmit(,8)
cookie=0x236e941d, duration=408.874s, table=0, n_packets=11, n_bytes=846, priority=100,in_port=aceac9829941d11 actions=load:0x11->NXM_NX_REG13[],load:0x2->NXM_NX_REG11[],load:0x7->NXM_NX_REG12[],load:0x4->OXM_OF_METADATA[],load:0x3->NXM_NX_REG14[],resubmit(,8)
cookie=0x3facf689, duration=405.581s, table=0, n_packets=11, n_bytes=846, priority=100,in_port="363ba22029cd92b" actions=load:0x12->NXM_NX_REG13[],load:0x2->NXM_NX_REG11[],load:0x7->NXM_NX_REG12[],load:0x4->OXM_OF_METADATA[],load:0x4->NXM_NX_REG14[],resubmit(,8)
cookie=0xe7c8c4bb, duration=405.570s, table=0, n_packets=11, n_bytes=846, priority=100,in_port="6a62cde0d50ef44" actions=load:0x13->NXM_NX_REG13[],load:0x2->NXM_NX_REG11[],load:0x7->NXM_NX_REG12[],load:0x4->OXM_OF_METADATA[],load:0x5->NXM_NX_REG14[],resubmit(,8)
cookie=0x99a0ffc1, duration=59.391s, table=0, n_packets=8, n_bytes=636, priority=100,in_port="5ff3bfaaa4eb622" actions=load:0x14->NXM_NX_REG13[],load:0x2->NXM_NX_REG11[],load:0x7->NXM_NX_REG12[],load:0x4->OXM_OF_METADATA[],load:0x6->NXM_NX_REG14[],resubmit(,8)
cookie=0xe1b5c263, duration=59.365s, table=0, n_packets=8, n_bytes=636, priority=100,in_port="8d9e0bc76347e59" actions=load:0x15->NXM_NX_REG13[],load:0x2->NXM_NX_REG11[],load:0x7->NXM_NX_REG12[],load:0x4->OXM_OF_METADATA[],load:0x7->NXM_NX_REG14[],resubmit(,8)
However, table 0 can be better understood by looking at its logical
representation:
::
$ ovs-flowviz -i flows.txt -f "table=0" openflow logic
Ofproto Flows (logical)
└── ** TABLE 0 **
├── priority=180 priority,vlan_tci ---> conjunction ( x 1 )
├── priority=180 priority,conj_id,in_port,vlan_tci ---> load,load,load,load,load,mod_dl_src resubmit(,8), ( x 1 )
├── priority=100 priority,in_port ---> move,move,move resubmit(,40), ( x 2 )
├── priority=100 priority,in_port ---> load,load,load,load,load resubmit(,8), ( x 16 )
├── priority=100 priority,in_port,vlan_tci ---> load,load,load,load,load resubmit(,8), ( x 1 )
├── priority=100 priority,in_port,dl_vlan ---> strip_vlan,load,load,load,load,load resubmit(,8), ( x 1 )
└── priority=0 priority ---> drop, ( x 1 )
In only a few logical blocks, there is a good overview of what this table is
doing. It looks like it's adding metadata based on input ports and vlan
IDs and mainly sending traffic to table 8.
A possible next step might be to look at table 8, and in this case, filter out
the flows that have not been hit by actual traffic.
This is quite easy to do with the arithmetic filtering expressions:
::
$ ovs-flowviz -i flows.txt -f "table=8 and n_packets>0" openflow logic
Ofproto Flows (logical)
└── ** TABLE 8 **
├── priority=50 priority,reg14,metadata,dl_dst ---> load resubmit(,9), ( x 3 )
└── priority=50 priority,metadata ---> load,move resubmit(,73),resubmit(,9), ( x 2 )
At this point, understanding the output might be difficult without relating it
to the matadata OVN stored in the previous table. This is where
``ovs-flowviz``'s OVN integration is useful:
::
$ export OVN_NB_DB=tcp:172.18.0.4:6641
$ export OVN_SB_DB=tcp:172.18.0.4:6642
$ ovs-flowviz -i flows.txt -f "table=8 and n_packets>0" openflow logic --ovn-detrace
Ofproto Flows (logical)
└── ** TABLE 8 **
├── cookie=0xe10c34ee priority=50 priority,reg14,metadata,dl_dst ---> load resubmit(,9), ( x 1 )
│ └── OVN Info
│ ├── * Logical datapaths:
│ ├── * "ovn_cluster_router" (366e1c41-0f3d-4420-b796-10692b64e3e4)
│ ├── * Logical flow: table=0 (lr_in_admission), priority=50, match=(eth.mcast && inport == "rtos-ovn-worker2), actions=(xreg0[0..47] = 0a:58:0a:f4:01:01; next;)
│ └── * Logical Router Port: rtos-ovn-worker2 mac 0a:58:0a:f4:01:01 networks ['10.244.1.1/24'] ipv6_ra_configs {}
├── cookie=0x11e1adbc priority=50 priority,reg14,metadata,dl_dst ---> load resubmit(,9), ( x 1 )
│ └── OVN Info
│ ├── * Logical datapaths:
│ ├── * "GR_ovn-worker2" (c07f8387-6479-4e81-9304-9f8e54f81c56)
│ ├── * Logical flow: table=0 (lr_in_admission), priority=50, match=(eth.mcast && inport == "rtoe-GR_ovn-worker2), actions=(xreg0[0..47] = 02:42:ac:12:00:03; next;)
│ └── * Logical Router Port: rtoe-GR_ovn-worker2 mac 02:42:ac:12:00:03 networks ['172.18.0.3/16'] ipv6_ra_configs {}
├── cookie=0xf42133f priority=50 priority,reg14,metadata,dl_dst ---> load resubmit(,9), ( x 1 )
│ └── OVN Info
│ ├── * Logical datapaths:
│ ├── * "GR_ovn-worker2" (c07f8387-6479-4e81-9304-9f8e54f81c56)
│ ├── * Logical flow: table=0 (lr_in_admission), priority=50, match=(eth.dst == 02:42:ac:12:00:03 && inport == "rtoe-GR_ovn-worker2), actions=(xreg0[0..47] = 02:42:ac:12:00:03; next;)
│ └── * Logical Router Port: rtoe-GR_ovn-worker2 mac 02:42:ac:12:00:03 networks ['172.18.0.3/16'] ipv6_ra_configs {}
└── cookie=0x43a0327 priority=50 priority,metadata ---> load,move resubmit(,73),resubmit(,9), ( x 2 )
└── OVN Info
├── * Logical datapaths:
├── * "ovn-worker" (24280d0b-fee0-4f8e-ba4f-036a9b9af921)
├── * "ovn-control-plane" (3262a782-8961-416b-805e-08233e8fda72)
├── * "ext_ovn-worker2" (3f88dcd2-c56d-478f-a3b1-c7aee2efe967)
├── * "ext_ovn-worker" (5facbaf0-485d-4cf5-8940-eff9678ef7bb)
├── * "ext_ovn-control-plane" (8b0aecb6-b05a-48a7-ad09-72524bb91d40)
├── * "join" (e2dc230e-2f2a-4b93-93fa-0fe495163514)
├── * "ovn-worker2" (f7709fbf-d728-4cff-9b9b-150461cc75d2)
└── * Logical flow: table=0 (ls_in_check_port_sec), priority=50, match=(1), actions=(reg0[15] = check_in_port_sec(); next;)
``ovs-flowviz`` has automatically added the `cookie` to the logical block key
so more blocks have been printed. In exchange, it has looked up each cookie on
the running OVN databases and inserted the known information on each
block.
The logical flow that generated each OpenFlow flow and the logical datapath
it belongs to are now printed, making OVN's pipeline clearer.
Visualizing datapath flow trees
-------------------------------
Another typical usecase that can lead to eyestrain is understanding datapath
conntrack recirculations.
OVS makes heavy use of connection tracking and the ``recirc()`` action
to build complex datapaths. Typically, OVS will insert a flow that,
when matched, will send the packet through conntrack (using the ``ct`` action)
and recirculate it with a particular recirculation id (``recirc_id``). Then,
flows matching on that ``recirc_id`` will be matched and further process the
packet. This can happen more than once for a given packet.
This sequential set of events is, however, difficult to visualize when you
look at a datapath flow dump. Flows are unordered so recirculations need to
be followed manually (typically, with heavy use of "grep").
For this use-case, ``ovs-flowviz datapath tree`` format can be extremely
useful. It builds a hierarchical tree based on the ``recirc_id``, ``in_port``
and ``recirc()`` actions.
Furthermore, it is common to end up with multiple flows that have the same
list of actions. An example of this is a number flows that perform mac/vlan
checks for a given port and send the traffic though the same conntrack zone.
In order to better visualize this and reduce the amount of duplicated flows
that are printed in this view, these flows are combined into a block, and the
match keys that are equal for all flows are removed.
For example:
::
Datapath Flows (logical)
└── ╭────────────────────────────────╮
│ [recirc_id(0x0) in_port(eth0)] │
╰────────────────────────────────╯
└── ╭───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ recirc_id(0),dp_hash(0/0),skb_priority(0/0),in_port(eth0),skb_mark(0/0),ct_state(0/0),ct_zone(0/0),ct_mark(0/0),ct_label(0/0),eth(src=0a:58:0a:84:00:07,dst=22:a1:5d:dc:95:50),eth_type(0x0800),ipv4(src=10.132.0.7,dst=1 │
│ 0.128.0.0/255.128.0.0,proto=6,tos=0/0,ttl=0/0,frag=no),tcp(src=0/0,dst=0/0),tcp_flags(0/0), packets:4924, bytes:468961, │
│ recirc_id(0),dp_hash(...),skb_priority(...),in_port(eth0),skb_mark(...),ct_state(...),ct_zone(...),ct_mark(...),ct_label(...),eth(src=0a:58:0a:84:00:07,dst=0a:58:0a:84:00:01),eth_type(......),ipv4(src=10.132.0.7,dst=1 │
│ 0.0.0.0/255.255.128.0,proto=17,tos=0/0,ttl=0/0,frag=no),udp(src=32768/0x8000,dst=0/0), packets:711, bytes:114236, │
│ recirc_id(0),dp_hash(...),skb_priority(...),in_port(eth0),skb_mark(...),ct_state(...),ct_zone(...),ct_mark(...),ct_label(...),eth(src=0a:58:0a:84:00:07,dst=0a:58:0a:84:00:14),eth_type(......),ipv4(src=10.132.0.7,dst=1 │
│ 0.128.0.0/255.128.0.0,proto=17,tos=0/0,ttl=0/0,frag=no),udp(src=4096/0xf000,dst=0/0), packets:140, bytes:114660, │
│ recirc_id(0),dp_hash(...),skb_priority(...),in_port(eth0),skb_mark(...),ct_state(...),ct_zone(...),ct_mark(...),ct_label(...),eth(src=0a:58:0a:84:00:07,dst=0a:58:0a:84:00:22),eth_type(......),ipv4(src=10.132.0.7,dst=1 │
│ 0.128.0.0/255.128.0.0,proto=6,tos=0/0,ttl=0/0,frag=no),tcp(src=0/0,dst=0/0),tcp_flags(0/0), packets:1, bytes:66, │
│ recirc_id(0),dp_hash(...),skb_priority(...),in_port(eth0),skb_mark(...),ct_state(...),ct_zone(...),ct_mark(...),ct_label(...),eth(src=0a:58:0a:84:00:07,dst=0a:58:0a:84:00:09),eth_type(......),ipv4(src=10.132.0.7,dst=1 │
│ 0.128.0.0/255.128.0.0,proto=17,tos=0/0,ttl=0/0,frag=no),udp(src=4096/0xf000,dst=0/0), packets:0, bytes:0, │
╰───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
└── ╭───────────────────────────────────────╮
│ actions: ct(zone=32,nat),recirc(0xc1) │
╰───────────────────────────────────────╯
└── ╭─────────────────────────────────╮
│ [recirc_id(0xc1) in_port(eth0)] │
╰─────────────────────────────────╯
├── ╭───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ │ recirc_id(0xc1),dp_hash(0/0),skb_priority(0/0),in_port(eth0),skb_mark(0/0),ct_state(0x2a/0x3f),ct_zone(0/0),ct_mark(0/0xf),ct_label(0/0),eth(src=0a:58:0a:84:00:07,dst=22:a1:5d:dc:95:50),eth_type(0x0800),ip │
│ │ v4(src=0.0.0.0/0.0.0.0,dst=0.0.0.0/0.0.0.0,proto=6,tos=0/0,ttl=0/0,frag=no),tcp(src=0/0,dst=0/0),tcp_flags(0/0), packets:4924, bytes:468961, │
│ ╰───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
│ └── ╭───────────────────────────────────────╮
│ │ actions: ct(zone=14,nat),recirc(0xc2) │
│ ╰───────────────────────────────────────╯
│ └── ╭─────────────────────────────────╮
│ │ [recirc_id(0xc2) in_port(eth0)] │
│ ╰─────────────────────────────────╯
│ └── ╭───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ │ recirc_id(0xc2),dp_hash(0/0),skb_priority(0/0),in_port(eth0),skb_mark(0/0),ct_state(0x2a/0x3f),ct_zone(0/0),ct_mark(0/0x1),ct_label(0/0),eth(src=00:00:00:00:00:00/00:00:00:00:00:00,dst=00:00:00 │
│ │ :00:00:00/01:00:00:00:00:00),eth_type(0x0800),ipv4(src=0.0.0.0/0.0.0.0,dst=0.0.0.0/0.0.0.0,proto=0/0,tos=0/0,ttl=0/0,frag=no), packets:4924, bytes:468961, │
│ ╰───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
│ └── ╭──────────────────────╮
│ │ actions: ovn-k8s-mp0 │
│ ╰──────────────────────╯
├── ╭───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ │ recirc_id(0xc1),dp_hash(0/0),skb_priority(0/0),in_port(eth0),skb_mark(0/0),ct_state(0x2a/0x3f),ct_zone(0/0),ct_mark(0/0xf),ct_label(0/0),eth(src=0a:58:0a:84:00:07,dst=0a:58:0a:84:00:14),eth_type(0x0800),ip │
│ │ v4(src=0.0.0.0/0.0.0.0,dst=0.0.0.0/0.0.0.0,proto=17,tos=0/0,ttl=0/0,frag=no),udp(src=4096/0xf000,dst=0/0), packets:140, bytes:114660 │
│ ╰───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
The above shows a part of a bigger tree with an initial block of flows
at ``recirc_id(0)`` which match on different destination Ethernet
addresses and protocols, and send traffic through conntrack (zone 32).
Then some additional flows at ``recirc_id(0xc1)`` process each
connection independently. One of them, shown in the example, sends packets
through conntrack zone 14, and after another recirculation the packet is
ultimately sent through a port.
This is a truly complex multi-zone conntrack pipeline that is now significantly
clearer thanks to this visualization.
Also note, the flows in the block are conveniently sorted by sent packets.
This example shows only a single "subtree". Even though the combination of
flows with the same action helps, if we use this command to display a large
dump, the output can be verbose. There are two, combinable, mechanisms that
can help.
Plotting datapath trees
~~~~~~~~~~~~~~~~~~~~~~~
By using the ``ovs-flowviz datapath html`` format, long datapath trees can
be displayed in an interactive HTML table. The resulting web page allows
subtrees to be expanded and collapsed, allowing focus on the desired
information.
The ``ovs-flowviz datapath graph`` format generates a graphviz
graph definition where blocks of flows with the same ``recirc_id`` match
are arranged together, and edges are created to represent recirculations.
This format comes with further features such as displaying the conntrack
zones, which are key to understanding what the datapath is really doing with a
packet.
The ``html`` and ``graph`` can also be combined.
``ovs-flowviz datapath graph --html`` command will output an interactive
HTML table alongside a SVG graphical representation of the flows. Flows in the
SVG representation link to the corresponding entry in the HTML table.
Filtering
~~~~~~~~~
As well as allowing expanding and collapsing subtrees, filtering can be used.
However, filtering works in a slightly different way than it does with OpenFlow
flows. Instead of just removing non-matching flows, the output of a filtered
datapath flow tree will show full sub-trees containing at least one flow that
satisfies the filter.
For example, the following command allows understanding the flows in the above
example in the context of traffic going out on port ``ovn-k8s-mp0``:
::
$ ovs-appctl dpctl/dump-flows | ovs-flowviz -f "output.port=ovn-k8s-mp0" datapath tree
The resulting flow tree will contain all of the flows above, including those
with ``recirc_id(0)`` and ``recirc_id(0xc1)`` that don't actually output
traffic to port ``ovn-k8s-mp0``. This is because they are part of a subtree
that contains flows that output packets on port ``ovn-k8s-mp0``
This provides a "full picture" of how traffic, ending up in a particular
port, is being processed.
.. _ovs-flowviz(8): https://docs.openvswitch.org/en/latest/ref/ovs-flowviz.8

View File

@ -58,3 +58,4 @@ OVS
userspace-checksum-offloading
userspace-tx-steering
usdt-probes
flow-visualization

4
NEWS
View File

@ -5,6 +5,10 @@ Post-v3.4.0
that does not have a specific value defined, rather than being
treated as a global value, aligning the behavior with that of
the kernel datapath.
- Python:
* Added tool called "ovs-flowviz" capable of parsing OpenFlow
and datapath flow dumps and displaying them in several different
formats.
v3.4.0 - 15 Aug 2024

View File

@ -500,6 +500,7 @@ fi
%{_mandir}/man8/ovs-ctl.8*
%{_mandir}/man8/ovs-dpctl.8*
%{_mandir}/man8/ovs-dpctl-top.8*
%{_mandir}/man8/ovs-flowviz.8*
%{_mandir}/man8/ovs-kmod-ctl.8*
%{_mandir}/man8/ovs-ofctl.8*
%{_mandir}/man8/ovs-pki.8*

View File

@ -223,6 +223,7 @@ exit 0
/usr/share/man/man8/ovs-ctl.8.gz
/usr/share/man/man8/ovs-dpctl.8.gz
/usr/share/man/man8/ovs-dpctl-top.8.gz
/usr/share/man/man8/ovs-flowviz.8.gz
/usr/share/man/man8/ovs-kmod-ctl.8.gz
/usr/share/man/man8/ovs-ofctl.8.gz
/usr/share/man/man8/ovs-parse-backtrace.8.gz