Magpie

Magpie is a charm used for testing the networking of a Juju provider/substrate.

It provides tools for testing:

• DNS functionality
• network connectivity between nodes (iperf, ping)
• network benchmarking
• MTU
• local hostname lookup

Usage

Deploy the charm to two or more units, then run the provided actions to retrieve debug information about the nodes or run network diagnostic tests.

juju deploy magpie -n 3

juju actions magpie
juju run magpie/leader info
juju run magpie/leader ping
# etc.

Check the charm config before deploying for values you may wish to tweak, and see the parameters accepted by each action.

TODO: document each action and the expected results

Network spaces

If you use network spaces in your Juju deployment (as you should) use --bind '<space-name> magpie=<space-name>' to force magpie to test that particular network space.

It is possible to deploy several magpie charms (as different Juju applications) to the same server each in a different network space.

Example:

juju deploy magpie magpie-space1 --bind "space1 magpie=space1" -n 5 --to 0,2,1,4,3
juju deploy magpie magpie-space2 --bind "space2 magpie=space2" -n 3 --to 3,2,0
juju deploy magpie magpie-space3 --bind "space3 magpie=space3" -n 4 --to 3,2,1,0
juju deploy magpie magpie-space4 --bind "space4 magpie=space4" -n 4 --to 3,2,1,0

Benchmarking network with iperf and grafana

Assumes juju 3.1

Step 1, deploy COS:

# Deploy COS on microk8s.
# https://charmhub.io/topics/canonical-observability-stack/tutorials/install-microk8s
juju bootstrap microk8s microk8s
juju add-model cos
juju deploy cos-lite

# Expose the endpoints for the magpie model to consume.
juju offer grafana:grafana-dashboard
juju offer prometheus:receive-remote-write

Step 2, deploy magpie and relate to COS

juju switch <controller for cloud to be benchmarked>
juju add-model magpie

juju consume microk8s:cos.prometheus
juju consume microk8s:cos.grafana

# adjust as required
juju deploy magpie -n 3
juju deploy ./magpie_ubuntu-22.04-amd64.charm -n 3

juju deploy grafana-agent --channel edge
juju relate magpie grafana-agent
juju relate grafana-agent prometheus
juju relate grafana-agent grafana

Step 3, run the iperf action and view results in grafana:

# adjust as needed
juju run magpie/0 iperf

# you may wish to run against one unit pair at a time:
juju run magpie/0 iperf units=magpie/1
juju run magpie/0 iperf units=magpie/2
# etc.

Obtain details to access grafana from COS:

juju show-unit -m microk8s:cos catalogue/0 --format json | jq -r '.["catalogue/0"]."relation-info"[] | select(."application-data".name == "Grafana") | ."application-data".url'
juju config -m microk8s:cos grafana admin_user
juju run -m microk8s:cos grafana/0 get-admin-password

Find the dashboard titled "Magpie Network Benchmarking", and limit the time range as required.

Bonded links testing and troubleshooting

Network bonding enables the combination of two or more network interfaces into a single-bonded (logical) interface, which increases the bandwidth and provides redundancy. While Magpie does some sanity checks and could reveal some configuration problems, this part of README contains some advanced troubleshooting information, which might be useful, while identifying and fixing the issue.

There are six bonding modes:

balance-rr

Round-robin policy: Transmit packets in sequential order from the first available slave through the last. This mode provides load balancing and fault tolerance.

active-backup

Active-backup policy: Only one slave in the bond is active. A different slave becomes active if, and only if, the active slave fails. The bond's MAC address is externally visible on only one port (network adapter) to avoid confusing the switch. This mode provides fault tolerance. The primary option affects the behavior of this mode.

balance-xor

XOR policy: Transmit based on selectable hashing algorithm. The default policy is a simple source+destination MAC address algorithm. Alternate transmit policies may be selected via the xmit_hash_policy option, described below. This mode provides load balancing and fault tolerance.

broadcast

Broadcast policy: transmits everything on all slave interfaces. This mode provides fault tolerance.

802.3ad (LACP)

Link Aggregation Control Protocol (IEEE 802.3ad LACP) is a control protocol that automatically detects multiple links between two LACP enabled devices and configures them to use their maximum possible bandwidth by automatically trunking the links together. This mode has a prerequisite - the switch(es) ports should have LACP configured and enabled.

balance-tlb

Adaptive transmit load balancing: channel bonding that does not require any special switch support. The outgoing traffic is distributed according to the current load (computed relative to the speed) on each slave. Incoming traffic is received by the current slave. If the receiving slave fails, another slave takes over the MAC address of the failed receiving slave.

balance-alb

Adaptive load balancing: includes balance-tlb plus receive load balancing (rlb) for IPV4 traffic, and does not require any special switch support. The receive load balancing is achieved by ARP negotiation.

The most commonly used modes are active-backup and 802.3ad (LACP), and while active-backup does not require any third party configuration, it has its own cons - for example, it can't multiply the total bandwidth of the link, while 802.3ad-based bond could utilize all bond members, therefore multiplying the bandwidth. However, in order to get a fully working LACP link, an appropriate configuration has to be done both on the actor (link initiator) and partner (switch) side. Any misconfiguration could lead to the link loss or instability, therefore it's very important to have correct settings applied to the both sides of the link.

A quick overview of the LACP link status could be obtained by reading the /proc/net/bonding/<bond_name> file.

$ sudo cat /proc/net/bonding/bondM
Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011)

Bonding Mode: IEEE 802.3ad Dynamic link aggregation
Transmit Hash Policy: layer3+4 (1)
MII Status: up
MII Polling Interval (ms): 100
Up Delay (ms): 0
Down Delay (ms): 0

802.3ad info
LACP rate: fast
Min links: 0
Aggregator selection policy (ad_select): stable
System priority: 65535
System MAC address: 82:23:80:a1:a9:d3
Active Aggregator Info:
 Aggregator ID: 1
 Number of ports: 2
 Actor Key: 15
 Partner Key: 201
 Partner Mac Address: 02:01:00:00:01:01

Slave Interface: eno3
MII Status: up
Speed: 10000 Mbps
Duplex: full
Link Failure Count: 0
Permanent HW addr: 3c:ec:ef:19:eb:30
Slave queue ID: 0
Aggregator ID: 1
Actor Churn State: none
Partner Churn State: none
Actor Churned Count: 0
Partner Churned Count: 0
details actor lacp pdu:
    system priority: 65535
    system mac address: 82:23:80:a1:a9:d3
    port key: 15
    port priority: 255
    port number: 1
    port state: 63
details partner lacp pdu:
    system priority: 65534
    system mac address: 02:01:00:00:01:01
    oper key: 201
    port priority: 1
    port number: 12
    port state: 63

Slave Interface: eno1
MII Status: up
Speed: 10000 Mbps
Duplex: full
Link Failure Count: 0
Permanent HW addr: 3c:ec:ef:19:eb:2e
Slave queue ID: 0
Aggregator ID: 1
Actor Churn State: none
Partner Churn State: none
Actor Churned Count: 0
Partner Churned Count: 0
details actor lacp pdu:
    system priority: 65535
    system mac address: 82:23:80:a1:a9:d3
    port key: 15
    port priority: 255
    port number: 2
    port state: 63
details partner lacp pdu:
    system priority: 65534
    system mac address: 02:01:00:00:01:01
    oper key: 201
    port priority: 1
    port number: 1012
    port state: 63

The key things an operator should take a look at is:

• LACP rate
• Actor Churn State
• Partner Churn State
• Port State

LACP rate

The Link Aggregation Control Protocol (LACP) provides a standardized means for exchanging information between Partner Systems on a link to allow their Link Aggregation Control instances to reach agreement on the identity of the LAG to which the link belongs, move the link to that LAG, and enable its transmission and reception functions in an orderly manner. The protocol depends upon the transmission of information and state, rather than the transmission of commands. LACPDUs (LACP Data Unit) sent by the first party (the Actor) convey to the second party (the Actor’s protocol Partner) what the Actor knows, both about its own state and that of the Partner.

Periodic transmission of LACPDUs occurs if the LACP Activity control of either the Actor or the Partner is Active LACP. These periodic transmissions will occur at either a slow or fast transmission rate depending upon the expressed LACP_Timeout preference (Long Timeout or Short Timeout) of the Partner System.

Actor/Partner Churn State

In general, "Churned" port status means that the parties are unable to reach agreement upon the desired state of a link. Under normal operation of the protocol, such a resolution would be reached very rapidly; continued failure to reach agreement can be symptomatic of component failure, of the presence of non-standard devices on the link concerned, or of mis-configuration. Hence, detection of such failures is signalled by the Churn Detection algorithm to the operator in order to prompt administrative action to further resolution.

Port State

Both of the Actor and Partner state are variables, encoded as individual bits within a single octet, as follows.

0. LACP_Activity: Device intends to transmit periodically in order to find potential members for the aggregate. Active LACP is encoded as a 1; Passive LACP as a 0.
1. LACP_Timeout: This flag indicates the Timeout control value with regard to this link. Short Timeout is encoded as a 1; Long Timeout as a 0.
2. Aggregability: This flag indicates that the system considers this link to be Aggregateable; i.e., a potential candidate for aggregation. If FALSE (encoded as a 0), the link is considered to be Individual; i.e., this link can be operated only as an individual link. Aggregatable is encoded as a 1; Individual is encoded as a 0.
3. Synchronization: Indicates that the bond on the transmitting machine is in sync with what’s being advertised in the LACP frames, meaning the link has been allocated to the correct LAG, the group has been associated with a compatible Aggregator, and the identity of the LAG is consistent with the System ID and operational Key information transmitted. "In Sync" is encoded as a 1; "Out of sync" is encoded as a 0.
4. Collecting: Bond is accepting traffic received on this port, collection of incoming frames on this link is definitely enabled and is not expected to be disabled in the absence of administrative changes or changes in received protocol information. True is encoded as a 1; False is encoded as a

5. Distributing: Bond is sending traffic using these ports encoded. Same as above, but for egress traffic. True is encoded as a 1; False is encoded as a 0.
6. Defaulted: Determines, whether the receiving bond is using default (administratively defined) parameters, if the information was received in an LACP PDU. Default settings are encoded as a 1, LACP PDU is encoded as 0.
7. Expired: Is the bond in the expired state. Yes encoded as a 1, No encoded as a 0.

In the example output above, both of the port states are equal to 63. Let's decode:

$ python3
Python 3.8.4 (default, Jul 17 2020, 15:44:37)
[Clang 11.0.3 (clang-1103.0.32.62)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> bin(63)
'0b111111'

Reading right to the left:

LACP Activity: Active LACP Timeout: Short Aggregability: Link is Aggregatable Synchronization: Link in sync Collecting: True - bond is accepting the traffic Distributing: True - bond is sending the traffic Defaulted: Info received from LACP PDU Expired: False - link is not expired

The above status represents the fully healthy bond without any LACP-related issues. Also, for the operators' convenience, the lacp_decoder.py script could be used to quickly convert the status to some human-friendly format.

However, the situations where one of the links is misconfigured are happening too, so let's assume we have the following:

$ sudo cat /proc/net/bonding/bondm
Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011)

Bonding Mode: IEEE 802.3ad Dynamic link aggregation
Transmit Hash Policy: layer3+4 (1)
MII Status: up
MII Polling Interval (ms): 100
Up Delay (ms): 0
Down Delay (ms): 0

802.3ad info
LACP rate: fast
Min links: 0
Aggregator selection policy (ad_select): stable
System priority: 65535
System MAC address: b4:96:91:6d:20:fc
Active Aggregator Info:
        Aggregator ID: 2
        Number of ports: 1
        Actor Key: 9
        Partner Key: 32784
        Partner Mac Address: 00:23:04:ee:be:66

Slave Interface: enp197s0f2
MII Status: up
Speed: 100 Mbps
Duplex: full
Link Failure Count: 0
Permanent HW addr: b4:96:91:6d:20:fe
Slave queue ID: 0
Aggregator ID: 1
Actor Churn State: churned
Partner Churn State: none
Actor Churned Count: 1
Partner Churned Count: 0
details actor lacp pdu:
    system priority: 65535
    system mac address: b4:96:91:6d:20:fc
    port key: 7
    port priority: 255
    port number: 1
    port state: 7
details partner lacp pdu:
    system priority: 32667
    system mac address: 00:23:04:ee:be:66
    oper key: 32784
    port priority: 32768
    port number: 16661
    port state: 13

Slave Interface: enp197s0f0
MII Status: up
Speed: 1000 Mbps
Duplex: full
Link Failure Count: 0
Permanent HW addr: b4:96:91:6d:20:fc
Slave queue ID: 0
Aggregator ID: 2
Actor Churn State: none
Partner Churn State: none
Actor Churned Count: 0
Partner Churned Count: 0
details actor lacp pdu:
    system priority: 65535
    system mac address: b4:96:91:6d:20:fc
    port key: 9
    port priority: 255
    port number: 2
    port state: 63
details partner lacp pdu:
    system priority: 32667
    system mac address: 00:23:04:ee:be:66
    oper key: 32784
    port priority: 32768
    port number: 277
    port state: 63

As we could see, one of the links has different port states for both partner and actor, while the second one has 63 for both - meaning, the first one is problematic and we'd need to dive more into this problem.

Let's decode both of the statuses, using the mentioned script:

$ python ./lacp-decoder.py 7 13
(Equal for both ports) LACP Activity: Active LACP
LACP Timeout: Short (Port 1) / Long (Port 2)
(Equal for both ports) Aggregability: Aggregatable
Synchronization: Link out of sync (Port 1) / Link in sync (Port 2)
(Equal for both ports) Collecting: Ingress traffic: Rejecting
(Equal for both ports) Distributing: Egress traffic: Not sending
(Equal for both ports) Is Defaulted: Settings are received from LACP PDU
(Equal for both ports) Link Expiration: No

The above output means that there are two differences between these statuses: LACP Timeout and Synchronization. That means two things:

1. the Partner side (a switch side in most of the cases) has incorrectly configured LACP timeout control. To resolve this, an operator has to either change the LACP rate from the Actor (e.g a server) side to "Slow", or adjust the Partner (e.g switch) LACP rate to "Fast".
2. the Partner side considers this physical link as a part of a different link aggregation group. The switch config has to be revisited and link aggregation group members need to be verified again, ensuring there is no extra or wrong links configured as part of the single LAG.

After addressing the above issues, the port state will change to 63, which means "LACP link is fully functional".

Bugs

Please report bugs on Launchpad.

For general questions please refer to the OpenStack Charm Guide.