Update the controller on son_slave connect/disconnect event #931

morantr · 2020-03-08T17:16:53Z

In the past, we used to have a TCP/IP socket between the Controller and the Agent.
When a radio went down, the son_slave perform reset and closed the socket to the controller.
When that happened, the socket on the controller side was closed too, and the controller removed the node of the son_slave from the database.

Nowadays, we have a BUS infrastructure that replaces the TCP/IP socket. This means that if the son_slave perform reset,
No socket is closed and the controller doesn’t aware that and Agent radio is down.

This PP adds a Topology Query message send after the controller receives M1
and add missing fronthaul radio information to Topology Response message, so the Controller
could remove from its database inactive radio (son_salve) nodes.

#925

tomereli

Mainly nitpick comments, but not approving yet cause I think this should be rebased on #900 instead of using dummy data.

controller/src/beerocks/master/son_master_thread.cpp

agent/src/beerocks/slave/backhaul_manager/backhaul_manager_thread.cpp

controller/src/beerocks/master/son_master_thread.cpp

tomereli

Since support for WIPHY commands via NL80211 is not ready yet - approved, but please add a task and a TODO comment in the code and commit message which mention that.

arnout

Looks great! I have a few important comments, but everything can be fixed easily, so approve with don't merge.

framework/tlvf/yaml/tlvf/ieee_1905_1/s802_11SpecificInformation.yaml

agent/src/beerocks/slave/backhaul_manager/backhaul_manager_thread.cpp

arnout · 2020-03-12T13:37:18Z

agent/src/beerocks/slave/backhaul_manager/backhaul_manager_thread.cpp

+            // For now, put zeros and when the Agent management will be move to unified Agent thread
+            // this field will be filled. #435


This is not true. The unified agent is about unifying the ap manager and monitor threads. There would still be a separate process for each radio. The idea is that the per-radio process needs to link to bwl and is more prone to crashing, plus it may need more capabilities or different SELinux configuration.

The topology response will always be constructed by the main agent process, not by the per-radio process, so #435 will not solve this.

Unify Agent is the entity that will connect to two fronthaul managers (per radio) that each one will have a monitor and ap manager. Meaning, it will be the main Agent process.
The backhaul manager will be responsible only to link establishment and remove from him all the Agent stuff.

controller/src/beerocks/master/son_master_thread.cpp

arnout · 2020-03-12T16:23:03Z

controller/src/beerocks/master/son_master_thread.cpp

+    auto fronthaul_radios_on_db = database.get_node_children(
+        network_utils::mac_to_string(al_mac), beerocks::TYPE_SLAVE, beerocks::STATE_CONNECTED);
+
+    std::unordered_set<std::string> reported_fronthaul_radios;


Suggested change

std::unordered_set<std::string> reported_fronthaul_radios;

std::unordered_set<sMacAddress> reported_fronthaul_radios;

I did it because of the database return set of std::string.
I don't want to change the database in this PR scope.
Let's give #524 more priority.

Still, the decision is to use sMacAddress instead of std::string even if we don't change everything at once, so please change to sMacAddress as part of this PR.

Indeed. It's OK for fronthaul_radios_on_db because that is indeed returned by the database. But reported_fronthaul_radios gets filled with the MAC addresses that come from the device information so they're not strings. And where you do reported_fronthaul_radios.find(fronthaul_radio_on_db) you can easily do reported_fronthaul_radios.find(network_utils::string_to_mac(fronthaul_radio_on_db)) instead.

Anyway, I'm not going to block the merge for this. But please take it into account next time.

arnout · 2020-03-12T16:25:42Z

controller/src/beerocks/master/son_master_thread.cpp

+            // const auto &iface_bssid = media_info->network_membership;
+            // const auto iface_bw     = media_info->ap_channel_bandwidth;
+            // const auto iface_cf1    = media_info->ap_channel_center_frequency_index1;
+            // const auto iface_cf2    = media_info->ap_channel_center_frequency_index2;


Suggested change

// const auto &iface_bssid = media_info->network_membership;

// const auto iface_bw = media_info->ap_channel_bandwidth;

// const auto iface_cf1 = media_info->ap_channel_center_frequency_index1;

// const auto iface_cf2 = media_info->ap_channel_center_frequency_index2;

// In MultiAP context, bssid, bw, cf1 and cf2 are not actually useful since they are more accurately

// represented in other TLVs (operating channel report, AP operational BSS).

AFAIR, the current channel selection task does use it. I'm not sure that we will not use it. Keep it for now.

Our mindset is to delete what we don't use now - so I don't see why we should keep it. In addition, the channel selection task will be stripped down to nothing very soon so we don't really care if it uses it since its obsolete.

Channel selection task will use this information - but coming from the AP Radio Basic Capabilities and the Channel Preference Report, which is a lot more accurate than the limited information given in this TLV.

I hoped that my suggested edit expressed this feeling accurately...

After further discusion - it turns out that p_channel_center_frequency_index2 is missing from the Operating Channel Report TLV, so we will need to use this for 80+80MHz operation... Which is super-weird...

controller/src/beerocks/master/son_master_thread.cpp

Send topology message query upon receiving auto-config m1 message. Signed-off-by: Moran Shoeg <[email protected]>

Change the name of the member "ap_channel_band" to "ap_channel_bandwidth". The term "band" is wrong and not reflects the field's meaning. Update auto-generated code. Signed-off-by: Moran Shoeg <[email protected]>

The controller needs to be able to add and remove the radio node from its database based on the data that is coming from the TOPOLOGY RESPONSE message. Add dummy data on the wireless fronthaul interfaces in the tlvDeviceInformation which is in the Topology Response message. The reason that it is dummy values is that this is something that cannot be done at this moment because the MediaType field must be computed with information obtained through NL80211_CMD_GET_WIPHY command of DWPAL, which is currently not supported. See "Send standard NL80211 commands using DWPAL #782 The filled band data is based on the information we have about the band but it is not accurate. The fields ap_channel_bandwidth, ap_channel_center_frequency_index1 and ap_channel_center_frequency_index2 we could get (the bw for sure, the others not so sure), but it not worth it since it will require to create more internal messaging which will be deleted after moving to a unified Agent (#435). Signed-off-by: Moran Shoeg <[email protected]>

When son_slave socket disconnects or connects, send a TOPOLOGY NOTIFICATION with no TLVs to the controller. The controller will send TOPOLOGY QUERY and on the response, it will be notified which radio (son_slave) when up or down. Signed-off-by: Moran Shoeg <[email protected]>

Subscribe the controller to TOPOLOGY RESPONSE MESSAGE and add handler function to it with parsing of device information type TLV. Add a piece of code that checks whether an unreported fronthaul radio interface exists on the controller database, and if not, remove its node. Signed-off-by: Moran Shoeg <[email protected]>

Send TOPOLOGY_QUERY_MESSAGE when the controller receives TOPOLOGY_NOTIFICATION_MESSAGE with no Client Association Event TLV. This will result in getting TOPOLOGY_RESPONSE_MESSAGE with all the information the controller needs to update the topology on its database. Signed-off-by: Moran Shoeg <[email protected]>

morantr requested review from arnout, rmelotte and tomereli as code owners March 8, 2020 17:16

github-actions bot assigned morantr Mar 8, 2020

morantr force-pushed the feature/detect-son-slave-reset-with-topology-flow branch from 2073992 to ee9f665 Compare March 8, 2020 17:45

morantr added the don't merge This PR is not ready for merge or review label Mar 8, 2020

morantr force-pushed the feature/detect-son-slave-reset-with-topology-flow branch from e5a5b5a to c530f4e Compare March 10, 2020 12:42

morantr removed the don't merge This PR is not ready for merge or review label Mar 10, 2020

morantr force-pushed the feature/detect-son-slave-reset-with-topology-flow branch from c530f4e to 2398b0d Compare March 10, 2020 16:45

tomereli reviewed Mar 11, 2020

View reviewed changes

morantr force-pushed the feature/detect-son-slave-reset-with-topology-flow branch 3 times, most recently from 3f8e8fd to 68a848b Compare March 12, 2020 08:25

morantr requested a review from tomereli March 12, 2020 08:25

tomereli approved these changes Mar 12, 2020

View reviewed changes

morantr force-pushed the feature/detect-son-slave-reset-with-topology-flow branch 3 times, most recently from 6cfea97 to 0d50037 Compare March 12, 2020 16:23

arnout added the don't merge This PR is not ready for merge or review label Mar 12, 2020

arnout approved these changes Mar 12, 2020

View reviewed changes

morantr force-pushed the feature/detect-son-slave-reset-with-topology-flow branch from 0d50037 to e009eb7 Compare March 15, 2020 12:04

morantr requested a review from arnout March 15, 2020 12:15

arnout removed the don't merge This PR is not ready for merge or review label Mar 16, 2020

morantr added 5 commits March 16, 2020 11:40

controller: Send TOPOLOGY_QUERY_MESSAGE upon receiving m1

cfecfcb

Send topology message query upon receiving auto-config m1 message. Signed-off-by: Moran Shoeg <[email protected]>

tlvf: Change tlv member name

9f75d87

Change the name of the member "ap_channel_band" to "ap_channel_bandwidth". The term "band" is wrong and not reflects the field's meaning. Update auto-generated code. Signed-off-by: Moran Shoeg <[email protected]>

arnout force-pushed the feature/detect-son-slave-reset-with-topology-flow branch from e009eb7 to 39c59c6 Compare March 16, 2020 11:41

mergify bot merged commit 68dbd33 into master Mar 16, 2020

mergify bot deleted the feature/detect-son-slave-reset-with-topology-flow branch March 16, 2020 11:52

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Update the controller on son_slave connect/disconnect event #931

Update the controller on son_slave connect/disconnect event #931

morantr commented Mar 8, 2020 •

edited

Loading

tomereli left a comment

tomereli left a comment

arnout left a comment

arnout Mar 12, 2020

morantr Mar 12, 2020

arnout Mar 12, 2020

morantr Mar 15, 2020

tomereli Mar 15, 2020

arnout Mar 16, 2020

arnout Mar 12, 2020

morantr Mar 15, 2020

tomereli Mar 15, 2020

arnout Mar 16, 2020 •

edited

Loading

arnout Mar 16, 2020

		// For now, put zeros and when the Agent management will be move to unified Agent thread
		// this field will be filled. #435

	std::unordered_set<std::string> reported_fronthaul_radios;
	std::unordered_set<sMacAddress> reported_fronthaul_radios;

Update the controller on son_slave connect/disconnect event #931

Update the controller on son_slave connect/disconnect event #931

Conversation

morantr commented Mar 8, 2020 • edited Loading

tomereli left a comment

Choose a reason for hiding this comment

tomereli left a comment

Choose a reason for hiding this comment

arnout left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

arnout Mar 16, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

morantr commented Mar 8, 2020 •

edited

Loading

arnout Mar 16, 2020 •

edited

Loading