Cisco MDT input: minor improvements and support for NX-OS telemetry extensions #6177

sbyx · 2019-07-29T06:47:31Z

This adds support for the platform-specific telemetry features of Cisco NX-OS platforms including support for gzip compression (via native GRPC compression) and chunking data over different GRPC messages as well as decoding logic for NX-OS DME and NX-API data structures over telemetry.

For the latter see as reference:

In addition the following other things are addressed:

Correctly close the listener in error case when setting up.
Simplify field-name generation by getting rid of passed around bytes.Buffer instances

danielnelson

Let's add some unit tests for the NX-OS message handling too.

danielnelson · 2019-08-27T01:21:35Z

plugins/inputs/cisco_telemetry_mdt/cisco_telemetry_mdt.go

+		if packet.TotalSize == 0 {
+			c.handleTelemetry(packet.Data)
+		} else if int(packet.TotalSize) <= c.MaxMsgSize {
+			chunkBuffer.Write(packet.Data)


Why do we need a bytes.Buffer, why not just send the packet.Data into handleTelemetry?

NX-OS can send data which is too large for a single GRPC message and if so uses "application layer" chunking where one telemetry message is reassembled from multiple GRPC messages.

danielnelson · 2019-08-27T01:49:01Z

plugins/inputs/cisco_telemetry_mdt/cisco_telemetry_mdt.go

+		for _, subfield := range field.Fields {
+			c.parseGPBKVField(subfield, name, path, timestamp, tags, fields)
+		}
+	} else if c.DecodeNXOS && len(field.Fields) > 0 { // NX-OS extended decoding logic


Can we inspect the field and determine if it is NXOS data automatically? If so we could remove the option and handle it automatically.

Done in latest update.

danielnelson · 2019-08-27T01:52:29Z

plugins/inputs/cisco_telemetry_mdt/cisco_telemetry_mdt.go

+		}
+		if rn, hasRN := values["rn"]; hasRN {
+			// Promote the relative name of the entry from a value to a key
+			tags[prefix] = fmt.Sprint(rn)


Using fmt.Sprint here can be a bit risky, since it is assuming the interface{} has the proper string representation. It would be better to create a helper function with a type switch that converts known types to a string.

Done in latest update.

sbyx · 2019-09-06T14:57:10Z

I have reworked some of the code a bit for better clarity. I have also added a feature (EmbeddedTags) to address issue #6322 where the device emits unexpected nested content that contains additional keys.

danielnelson

Here is my review, overall I think this code needs more work yet, some of the complexity is unavoidable due to the nested structure of the messages but we should try to flatten the processing so that it is as straightforward as we can get it.

danielnelson · 2019-09-13T23:52:46Z

plugins/inputs/cisco_telemetry_mdt/README.md

@@ -32,6 +32,9 @@ The TCP dialout transport is supported on IOS XR (32-bit and 64-bit) 6.1.x and l
 ## Define aliases to map telemetry encoding paths to simple measurement names
 [inputs.cisco_telemetry_mdt.aliases]
   ifstats = "ietf-interfaces:interfaces-state/interface/statistics"
+
+ ## Define (for certain nested telemetry measurements with embedded tags) which fields are tags
+ # enbedded_tags = ["Cisco-IOS-XR-qos-ma-oper:qos/interface-table/interface/input/service-policy-names/service-policy-instance/statistics/class-stats/class-name"]


Move this option up above the aliases subtable so it doesn't fall into the table. Also s/enbedded_tags/embedded/

danielnelson · 2019-09-13T23:54:52Z

plugins/inputs/cisco_telemetry_mdt/cisco_telemetry_mdt.go

+	// Fill extra tags
+	c.extraTags = make(map[string]map[string]struct{})
+	for _, tag := range c.EmbeddedTags {
+		i := strings.LastIndexByte(tag, '/')


Try to use path.Base here to get the last element of the path. You can also use path.Dir for everything up to the last item.

danielnelson · 2019-09-14T00:25:01Z

plugins/inputs/cisco_telemetry_mdt/cisco_telemetry_mdt.go

-				fields = make(map[string]interface{}, len(field.Fields))
-				for _, subfield := range field.Fields {
-					c.parseGPBKVField(subfield, &namebuf, telemetry.EncodingPath, timestamp, tags, fields)
+				if tags != nil {


This flow is a bit confusing for me, it looks like there is some sort of expected ordering here: "keys" should come before "content"? To what degree is this guaranteed? Is it possible that "keys" is unset? Can there be more than one "keys" or "content" section? Are there other sections to ignore?

Not knowing these answers it is hard for me know how to fix, but one way I think this code to be more straightforward is that instead of looping over the fields and having a switch statement, have a function that gets the fields, then you can do the flow in a way that has less looping:

func GetTelemetryField(f *telemetry.Field, name string) *telemetry.Field

keysField, err := GetTelemetryField(gpvkb.Fields, "keys") if err != nil { return err } tags := err := parseKeyField(keysField, tags) if err != nil { return err } contentField, err := GetTelemetryField(gpvkb.Fields, "content") if err != nil { return err } // tags were definitely found. err := parseContentField(contentField, tags) if err != nil { return err }

Yes, so far the assumption was there was a "keys"-field followed by a "content"-field which holds true for all datasets I have seen so far. But since it does not hurt I have now made a simple loop which just looks for these 2 fields afterwards a check if both have been found and an early continue if one or both are missing.

danielnelson · 2019-09-14T00:25:54Z

plugins/inputs/cisco_telemetry_mdt/cisco_telemetry_mdt.go

+	if len(name) == 0 {
+		name = prefix
+	} else if len(prefix) > 0 {
+		name = prefix + "/" + localname


What if localname = ""? Do we still want to do this?

Line 371 here checks if localname is empty (name is set to localname one line before) but I will clarify.

danielnelson · 2019-09-14T00:26:40Z

plugins/inputs/cisco_telemetry_mdt/cisco_telemetry_mdt.go

+	localname := strings.Replace(field.Name, "-", "_", -1)
+	name := localname
+	if len(name) == 0 {
+		name = prefix


What if prefix is "", is this still right? Is this to set the top "keys" field to ""?

Children of "keys" are iterated with prefix set to "". This branch is mainly necessary for NX-OS devices sending fields with empty name that have children. Without this we would get fieldnames with multiple /. See e.g. the format in the testcases for NX-OS decoding.

danielnelson · 2019-09-14T00:52:18Z

plugins/inputs/cisco_telemetry_mdt/cisco_telemetry_mdt.go

-		if fields != nil {
-			fields[namebuf.String()] = value
+	if tag := decodeTag(field); len(tag) > 0 {
+		localname := strings.Replace(field.Name, "-", "_", -1)


Does this need done again? Isn't it the same as on L369?

Nope not necessary, will be removed.

danielnelson · 2019-09-14T01:05:47Z

plugins/inputs/cisco_telemetry_mdt/cisco_telemetry_mdt.go

+		if len(rn) > 0 {
+			tags[prefix] = rn
+		} else if !dn { // Check for distinguished name being present
+			c.acc.AddError(fmt.Errorf("NX-OS decoding failed: missing dn field"))


I think we should stop processing the attributes and return an error here.

Agreed. I will add a return here. This is checking for a possible corner case I haven't encountered in practice yet. I will add a return statement.

So parsing can continue with the next valid field.

danielnelson · 2019-09-14T01:11:58Z

plugins/inputs/cisco_telemetry_mdt/cisco_telemetry_mdt.go

+
+	extraTags := c.extraTags[path+"/"+name]
+
+	if value := decodeValue(field); value != nil {


This seems like an error, can we return early if decode fails? We should try to do this whenever we can, if the message isn't valid then lets skip it instead of trying to produce something which may not be right.

value, err := decodeValue(field) if err != nil return err }

value being nil is legitimate when the field has children. usually a field has either children or a value (e.g. a tree-structure where only leafs carry values). I have added a return statement inside this branch, i.e. if it is a value, stop processing children.

danielnelson · 2019-09-14T01:14:14Z

plugins/inputs/cisco_telemetry_mdt/cisco_telemetry_mdt.go

 	}

-	namebuf.Truncate(namelen)
+	if nxAttributes != nil {


Return early if unset.

Okay reorganized this part a bit, to unindent some longer code branches.

sbyx · 2019-09-14T13:54:16Z

Tried to address all open points with the new commit. Please let me know if anything still remains to be changed.

…#6177)

sbyx force-pushed the telemetry-nxos branch from c2c98fe to be4d70b Compare July 29, 2019 11:33

danielnelson added the feat Improvement on an existing feature such as adding a new setting/mode to an existing plugin label Jul 31, 2019

danielnelson self-requested a review July 31, 2019 04:34

danielnelson suggested changes Aug 27, 2019

View reviewed changes

sbyx added 3 commits September 6, 2019 13:35

Cisco MDT: add support for NX-OS telemetry extensions

4156d0a

Update Gopkg.lock

4e873dd

Rework NX-OS support, add support for embedded tags

4e94071

sbyx force-pushed the telemetry-nxos branch from be4d70b to 4e94071 Compare September 6, 2019 14:08

danielnelson suggested changes Sep 14, 2019

View reviewed changes

Some clarifications and cleanup

7036ffe

russorat added this to the 1.13.0 milestone Sep 16, 2019

danielnelson approved these changes Sep 16, 2019

View reviewed changes

danielnelson merged commit ee9d0fc into influxdata:master Sep 16, 2019

bitcharmer pushed a commit to bitcharmer/telegraf that referenced this pull request Oct 18, 2019

Support NX-OS telemetry extensions in cisco_telemetry_mdt (influxdata…

f646c52

…#6177)

athoune pushed a commit to bearstech/telegraf that referenced this pull request Apr 17, 2020

Support NX-OS telemetry extensions in cisco_telemetry_mdt (influxdata…

e4d63e8

…#6177)

idohalevi pushed a commit to idohalevi/telegraf that referenced this pull request Sep 29, 2020

Support NX-OS telemetry extensions in cisco_telemetry_mdt (influxdata…

08eb6cb

…#6177)

arstercz pushed a commit to arstercz/telegraf that referenced this pull request Mar 5, 2023

Support NX-OS telemetry extensions in cisco_telemetry_mdt (influxdata…

e00976d

…#6177)

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Cisco MDT input: minor improvements and support for NX-OS telemetry extensions #6177

Cisco MDT input: minor improvements and support for NX-OS telemetry extensions #6177

sbyx commented Jul 29, 2019

danielnelson left a comment

danielnelson Aug 27, 2019

sbyx Sep 2, 2019

danielnelson Aug 27, 2019

sbyx Sep 6, 2019

danielnelson Aug 27, 2019

sbyx Sep 6, 2019

sbyx commented Sep 6, 2019

danielnelson left a comment

danielnelson Sep 13, 2019

sbyx Sep 14, 2019

danielnelson Sep 13, 2019

sbyx Sep 14, 2019

danielnelson Sep 14, 2019

sbyx Sep 14, 2019

danielnelson Sep 14, 2019

sbyx Sep 14, 2019

danielnelson Sep 14, 2019

sbyx Sep 14, 2019

danielnelson Sep 14, 2019

sbyx Sep 14, 2019

danielnelson Sep 14, 2019

sbyx Sep 14, 2019

sbyx Sep 14, 2019

danielnelson Sep 14, 2019

sbyx Sep 14, 2019

danielnelson Sep 14, 2019

sbyx Sep 14, 2019

sbyx commented Sep 14, 2019


		extraTags := c.extraTags[path+"/"+name]

		if value := decodeValue(field); value != nil {

Cisco MDT input: minor improvements and support for NX-OS telemetry extensions #6177

Cisco MDT input: minor improvements and support for NX-OS telemetry extensions #6177

Conversation

sbyx commented Jul 29, 2019

danielnelson 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

sbyx commented Sep 6, 2019

danielnelson 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

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

sbyx commented Sep 14, 2019