[LaunchDarkly Segment] What is the difference between rules.clauses.context_kind and rule.rollout_context_kind?

[ozheng-circle]

I am confused specifically about rule.rollout_context_kind.

It makes sense that rules.clauses.context_kind targets specific context kind attributes. However, how does rule.rollout_context_kind work alongside a flag?

1. rule.rollout_context_kind , so if my segment rule matches say, person context but rollout context is set to user (also by default), does this mean the segment would evaluate false if my flag is evaluating a person context?
2. What is the point if I specify these two attributes differently? In other words, if my rule is meant for a person context but rollout context is for user, it wouldn't mean anything.
3. In doc, bucket_by says The attribute by which to group users together, so is this only applied if rule.rollout_context_kind=user? What if my rollout context kind is different? Additionally, it does not seem like to me in the UI there's an option to specify rollout by attributes?

It makes more sense to me that both of these context kinds are always the same

[ozheng-circle]

Oh also just discovered that the documentation is incorrect:

context_kind (String) The context kind associated with this rule clause. This argument is only valid if rollout_weights is also specified. If omitted, defaults to user.

but rollout_weights is not an attribute, did you mean weight?

[sloloris]

Hi @ozheng-circle!

Whether or not both of these properties (rules.clauses.context_kind and rule.rollout_context_kind) apply for you depends on how you are targeting. For segments, these specifically apply for rule-based segments. In response to your question (1), this would depend on whether you were evaluating on a person context only or whether you were evaluating on a multi-context containing both context types.

You may have rules that target on different types on contexts; for instance, you may have a flag that you want to turn on for specific users (perhaps the people who are working on it / testing it), thus targeting your rule on a user context kind, but then be rolling out your release based on organization contexts to ensure that different users within the same customer company do not receive different variations.

Similarly, if you have a rule-based segment, you may want your segment to include trial account contexts that are also in a particular AWS region (possibly another context, depending on how you have your contexts set up), and then roll out to a subset of users within that. Something like this:

For which the created rules response would look like this:

{
    "rules": [
        {
            "_id": "dba8f7f2-853c-4545-b1ca-71d3ea53ffbe",
            "clauses": [
                {
                    "_id": "2d95d8e0-5ec0-4e11-bb61-c6682384e23a",
                    "attribute": "isTrial",
                    "op": "in",
                    "values": [
                        true
                    ],
                    "contextKind": "account",
                    "negate": false
                },
                {
                    "_id": "10aee091-99a6-4f7b-98be-4cb432d7810f",
                    "attribute": "key",
                    "op": "in",
                    "values": [
                        "us-east-1"
                    ],
                    "contextKind": "aws-region",
                    "negate": false
                }
            ],
            "weight": 20000,
            "rolloutContextKind": "user",
            "description": "demo rule"
        }
    ], // etc rest of API response
}

So it's very possible that in your case (2) applies - depending on how you have your contexts set up, this may be a moot feature. Multi-contexts were only released last year so I'd imagine not all organizations are taking advantage of them yet.

I hope this helps!

Additionally, it does seem like you have discovered some errors in our docs - rollout_weights is a flag attribute, not a segment attribute. I believe that sentence was incorrectly copy-pasted from somewhere else. Additionally, the wording you draw attention to in (3) should say 'contexts' in lieu of 'users'. Thank you for drawing our attention to these! We will make sure to correct that for the next release.

I will leave this issue open for another week or two in case you have any follow-up questions and then will close :) Have a great weekend!

[ozheng-circle]

@sloloris Thank you for the thorough and timely response! You have solved my questions sufficiently. Marking this as closed.

[ozheng-circle]

Actually @sloloris I do have another question. For percentage rollout, if I evaluate with a context/multi-context that does contain my rollout context kind, what would happen for 1) the segment result 2) a flag's custom rule result?

[ozheng-circle]

And, rules.clauses.context_kind doesn't have anything to do with weight since it's specific to the rule? I believe the documentation is incorrect on that.

Another thing is in the segment UI, there doesn't appear to be an option to bucket by a rollout context's attribute.

[sloloris]

Hi @ozheng-circle ! Happy to help :)

And, rules.clauses.context_kind doesn't have anything to do with weight since it's specific to the rule? I believe the documentation is incorrect on that.

This is correct! I have removed that sentence entirely and will roll the updated documentation into a release later this week.

Another thing is in the segment UI, there doesn't appear to be an option to bucket by a rollout context's attribute.

This option will only appear for rule-based segments. Unfortunately, you have to choose the type of segment (rule-based or list-based) upon segment creation. If you are working with a list-based segment, the option will not appear.

For percentage rollout, if I evaluate with a context/multi-context that does contain my rollout context kind, what would happen for 1) the segment result 2) a flag's custom rule result?

I'm not 100% sure I've understood your question correctly, but I am guessing that in (1) you are asking how the segment with a rollout context kind set will evaluate a multi-context. In this case, I am fairly certain that it would probabilistically (depending on how far along in your rollout you are) include or exclude that multi-context, if that makes sense. We then save the configuration. For example, say you're 60% rolled out on organization kind, and you have a multi-context that includes an organization and user context. At this point, you should be rolled out to 60% of organizations, so you'd have a 60% chance of that multi-context being included in the segment. The user context is irrelevant here even though it is included in the multi-context - unless there is a prior segment rule that applies.

For (2), if you mean how a flag would evaluate with similar rules to the above, it would be pretty much the same. If you are evaluating on a segment, it would depend entirely on whether or not the context had made it into that segment. However, it's important to always keep in mind that flag and segment rules evaluate in order - a context will be evaluated by each rule in order and only if none of the rules apply to it will the default be applied. If an earlier rule puts it in a non-default variation, later rules will not apply.

This is not my area so I will follow up with someone who works on this specifically to make sure I haven't given you any false information. If this was not your question, or if you have any further questions, please reach out to our support team! They're much more well-acquainted with broad areas of the product and can probably give you more confident answers.

Have a great week!

[ozheng-circle]

Ah sorry there was a typo:

if I evaluate with a context/multi-context that does NOT contain my rollout context kind,...

This option will only appear for rule-based segments

Is there a way to explicitly set up rule-based segment in terraform? if not, is the attribute still being used for bucketing?