Multi-valued arguments

[rashidkpc]

Currently support for multi-valued arguments to functions are handled independently by the receiving function. This requires users to pass a specially formatted string, for which there is currently a basic convention:

.es(metric="cardinality:user")

This works well enough for simple key-value situations. But it falls apart for more complex cases. For example, the cardinalitymetric above can take a precision_threshold parameter. How do we handle that?

It would be easy to add a 3rd parameter, eg cardinality:user:200, but this won't be shared metric aggs, and all of the arguments will be positional. Gross.

I could add an alternate form allows for an opts function that generates an object to be passed into complex arguments

.es(metric=.opts(type=cardinality, field=user, precision_threshold=200))

Pros

• Super flexible
• Does not introduce new language constructs
• Can use existing parser, would only require the ability to define arguments that don't have positionally.
• Could use it for multi-value aggs. Eg .es(metric=.opts(type=percentiles, field=bytes per=1, per=10, per=50, per=90))

Cons

• .opts() is essentially a datasource, but it only does anything if passed as an argument somewhere that makes sense, and there's nothing you'd chain to it to.
• .opts() would need to be excluded from autocomplete. This is the easy way to handle it. In reality, better context support in the autocompletion would be ideal.
• Auto completion for .opts() will be really hard

I think the key here is that this would really need to be optional, as it creates some really long arguments

[yuvalzaltz]

I like the opts option which provides max flex.
Autocompletion is a problem but if it would be documented in docs i guess it would be good enough.

[rashidkpc]

The more I think about this the more I think an improved expression editor is a pre-req

[eprothro]

+1 for this problem being solved. As a beginner I did:

As a beginner I tried .es(metric=percentiles:service) and saw an empty chart. Realizing I needed to provide options, I expected something like:

.es(metric=.metric(service, percentiles, percents=[68,95,99]))

[v01t]

+1
Missing percentiles metric, expect to have it before... by this missing functionality all latency calculations become very limited (max,min,avg)

Please add it.

Thanks for your great job.

[daveres]

+1

[tommikiviniemi-srs]

This would be hugely appreciated by us. +1.