Proposal: Add type and unit metadata labels #39

dashpole · 2024-10-24T15:20:13Z

Some screenshots from the Prometheus UI after this PoC: prometheus/prometheus@main...dashpole:prometheus:type_and_unit_labels

Querying for `unit`

Signed-off-by: David Ashpole <[email protected]>

ArthurSens

Great start! I think we need to discuss a bit more whether to handle __* labels in the PromQL engine or the UI. Maybe there's a chance we're doing a breaking change...?

proposals/2024-09-25_metadata-labels.md

ArthurSens · 2024-10-28T18:40:09Z

proposals/2024-09-25_metadata-labels.md

+* Users might be surprised by, or dislike the additional suffixes and delimiters in the metric name results
+  * Mitigation: Opt-in for query engines?
+
+### "Hide" __type__ and __unit__ labels in PromQL, instead of UI


I wonder how safe this is. If PromQL does not hide the return, new labels will suddenly appear in users' queries.

I'm also not sure about the potential impact other than dashboards/alerts having new labels, could that be a problem?

I'm definitely open to this alternative. I like having the option of showing __type__ and __unit__ in the UI, but if it has to wait for Prom 4.0 because it is breaking, that seems like a significant downside.

Well, it could be also merge of both, where PromQL only returns those labels based on something e.g. on conflict or when type or unit is selectors.

The downside is that we now have to really deeply design groupings and joins.. should they use those labels or not. With the current proposal the answer is trivial and straightforward. This helps with creating alerts and recording rules etc

I think I would be keen to try the current approach e.g. behind feature flag and see who will use it and how to tell more. cc @juliusv @roidelapluie WDYT from UX point of view for non-UI things?

proposals/2024-09-25_metadata-labels.md

Signed-off-by: David Ashpole <[email protected]>

beorn7

Thank you very much for proposing this. I love it in general, and as said elsewhere, I had a very similar idea a while ago for completely different reasons (mostly to avoid mixed series with native histograms and float-based sample types). Finding a similar solution for unrelated problems seems like a signal that this could be actually useful. However, I have also run into some conundrums, which was the reason why I put the whole idea on the backburner. I'll try to sketch out my train of thought here:

The __name__ label is the precedent here for "special labels", and as such it gives us a hint of what the issues might be. It's special power is not just about how to display it. It is treated specially for label matching and aggregation, and it is removed by most operations, following the argument that a rate of process_cpu_seconds_total is not process_cpu_seconds_total anymore. We have to ask ourselves the same questions for __unit__ and __type__.

In the brave new world, process_cpu_seconds_total would probably look like process_cpu{__type__="counter", __unit__="seconds"}. If we rate that, the outcome is not a counter anymore (but a gauge), and it is actually unit-less (seconds per second, i.e. it is the CPU usage ratio. In wishful thinking, the outcome should be process_cpu_usage{__type__="gauge", __unit__="ratio"}. Of course, we cannot easily come up with a general procedure to make up new names (which is the reason why current PromQL simply removes the name), and even changing the unit is tough. Changing the type might actually be feasible (we explored that a bit with native histograms, which "know" if they are a counter histogram or a gauge histogram). So maybe we can come up with "type translation rules" for all PromQL operations, and then maybe drop the unit in doubt (when calculating a rate or multiplying) and keep it when it makes sense (a sum aggregation or adding). But you will notice how deep we are in the woods here already.

The next thing is aggregation and label matching. As said, __name__ is also special here, but we probably need to be special in different ways for __unit__ and __type__. For example if we simply do a + b, we probably want to include unit and type in the label match (only add gauges to gauges, only add seconds to seconds etc.). However, a * b is already very different. b might be a scaling factor, i.e. a unit-less gauge. a could be counters or histograms. So in this case, we would like to exclude the unit and type from the label match.

It gets very confusing quickly. I wish we could just add the __unit__ and __type__ as an experiment without further special treatment and see how people cope with it. But right now my worry is it will hit too many road bumps…

beorn7 · 2024-11-07T13:53:48Z

proposals/2024-09-25_metadata-labels.md

+
+* [Required] OpenTelemetry users can query for the original names of their metrics.
+* [Required] Users can filter by the metric type or unit in PromQL queries.
+* [Required] PromQL returns a warning when querying across metrics with different units or types.


And PromQL could finally warn reliably about "inappropriate" operations (apply rate to a gauge, delta to a counter, …). (Recently, Prometheus has added an info-level annotation for cases where rate is applied to a metric that doesn't end on _total etc., but that's error prone (gauges could have a name ending on _total, and not all counters end on _total).)

By making unit and type a label (and thus part of the series' identity) it becomes implicitly harder to accidentally query across metrics with different unit or type anyway (there aren't any "mixed series" anymore). Maybe that is something we could work into the goals as preventing the mismatch from happening in the first place is a much more attractive selling point than just stating that we can "return a warning".

I'd say that's a happy side effect of the proposal, but not a requirement :)

Warning about inappropriate operations sgtm. Added to goals.

In the context of preventing collisions from happening in the first place, i've been thinking about what best practices should be. If users write queries and always include unit and type filters in queries, it is impossible for mismatch to happen in the first place. It also makes queries more readable, as you can easily tell the type and unit of the metric when reading a query. But it does seem quite verbose, and I'm pessimistic that users would actually do this if we tell them they should.

beorn7 · 2024-11-07T14:01:12Z

proposals/2024-09-25_metadata-labels.md

+
+When a query for a metric returns multiple metrics with a different `__type__` or `__unit__` label, but the same `__name__`, users see a warning in the UI.
+
+Users don't see the `__type__` or `__unit__` labels in the Prometheus UI next to other labels, but the unit displayed next to the value.


And maybe the type next to the metric name and labels, but as a colored badge or something (cf. how it's done in the PGW, which utilizes the full exposition data model so it even has the help string, which is of course out of scope here).

proposals/2024-09-25_metadata-labels.md

beorn7 · 2024-11-07T14:20:12Z

proposals/2024-09-25_metadata-labels.md

+* Doing this would require some changes to client libraries, which is significantly more work, and is harder to experiment with.
+* There can be conflicts between `# TYPE` and `# UNIT` metadata for a metric, and `__type__` and `__unit__` labels. This adds complexity to understanding the exposition format, and requires establishing rules for dealing with conflicts.
+
+### More metadata labels


Generally, I like the idea to model OTel's notion of "identifying metadata" as labels in Prometheus (to be distinguished from "non-identifying metadata", which is closer to Prometheus's original metadata idea).

Unfortunately OTel resource attributes are all considered identifying today (correct me if I'm wrong David).

There is work in progress to replace resource attributes with a new thing called Entity. Entity will be able to distinguish, using otel lingo, Identifying from Descriptive attributes, but this will take some years to get popular in the industry

OTel resource attributes are all considered identifying today, but we don't treat them all as identifying when translating from OTel -> Prometheus. Instead, we just treat service.name + service.instance.id + service.namespace as identifying. That is correct if the OTLP originates from an OTel-instrumented application, but not if it wasn't (e.g. comes from an otel collector receiver).

Anyways, what I had in mind was more things that users don't think of as "labels", like schema url, or maybe scope name + version. I think resource attributes make sense as "real" labels, rather than metadata.

beorn7 · 2024-11-12T10:37:40Z

It gets very confusing quickly. I wish we could just add the __unit__ and __type__ as an experiment without further special treatment and see how people cope with it. But right now my worry is it will hit too many road bumps…

Still in brainstorming mode, here is an idea how maybe a step zero could look like: Add the __unit__ and __type__ upon ingestion, but mostly ignore them in PromQL operations, maybe with the following exceptions:

__unit__ and __type__ can be used in label selectors. This allows the filtering (listed as a goal in this doc) and could also be used to increase readability of a PromQL expression.
__unit__ and __type__ can still be added via label_replace, e.g. to improve display of a result or to give the result of recording rules the unit and type information.

However, aggregations and label matches would ignore __unit__ and __type__ and would therefore work as before, and any operation would remove the __unit__ and __type__ label (with the exception of label_replace), both meant to circumvent any of the issues laid out above.

From there, we could cautiously move forward with more ideas. A step one might then be to handle __type__ in a more "native" fashion, i.e. attaching the "correct" __type__ label depending on the operation, still allowing the user to override via label_replace (like hard-casting a type in C), but maybe enforce a histogram type on native histograms (and prevent such a type on floats), so that we would avoid mixed series. Also, taking __type__ into account during aggregations might be possible.

Signed-off-by: David Ashpole <[email protected]>

bwplotka · 2024-11-12T21:09:49Z

unit and type can still be added via label_replace, e.g. to improve display of a result or to give the result of recording rules the unit and type information.

Hm, I think it would be also nice to display them when collision actually occurs 🤔

Signed-off-by: David Ashpole <[email protected]>

bwplotka

Nice work! Looking good, some comments.

proposals/2024-09-25_metadata-labels.md

bwplotka · 2024-11-25T15:11:47Z

proposals/2024-09-25_metadata-labels.md

+
+When a query for a metric returns multiple metrics with a different `__type__` or `__unit__` label, but the same `__name__`, users see a warning in the UI.
+
+Users don't see the `__type__` or `__unit__` labels in the Prometheus UI next to other labels, but the unit displayed next to the value.


Hm, so this is UI, but what we envision in PromQL response? Should it give those labels back in JSON? If yes, in what order?

This goes back to TSDB API too. I assume TSDB API provide those labels as normal labels (worth to clarify?)

Ok, the alternative explains it, should we mention that both API and PromQL contains those labels as-is?

I'm not familiar with the TSDB API. Can you link to it?

proposals/2024-09-25_metadata-labels.md

bwplotka · 2024-11-25T15:19:19Z

proposals/2024-09-25_metadata-labels.md

+
+### PromQL Changes
+
+When a query for a metric returns multiple metrics with a different `__type__` or `__unit__` label, but the same `__name__`, an info annotation will be returned with the PromQL response, which is otherwise unmodified.


Unmodified means, when TSDB storage gives you series with __type__ and __unit__ it does NOT strips/remotes those, right?

Correct. I'm trying to say that this is "just a warning", and doesn't impact results at all.

bwplotka · 2024-11-25T15:24:17Z

proposals/2024-09-25_metadata-labels.md

+
+### Prometheus Server Ingestion
+
+When receiving OTLP or PRW 2.0, or when scraping the text, OM, or proto formats, the type and unit of the metric are added as the `__type__` and `__unit__` labels, if not already present.


For exposition formats, I guess it means this would allow:

TYPE metric counter UNIT metric seconds HELP ... metric{__type__="gauge"}

Ideally I do opposite to what's written, so TYPE and UNIT wins here, or this could be blocked even by spec. Essentially I wonder about efficiency of parsers here.

For ingestion protocols, at least it feels metadata should override it (not what is written now) so it's easier for everyone.

Any further concerns around perf of parsing here for this @bboreham?

I was thinking about what would be on the /federate endpoint if the Prometheus server scraped conflicting metrics from applications. Our choices are:

[current] Only #UNIT metadata, and no __unit__ labels.

Only __unit__ labels, and no #UNIT metadata.

[proposal] Mixed __unit__ labels, and a #UNIT with one of the units the metric has.

1 will result in collisions, but maybe that is acceptable for the /federate endpoint. It isn't a regression from what we have today.

2 works, but would be breaking for existing users that are consuming #UNIT.

3 is confusing because of the mixing, but I wanted to leave that open as an option.

I'm OK with disallowing mixing __unit__ and #UNIT, since we could relax that in the future if needed.

We discussed offline, and:

would epic to follow 1 and allow multiple the same named metric for different types and units in OM, let's try proposing that, it would help massively with exporters and federate

type and unit from metadata is primary (in case of mix)

Should we allow type and unit at all in exposition, SHOULD NOT maybe?

I'm not too fond of adding __unit__ and __type__ as labels in the exposition format. I think they should stay as a single line at the top of the metric family like it is today.

The Unit/Type will be the same for all time series in a metric family. If they become labels in the exposition format, it will just be repeated information for all time series. It increases the payload, and we waste CPU time parsing it.

We have "Users see no difference to exposition formats." stated above, but if you put the type and especially the unit in the # UNIT part then you need them in the metric name as well, defeating the purpose. You're other option is to put them into the series on each line.

Looking at /metrics is for humans and follows what you see is what you get in terms of showing what you can query - implying that if you can query by __type__ or the API/UI returns the __type__ then you should put them on each line.

And the implication of that is that there needs to be at least one more efficient exposition format that is not so verbose for machines to read.

So I'm not sure this would work without either putting the new labels on every line or modifying OpenMetrics.

Of course the above argument doesn't stand if we follow @beorn7 's advice and not show __type__, __unit__ on the query API (/query , /query_range) , although I would love to have it on the metadata queries endpoint like (/series) for Grafana to be able to know what's a float series and what's a native histogram.

I think we should show __type__ and __unit__ in the query APIs. I'm just saying, in the zeroth step, we should remove those labels in PromQL operations (like the name is removed).

I agree with showing __type__ and __unit__ in query APIs, just like we show __name__.

We have "Users see no difference to exposition formats." stated above, but if you put the type and especially the unit in the # UNIT part then you need them in the metric name as well, defeating the purpose. You're other option is to put them into the series on each line.

That's a really good catch! This is something we'll probably need to tackle with OM 2.0. Is this what you had in mind with this issue @bwplotka prometheus/OpenMetrics#280?

bwplotka · 2024-11-25T16:02:51Z

proposals/2024-09-25_metadata-labels.md

+* Users might be surprised by, or dislike the additional suffixes and delimiters in the metric name results
+  * Mitigation: Opt-in for query engines?
+
+### "Hide" __type__ and __unit__ labels in PromQL, instead of UI


Well, it could be also merge of both, where PromQL only returns those labels based on something e.g. on conflict or when type or unit is selectors.

The downside is that we now have to really deeply design groupings and joins.. should they use those labels or not. With the current proposal the answer is trivial and straightforward. This helps with creating alerts and recording rules etc

I think I would be keen to try the current approach e.g. behind feature flag and see who will use it and how to tell more. cc @juliusv @roidelapluie WDYT from UX point of view for non-UI things?

Signed-off-by: David Ashpole <[email protected]>

dashpole · 2024-11-26T03:47:47Z

@beorn7 I've added "Aggregations and label matches ignore __unit__ and __type__ and any operation would remove the __unit__ and __type__ label (with the exception of label_replace)." based on your suggestion above. I've also added "Handle type and unit in PromQL operations" as a potential future extension.

beorn7 · 2024-11-26T15:46:05Z

Quick note about federation: To my knowledge, federation still completely ignores metadata, so the exposition format it generates has all metric types as "untyped" and no units and help strings anyway, simply because Prometheus doesn't know about metadata post-ingestion. So being able to funnel unit and type through the Prometheus TSDB via labels would be a direct improvement for federation.

ArthurSens · 2024-11-26T22:54:11Z

Quick note about federation: To my knowledge, federation still completely ignores metadata, so the exposition format it generates has all metric types as "untyped" and no units and help strings anyway, simply because Prometheus doesn't know about metadata post-ingestion. So being able to funnel unit and type through the Prometheus TSDB via labels would be a direct improvement for federation.

I think it would be better to enable metadata post-scrape/federation than adding type and unit as labels in the exposition format. See #39 (comment)

beorn7 · 2024-11-27T11:58:02Z

Quick note about federation: To my knowledge, federation still completely ignores metadata, so the exposition format it generates has all metric types as "untyped" and no units and help strings anyway, simply because Prometheus doesn't know about metadata post-ingestion. So being able to funnel unit and type through the Prometheus TSDB via labels would be a direct improvement for federation.

I think it would be better to enable metadata post-scrape/federation than adding type and unit as labels in the exposition format. See #39 (comment)

The problem is that federation breaks the assumption that all metrics of the same name (one metric family) all have the same metadata.

bwplotka · 2024-11-27T12:57:13Z

The problem is that federation breaks the assumption that all metrics of the same name (one metric family) all have the same metadata.

Yea, I wonder if we could lift this with OM 2.0. WDYT @beorn7 ?

Related discussion: #39 (comment)

beorn7 · 2024-11-27T13:28:05Z

The problem is that federation breaks the assumption that all metrics of the same name (one metric family) all have the same metadata.

Yea, I wonder if we could lift this with OM 2.0. WDYT @beorn7 ?

The current structure of both protobuf and text versions of OM doesn't really lend itself to multiple different type, unit, or help entries for metrics with the same name. So we needed a new structural way of specifying unit and type (let's ignore help for now) per metric (rather than per metric family). Ironically, we already have a place for things that are per metric. We call it labels. So let's put type and metric into labels? Wait… that's what this proposal actually proposes. 😁

bwplotka · 2024-11-27T15:06:45Z

Well yes, but we just discussed that type and unit per label will not be easily parsable in OM/text, so putting it somewhere else actually helps. There might be a big questions there:

Is Metric Family concept actually useful these days? Should OM 2.0 simply redefine it, and flat that out? Personally it's only confusing, we don't use this semantic on the storage or PromQL (AFAIK). 💣
What's wrong to keep metric families but make them unique per type and unit? (not only name)?

beorn7 · 2024-11-27T16:23:39Z

Well yes, but we just discussed that type and unit per label will not be easily parsable in OM/text, so putting it somewhere else actually helps.

I wouldn't say that this is actually the case. The way we parse the labels doesn't really create a lot of parsing cost if there are more labels. It does create more payload, but not so much in relative terms. The OM text format is designed to have a lot of repetitive information anyway. If we want to avoid repetitive information, we needed a fundamentally different structure.

Is Metric Family concept actually useful these days? Should OM 2.0 simply redefine it, and flat that out? Personally it's only confusing, we don't use this semantic on the storage or PromQL (AFAIK). 💣

It correctly models the per-scrape data model. It avoids (a part of the) repetition. Of course you can flat that out, but then you get the repetition that was marked above as "not easily parseable". My claim is that "flattening that out" is pretty much equivalent to "adding that as labels".

What's wrong to keep metric families but make them unique per type and unit? (not only name)?

Well, the text format originally was designed to not be ordered in any way. So every line was keyed by the metric family name, i.e. a metric line starts with the metric family name, and all the metadata lines contain the metric family name as the 2nd token after the hash. OM became somewhat dependent on order, but not consequently so (you could avoid a lot of repetition if you did that) (another of the design problems I see with OM). With the original idea of the text format structure, you needed to repeat not just the metric family name on each line, but the combination of metric family name, type, and unit (which, you guessed it, is equivalent to putting type and unit into a label). Alternatively, we could redesign the text format fundamentally and make it really depend on the order everywhere, but then we should do it thoroughly and avoid other repetitions, too (at the very least the metric family name).

Protobuf is a bit different as it is structured, and all the family members are a repeated Metric message inside a MetricFamily message. So keying on more than just the name is relatively easy to accomplish. This is another reason why I think we should "simply" solve the protobuf parsing performance issue and then go back to "if you want efficiency, use protobuf – if you want simplicity, use text" (simplicity as in "easy to create in a valid form", which means kicking out many of the requirements OM text introduced on top of the original text format, like order, whitespace, …).

bwplotka · 2024-11-28T13:10:15Z

The OM text format is designed to have a lot of repetitive information anyway. If we want to avoid repetitive information, we needed a fundamentally different structure.

I thought it's more the fact we usually need to know what's the metric type (and metric family name) ahead of time (e.g. histograms have a different flow). But you are right, perhaps we could make label based approach work fine. Then it's only human readability question and what's defined in SDK and what's queried.

It correctly models the per-scrape data model.

I think I got it now, FAIK this only makes a difference for unstructured types like classic histogram and summary where metric name you define in SDKs (aka metric family name) is != resulting metric name. In the word with native histograms and perhaps native summaries one day, it would make no different, right?

My claim is that "flattening that out" is pretty much equivalent to "adding that as labels".

Agree.

Well, the text format originally was designed to not be ordered in any way. So every line was keyed by the metric family name, i.e. a metric line starts with the metric family name, and all the metadata lines contain the metric family name as the 2nd token after the hash. OM became somewhat dependent on order, but not consequently so (you could avoid a lot of repetition if you did that) (another of the design problems I see with OM). With the original idea of the text format structure, you needed to repeat not just the metric family name on each line, but the combination of metric family name, type, and unit (which, you guessed it, is equivalent to putting type and unit into a label). Alternatively, we could redesign the text format fundamentally and make it really depend on the order everywhere, but then we should do it thoroughly and avoid other repetitions, too (at the very least the metric family name).

Got it, thank you for detailed explanation! I think we should discuss those options in a clear proposal, make some decision on this in OM 2.0, will add issue for this.

Protobuf is a bit different as it is structured, and all the family members are a repeated Metric message inside a MetricFamily message. So keying on more than just the name is relatively easy to accomplish. This is another reason why I think we should "simply" solve the prometheus/prometheus#14668 and then go back to "if you want efficiency, use protobuf – if you want simplicity, use text" (simplicity as in "easy to create in a valid form", which means kicking out many of the requirements OM text introduced on top of the original text format, like order, whitespace, …).

Nice! Proposing as an intention in our OM 2.0 doc

bwplotka · 2024-11-28T14:25:26Z

Created related issue on OM 2.0 prometheus/OpenMetrics#280

krajorama · 2024-11-28T17:28:42Z

proposals/2024-09-25_metadata-labels.md

+
+For example:
+
+* `my_metric{}` returns all series with any type or unit, including `__type__` and `__unit__` labels.


On the API I assume?

krajorama · 2024-11-28T17:28:55Z

proposals/2024-09-25_metadata-labels.md

+For example:
+
+* `my_metric{}` returns all series with any type or unit, including `__type__` and `__unit__` labels.
+* `my_metric{__unit__="seconds", __type__="counter"}` returns only series with the specified type and unit.


Both API and UI I assume?

krajorama · 2024-11-28T17:46:55Z

proposals/2024-09-25_metadata-labels.md

+
+### Prometheus Server Ingestion
+
+When receiving OTLP or PRW 2.0, or when scraping the text, OM, or proto formats, the type and unit of the metric are added as the `__type__` and `__unit__` labels, if not already present.


We have "Users see no difference to exposition formats." stated above, but if you put the type and especially the unit in the # UNIT part then you need them in the metric name as well, defeating the purpose. You're other option is to put them into the series on each line.

Looking at /metrics is for humans and follows what you see is what you get in terms of showing what you can query - implying that if you can query by __type__ or the API/UI returns the __type__ then you should put them on each line.

And the implication of that is that there needs to be at least one more efficient exposition format that is not so verbose for machines to read.

So I'm not sure this would work without either putting the new labels on every line or modifying OpenMetrics.

Proposal for type and unit metadata labels

25305b0

Signed-off-by: David Ashpole <[email protected]>

dashpole force-pushed the type_and_unit_labels branch from 8aab401 to 25305b0 Compare October 24, 2024 15:21

ArthurSens reviewed Oct 28, 2024

View reviewed changes

address feedback

e56b2be

Signed-off-by: David Ashpole <[email protected]>

dashpole force-pushed the type_and_unit_labels branch from 98a86bd to e56b2be Compare November 1, 2024 15:35

beorn7 reviewed Nov 7, 2024

View reviewed changes

dashpole added 2 commits November 12, 2024 20:49

add goal of warning on inappropriate operations

6cfdbd4

Signed-off-by: David Ashpole <[email protected]>

info annotation, rather than warning

de68600

Signed-off-by: David Ashpole <[email protected]>

remove dmention of unit in text format

e0d62d3

Signed-off-by: David Ashpole <[email protected]>

alexgreenbank mentioned this pull request Nov 19, 2024

Return UNIT information alongside queries prometheus/prometheus#12294

Open

bwplotka reviewed Nov 25, 2024

View reviewed changes

dashpole added 2 commits November 25, 2024 21:43

add native histogram motivation, and omit unit alternative

619537d

Signed-off-by: David Ashpole <[email protected]>

rename feature flag to identifying-type-and-unit

a2e5ed9

Signed-off-by: David Ashpole <[email protected]>

dashpole force-pushed the type_and_unit_labels branch from 15b0c44 to a2e5ed9 Compare November 26, 2024 02:52

dashpole added 4 commits November 26, 2024 03:12

add complex value alternative

75ccddd

Signed-off-by: David Ashpole <[email protected]>

add ability to relabel type and unit

d9e0ad3

Signed-off-by: David Ashpole <[email protected]>

preserve existing behavior for PromQL operations

f1760bd

Signed-off-by: David Ashpole <[email protected]>

add future extension to handle type and unit in promql operations

bce6d9c

Signed-off-by: David Ashpole <[email protected]>

dashpole force-pushed the type_and_unit_labels branch from d4ecde7 to bce6d9c Compare November 26, 2024 03:46

bwplotka mentioned this pull request Nov 28, 2024

OM 2.0: Decision around __type__ and __unit__ labels in Text format prometheus/OpenMetrics#280

Open

krajorama reviewed Nov 28, 2024

View reviewed changes


		When a query for a metric returns multiple metrics with a different `__type__` or `__unit__` label, but the same `__name__`, users see a warning in the UI.

		Users don't see the `__type__` or `__unit__` labels in the Prometheus UI next to other labels, but the unit displayed next to the value.


		### PromQL Changes

		When a query for a metric returns multiple metrics with a different `__type__` or `__unit__` label, but the same `__name__`, an info annotation will be returned with the PromQL response, which is otherwise unmodified.


		### Prometheus Server Ingestion

		When receiving OTLP or PRW 2.0, or when scraping the text, OM, or proto formats, the type and unit of the metric are added as the `__type__` and `__unit__` labels, if not already present.


		For example:

		* `my_metric{}` returns all series with any type or unit, including `__type__` and `__unit__` labels.

Proposal: Add type and unit metadata labels #39

Are you sure you want to change the base?

Proposal: Add type and unit metadata labels #39

Conversation

dashpole commented Oct 24, 2024 • edited Loading

Querying for __unit__

ArthurSens 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

beorn7 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ArthurSens Nov 7, 2024 • edited Loading

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

beorn7 commented Nov 12, 2024

bwplotka commented Nov 12, 2024

bwplotka 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

bwplotka Nov 25, 2024 • edited by dashpole Loading

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

ArthurSens Nov 26, 2024 • edited Loading

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

dashpole commented Nov 26, 2024

beorn7 commented Nov 26, 2024

ArthurSens commented Nov 26, 2024

beorn7 commented Nov 27, 2024

bwplotka commented Nov 27, 2024

beorn7 commented Nov 27, 2024 • edited Loading

bwplotka commented Nov 27, 2024 • edited Loading

beorn7 commented Nov 27, 2024

bwplotka commented Nov 28, 2024 • edited Loading

bwplotka commented Nov 28, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dashpole commented Oct 24, 2024 •

edited

Loading

Querying for `unit`

ArthurSens Nov 7, 2024 •

edited

Loading

bwplotka Nov 25, 2024 •

edited by dashpole

Loading

ArthurSens Nov 26, 2024 •

edited

Loading

beorn7 commented Nov 27, 2024 •

edited

Loading

bwplotka commented Nov 27, 2024 •

edited

Loading

bwplotka commented Nov 28, 2024 •

edited

Loading