OpenTelemetry Governance Panel - Reese Lee, Juraci Paixao Krohling, Alolita Sharma & Daniel Dyla
Summary
TLDRThe video script features a panel discussion with members of the OpenTelemetry governance committee, who introduce themselves and their roles. They delve into the committee's responsibilities, including managing product and feature backlogs, ensuring contributor and developer experience, and fostering community growth. The conversation also touches on the project's evolution, the addition of new signals, and the balance between supporting various programming languages and idiomatic APIs. The panelists emphasize the ongoing nature of OpenTelemetry's development and the importance of community involvement in shaping its future.
Takeaways
- π The OpenTelemetry governance committee (GC) is composed of various contributors who manage the project's direction, including product backlog and feature requests.
- π₯ The GC members wear multiple hats, acting as representatives of the project to the greater community and handling external relations and media interactions.
- π§ The committee is focused on improving contributor and developer experience, supporting maintainers, and ensuring the health of the project.
- π OpenTelemetry is not a monolithic project; it is composable, with a focus on being a de facto standard for observability in cloud-native environments.
- π There is a continuous effort to expand OpenTelemetry beyond traces, metrics, and logs to include new areas like profiling, client telemetry, and machine learning.
- π The GC is in the process of establishing a more structured project management approach, including the concept of 'projects' that need sponsors and clear active collaboration.
- π OpenTelemetry is designed to be vendor-neutral, allowing for changes in solutions while maintaining interoperability through the OpenTelemetry Protocol (OTLP) and semantic conventions.
- π The committee is keen on fostering a community that can absorb new requirements and technologies, such as those arising from AI applications and new hardware like GPUs.
- π There is an emphasis on creating a clear path for subject matter experts to contribute and eventually become spec approvers or even members of the technical committee (TC).
- π The GC is working on making the project more accessible and convenient for new contributors, with a new developer experience Sig being initiated to streamline APIs and installation processes.
- π The key to the success of new Special Interest Groups (SIGs) within OpenTelemetry is having clear goals, committed leads, and sponsorship from the GC.
Q & A
What is the primary role of the Governance Committee (GC) in the Open Telemetry project?
-The GC is responsible for overseeing the health of the Open Telemetry project. It manages the product and feature backlog, ensures contributor and developer experience concerns are addressed, supports maintainers, and acts as a strategic representative of the project to the greater Cloud native community.
What does the term 'Sig' refer to in the context of the Open Telemetry project?
-In the script, 'Sig' refers to a Special Interest Group, which is a subgroup within the Open Telemetry community that focuses on specific areas or aspects of the project.
How often is the Governance Committee elected by the community?
-The Governance Committee is elected by the community every two years.
What is the significance of having a project management structure in the Open Telemetry project as it grows?
-As the project grows, having a project management structure helps in organizing work more efficiently, ensuring that new areas of work are properly sponsored and that there is clear active collaboration among subject matter experts and maintainers.
What is the purpose of the liaison program mentioned by the GC?
-The liaison program aims to establish a line of communication between the Special Interest Groups (SIGs) and the GC, ensuring information flow and supporting maintainers more effectively.
How does the GC plan to engage with new contributors to the Open Telemetry project?
-The GC plans to reach out to new contributors more directly, acknowledging their contributions, and providing support to help them contribute further to the project.
What is the relationship between the Open Telemetry project and the Cloud Native Computing Foundation (CNCF)?
-The Open Telemetry project is part of the CNCF ecosystem and receives support in terms of resources, cloud credits, and promotional efforts from the CNCF and its member companies.
What is the current status of the Open Telemetry project in terms of its development focus on traces, metrics, and logs?
-While traces, metrics, and logs have been the primary focus for several years, the project is now considering expanding into other areas such as profiling, client telemetry, and machine learning, while still addressing any backlog in the original focus areas.
What is the process for deciding whether to open up a new SIG or introduce a new signal in the Open Telemetry project?
-The process involves creating a proposal in the form of a PR in the community repository, outlining the goals, leads, sponsors, and timeline. The GC reviews the proposal and assesses the availability of resources and the potential for the new SIG to succeed before making a decision.
How does the Open Telemetry project approach the balance between supporting different programming languages and adhering to language idioms?
-The project aims to strike a balance where the API feels consistent across languages for those who switch between them, while also being idiomatic to each language so that it feels natural to developers who have been using a particular language for a long time.
What is the significance of semantic conventions in the Open Telemetry project?
-Semantic conventions are important as they provide the necessary context and meaning to the telemetry data, enabling better storage, querying, and understanding by end users, which is crucial for effective observability.
Outlines
π£οΈ Introductions and Governance Committee Roles
The paragraph introduces various members of the Open Telemetry governance committee and their roles in different companies and projects. Rey, a senior developer relations engineer at New Relic, initiates the introductions and mentions the governance committee's purpose. Dan from Dynatrace, Austin Parker from Honeycomb, Judan King from Graphon Labs, Ted from the Open Tracing side, and Alita Sharma from Apple all introduce themselves, highlighting their contributions to Open Telemetry and related projects. The paragraph emphasizes the committee's responsibility for the project's health, handling product management, feature backlogs, and addressing contributor concerns.
π οΈ Governance Committee's Structure and Community Engagement
This section delves into the governance committee's evolving structure due to project growth, introducing the concept of 'projects' within Open Telemetry. It discusses the need for project sponsors, the importance of collaboration with subject matter experts, and the governance committee's role in facilitating community engagement. The paragraph also mentions the beta testing of community projects, the liaison program for better communication between the governance committee and maintainers, and the initiative to reach out to new contributors, reinforcing the committee's commitment to the project's health and community involvement.
π External Representation and Strategic Initiatives
The paragraph highlights the governance committee's role in representing Open Telemetry to the broader cloud-native community. It covers the committee's involvement in securing resources such as cloud credits and handling interactions with the media to promote the project. The members discuss their strategic role in engaging with large organizations to develop Open Telemetry strategies, emphasizing the importance of advocacy for the project and open-source observability within the cloud-native ecosystem.
π Future Directions and Continuous Improvement
The discussion in this paragraph revolves around the future of Open Telemetry, addressing the question of when the project might be considered 'done.' Panelists share insights on the continuous nature of open-source projects, the need for evolving standards, and the integration of new signals like profiling and machine learning. They emphasize the importance of community involvement in expanding the project's scope and improving the developer and end-user experience, acknowledging that the project will never be truly 'done' due to ever-evolving technology and requirements.
π New Observability Requirements and AI Integration
This section focuses on the emerging observability requirements brought by AI applications and the foundational work needed to support these new hardware and software developments. The panelists discuss the importance of Open Telemetry Protocol (OTP) and standardized metrics, tracing, and profiling in the context of GPUs and other advanced computational resources. They invite more contributors to join and support the new generation of applications, emphasizing the potential for Open Telemetry to adapt and grow with technological advancements.
π§ Developer Experience and Signal Prioritization
The paragraph discusses efforts to improve developer experience and end-user adoption, as well as the process for deciding when to open a new Special Interest Group (SIG) or introduce a new signal. The panelists describe the project process and template for proposing new SIGs, emphasizing the importance of having clear goals, leads, sponsors, and timelines. They also touch on the challenge of balancing the desire to experiment with the need for structured development and the committee's cautious approach to avoid spreading themselves too thin.
π Semantic Conventions and Signal Evolution
This section explores the concept of semantic telemetry and the importance of aligning new data types with existing signals in observability. The panelists discuss the evolution of semantic conventions, the challenge of instrumenting across different layers of technology, and the need for collaboration to maintain a baseline on protocol standardization. They also address the question of when to add new signals, emphasizing the importance of widespread adoption and the need for strict data structure guarantees for analysis.
ποΈ Balancing Language Implementations with Open Telemetry
The paragraph examines the challenge of creating a specification that is broad enough to accommodate various language environments while maintaining consistency across implementations. The panelists discuss the spec's generality and the interpretation allowed for language maintainers, the importance of balancing language-specific idioms with a consistent experience across languages, and the role of the Developer Experience SIG in addressing these concerns. They also highlight the importance of the Open Telemetry community's feedback in refining the API and spec.
π€ Integrating Expertise and TC Development
This section discusses the integration of subject matter experts into the Open Telemetry project and the development of the Technical Committee (TC). The panelists emphasize the importance of collaboration between experts and Open Telemetry architects, the need for a clear path to the TC, and the desire for the TC to include experts in every signal area. They also address the importance of feedback and the project's commitment to ensuring that all voices are heard and integrated effectively, regardless of whether individuals hold a formal TC position.
π Conclusion and Call for Continued Engagement
The final paragraph wraps up the discussion by emphasizing the ongoing nature of the project and the desire for continued community engagement. The panelists invite participants to connect with the governance committee and TC members during breaks or lunch, highlighting an upcoming panel with TC members. They stress the importance of feedback and the project's commitment to fostering an environment where anyone can have an impact, regardless of their position within the project.
Mindmap
Keywords
π‘Open Telemetry
π‘Governance Committee (GC)
π‘Technical Committee (TC)
π‘Semantic Conventions
π‘SIG (Special Interest Group)
π‘Observability
π‘Metrics
π‘Traces
π‘Profiling
π‘Context Propagation
π‘Vendor Neutrality
Highlights
Introduction of the governance committee members and their roles within the OpenTelemetry project.
Discussion on the responsibilities of the governance committee, including managing product and feature backlogs.
The importance of addressing contributor and developer experience concerns within the OpenTelemetry community.
The concept of project management and the need for structure as the OpenTelemetry project grows.
The role of the governance committee in sponsoring new projects and facilitating collaboration between subject matter experts.
Initiation of a GC liaison program to improve communication between Special Interest Groups (SIGs) and the governance committee.
Plans to reach out to new contributors more directly to acknowledge their contributions and support their integration into the project.
The governance committee's ultimate responsibility for the health of the OpenTelemetry project and fostering a supportive community.
The strategic representation of OpenTelemetry to the greater Cloud Native community and handling of external resources.
The interface role of the governance committee with the media to promote OpenTelemetry and open source observability.
The continuous evolution of OpenTelemetry and the challenges of determining when the project is 'done'.
The potential for new signals in OpenTelemetry, such as profiling and machine learning observability, and the criteria for their inclusion.
The balance between supporting different programming languages and maintaining idiomatic APIs in OpenTelemetry.
The developer experience SIG's focus on improving the convenience and accessibility of OpenTelemetry APIs.
The process for proposing and establishing new SIGs within the OpenTelemetry project, emphasizing the importance of clear goals and sponsorship.
The need for subject matter experts to work closely with the OpenTelemetry architecture team for successful project proposals.
The importance of the Technical Committee (TC) in evaluating designs and the criteria for becoming a TC member.
The project's aim to ensure that impact can be made without necessarily being a member of the TC, emphasizing the project's health and autonomy.
Transcripts
so my name is reys um my day job is at
New Relic where I'm a senior developer
uh relations
engineer and I also work on the end user
Sig um that Austin just mentioned so if
there's any events or any end user
resources you're interested in starting
up definitely come see me or uh Ren is
here as well I think that's just the two
of us but anyways I'm going to pass a
mic down the line here so everyone has a
chance to introduce themselves but we're
here to get to know the Govern committee
find out what they do and you all will
also have a chance to ask
questions I'll stand so I can be seen uh
I'm Dan daa uh my day job is at Din trce
I've been contributing to open Telemetry
since uh late
2019 uh it's been a while I've been the
maintainer of the JS Sig since then and
I've been on the governance committee
since uh late 2020
I totally forgot to introduce myself
earlier sorry I'm Austin Parker um I am
director of Open Source at honeycom doio
I've been on the governance committee um
since last year and I've been involved
in open Telemetry since um yeah late
2019 and then I was maintainer on open
tracing before
that I'm judan King I'm a software
engineer at graphon Labs um I work with
the open Telemetry or own the open
Telemetry project as my day job um and
and I've been with open climet since it
was open tracing right so um it's been a
while um I work out I work on the
collector as well so I feel like I know
the collector folks in this room here I
know the other uh I don't know 50% of
this room here from open Telemetry but
I'm here this week to get to know the
other half of the
room hey everybody I'm Ted uh I'm one of
the co-founders of the project from the
open tracing side of the family and uh
yeah I serve on the GC
and mostly run around putting up fires
and uh trying to make people
happy hi everyone very nice to be here
I'm Alita Sharma and uh I've been on the
project working uh contributing to
metrics uh metric ga uh Prometheus inop
uh for since 2000 so you know super
thrilled to see the project grow and
really take off and be this de facto
standard uh I lead AIML observability at
Apple so uh again lot of exciting things
going on in the in the world of IML
nowadays uh and super thrilled to you
know see the intersection of our
wonderful Community here at otel
Community day today so ree over to
you sorry I didn't mean to just
awkwardly pass in
front so I guess we'll get started so
what does the governance Comm comme
do well um I think uh it's a very good
question and we've asked ourselves that
too you know several times because I
think it's a good way to actually check
in on what you know the governance
committee is up to given we are you know
elected by the community uh every two
years the GC has a two-year term and you
know it rolls over across the GC so um I
would say a couple of areas that we
definitely you know have taken on and
have been working through is first of
all wearing a product hat you know
managing the product backlog uh and the
feature backlog if you will for the
project and looking at you know some of
the new uh issues that are coming up in
the spec and you know Ted and everybody
else on the GC actually has been
phenomenal in uh contributing to making
sure that you know we all actually look
at the backlog on a regular basis so we
do look at it on a weekly basis and the
second uh area of course is that making
sure that uh you know any kind of uh
contributor and developer experience uh
concerns that the project Community
might be seeing or any governance you
know concerns that they might be seeing
are brought up and uh supporting our
maintainers of course
Ted yeah um there besides dealing with
nonsense behind the scenes um that
better goes left un said uh there's uh
been this need as the project has grown
to have more and more
structure um we're always adding
structure every time we add a piece of
structure I always ask how did we get
this far without having this much
structure um and this year is no
different though I think we've really
turned it up quite a bit by specifically
adding uh a concept much more concrete
concept of of project management so we
haven't fully rolled this out to the
community because we're kind of beta
testing it and we want to get get
comfortable uh with it but basically we
have this concept of projects uh that
are themselves something that's evolved
uh projects need to have sponsors this
is something that we've learned is that
uh as a Project's grown we have a lot
more people coming in from the outside
when we're working on a new area it's
almost always a collaboration between
subject matter experts who understand
that area because we're just you know a
bunch of Trace jockeys so we don't
necessarily know anything about
profiling or client Telemetry or
anything like that so subject matter
experts come in but they don't know much
about open Telemetry and if we want to
make sure those groups can quickly is uh
come together on a design proposal that
is going to be accepted by the Community
uh we have to like sponsor those groups
and there needs to be like some clear
active collaboration and people in those
groups need to understand who they can
go talk to uh in open Telemetry about
like getting their stuff moving um and
the next step of that is like the GC
besides putting those groups together
needs to actually like keep tabs on them
uh and also like all of the rest of the
maintainers in the project um trying to
make sure everyone's a little more
connected we just started a new GC
liaison program just making sure there's
a line of communication between the the
sigs and the GC so that there's just
some information flow going back and
forth um we do have a maintainers
meeting but um just like rather than
waiting for maintainers to like bubble
things up just like outreaching to the
maintainers and just checking in with
them and uh also uh the next step of
that we'd like to start reaching out to
new contributors a little more directly
we feel like that's a role that GC
should play uh when we see new people
joining and making uh significant
contributions just letting them know
that they've been seen and you know uh
seeing what we can do to help them
contribute
more I guess there's nothing else to say
about that I mean um I guess I would
summarize all of that with the idea that
the GC is the ultimate responsible for
the health of the project and if it
means means going monthly to the
maintainers and asking them how they're
doing or if it means building handbooks
for new contributors then that's what we
we have to be um perhaps not doing
ourselves but um fostering a a smaller
community on executing on on those ideas
on those needs um and I think that's
what the GC does so the GC um is also
part of most of or every one of us uh we
wear different heads uh in the project
so we we are at the GC only for one hour
of the week I think uh but then acting
as um members of the project in in on
the remaining of the
week h two hours because there's triage
too yeah
yeah um I'll actually add something to
to your point about wearing multiple
hats one other thing that the
GC acts as is in a lot of ways like a
strategic representative of the project
to the greater Cloud native community so
there's a lot of like little things that
are very boring about that but for
example
um the open Telemetry now has access to
some like very large amount of like
Oracle Cloud credits um that sigs can
use there's also Amazon credits that we
have access to and there's a lot of like
resources that we can get from the cncf
from Member companies who make donations
of resources into the cncf
and so part of our job is like going out
and and handling that the the like
talking to people and making stuff
available and d d d d right we also wind
up getting ried into I say roped in um
we also wind up kind of being an
interface to the media so when we go to
cubon and you know we want to talk about
like here's all the cool things
happening in otel um we collaborate with
the cncf PR department to you know build
news releases and do interviews with
analysts and journalists about what's
going on um to help promote the project
obviously but also to help promote like
open source observability and and Cloud
native more generally right um and that
also leads to kind of we'll talk to
people at a strategic level in sort of
like large member organizations right
like larger Banks or you know people
that are trying to like develop an open
Telemetry strategy um we we become
advocates for the project and advocates
for like what we're doing so a lot of
that stuff is like really not visible to
um anyone you see the outputs you see
things like this or you see like the the
open Telemetry Observatory At cubec Con
but you know it's another one of those
function you know kind of those
invisible functions of the GC you have
anything else to add or is there more to
add I don't think so I think that
basically covered it do these things we
do the the maintainer track talks yeah
we do these panels and maintainer track
talks
um
yeah so for a long time the past few
years the focus has been traces metrics
logs are you all
done yeah it's it's it shipped projects
over
all
right yeah this is actually you know
back to the like project management this
is one of the reasons why I think we've
had to like step it up more recently was
like we didn't really need much of a
road map for the first several years
it's like what are we working on this
year pinky it's like the same thing
we're working on last year tra's metrics
and logs um but as we've like gotten
near the end of that there's been a
whole bunch of interest in expanding the
project in a bunch of different ways and
at the same time there are parts of
traces metrics and logs that we kind of
put aside uh we kind of have an 8020
rule we found in open Telemetry when it
comes to design there's often like 80%
that like everyone can totally agree on
and then there's like some bit where
it's like this is weird and we're not
totally sure exactly how to do this and
we can't find consensus on it quite yet
and we tend to be like well let's put
that aside so we can get the 80% through
but at this point there's now kind of a
backlog of of some of those things so I
don't think we're actually done with
traces metrics and logs uh but we do
need to find a balance between that and
also doing like profiling and client and
machine learning and you know all these
other things so figuring out like how
much bandwidth we actually have is
something we're thinking
about I I mean is anything ever done an
open
source yes okay well we seem to have a
split decision I I think the the key
thing really is that I don't think open
Telemetry is going to be done
until the until everyone agrees with
this um but really when you think about
it right like otel is one of the otel is
it's interesting because when you look
at it how you perceive it depends on
where you're kind of looking at it um if
you look at it from the perspective of
like well I've been using X thing for
like metrics and log aggregation uh you
know the collector is just another
filebeat or whatever right um it's
another way to scrape Prometheus end
points and it's a way to perpetuate you
know it's an open source way to do it
it's a nonproprietary way to do it cool
but it's another way to kind of
perpetuate the sort of observability
signals that you've already had and then
if you look at it from the other
perspective it's like oh there's
actually a lot you can do here because
of these fundamental changes that are
baked into the data model and baked into
the tooling and ecosystem and I I don't
think we're going to be done until most
people are looking at it from that
second perspective um quite
frankly yeah I see a few ways we can
evolve in the future one of them is um
the standards like semantic conventions
we have so many semantic conventions to
make stable and to convince people that
uh specific names are um I don't know
worse pursuing more than other formats I
don't know uh then the collector itself
we we are trying to get A1 uh towards
the end of the year or I don't know
sometime this year but uh at the same
time we have profiling coming so where
do we fit profiling The Collector and
what are the next steps for The
Collector so we have so many things
around like building collectors with a
builder and building interfaces to build
collectors and so it's we're never done
I think to to your point uh and then we
have new signals as well so we have we
can grow in so many directions and uh
and for those of you who are not
contributors to the project we need you
we need you to expand into those areas
that we know that we need but we don't
have the the capacity right
now I would also say that obviously we
do have new signals coming but even if
we didn't say we we froze the project
today and said we're not expanding scope
we still have many many years of
improving the experience more
instrumentation native instrumentation
like it's it's a project that will never
be done kind of by definition it can
always be
better one point
because um and I think we were all
having some conversations earlier today
and I would like to reiterate to
Daniel's point we won't be done because
you know we have a lot of work to do in
improving developer experience and end
user uh experience for adoption uh but
it's also that we are also seeing a new
generation of observability requirements
that are coming in with AI
applications and there is foundational
work to be done at every layer uh in
order to make that intersect with the
not only with OTP which is our protocol
data protocol but also with uh inherent
support and standardized metrics uh
tracing profiling coming in uh for um a
new class of uh Hardware which is uh
gpus and you know additional compute
that is coming in so it's very very
important that you know again if you say
kubernetes uh metrics you know
observability is it done not really uh
is it uh is tracing done not really
because you know you have also this
lower layer that is completely changing
the way that we are going to have you
know the uh the networks and the and the
uh applications that are supported
across that so with a new generation
come new requirements and I think that's
the cool thing about working in open
source that a project like open
Telemetry can actually absorb and you
know start an a a whole generation of
new sigs which actually can work on
these areas as you know developers in
that space need to be able to support
that so I really welcome you know and
would love to have more contributors
joining in to supp support that new
generation you know work towards the
spec work towards the um uh protocol
work towards you know the
implementations in the different
languages and the standardization of
metrics and traces and profiles which
are coming in to support that new
generation of applications yeah just a
quick shout out to the new convenience
Sig that's starting up um what's the
developer experience developer
experience that we're calling it now
okay well uh anyways the idea is there's
at least one of the ideas there is
there's a lot of places where we could
add sugar like our apis installers
configurators and everything are
designed for completeness and meeting
requirements so that they can work
everywhere and handle everything that of
course means for doing like the basic
normal things they can be like a bit
overly complicated so where in the see
so many heads nodding right now uh where
in the project can we add sugar on top
of these things things uh to make those
parts uh just faster and more convenient
I think this is a great place for end
users to get involved because you're all
the people who really know uh where the
pain points are uh oh um this is not a
bit otel will not be done until someone
other than open Telemetry has
implemented the
spec God yeah when we have our first
like hard hard implementation Fork
that's when we're done we're all going
to go home but seriously no I I do think
like alternative there's there's so much
there right like it's very easy to see
otel as this big monolith but it's it's
very composable and um I'm the future
holds a lot for it do we want to do
audience
[Music]
Q&A I I'm sure you all have many
interesting
questions feel free to raise your hand
and I
will come find you
ah brave
soul so how do you decide like uh
whether it's appropriate to open up a
new Sig or a new signal um when people
suggested the project because like
there's a lot everyone like has a lot of
ideas and whatnot but resources are
limited and every new like Sig that we
had on it's a more over I'm just
curious a good a great great
question I guess I'll use this uh so
this is a thing we've had to work on
because we want sigs to succeed and so
what we've developed is uh a project
process and um a project template so
when someone wants to propose a new Sig
uh there's a whole bunch of pieces uh
that need to be laid out so you create
an issue in the community repo um
actually a PR in the community repo um
with all of this information and we try
to work with that group to fill it out
so it's like what is the goal um who's
actually going to work on this project
who's going to be the lead who's going
to sponsor it who is committing to
actually write code and build the
Prototype tyes what is the rough
timeline that we think uh will will be
done and if we can get all of that
together and it seems like we aren't
spreading ourselves too thin then we'll
start the Sig but we're trying to be
very cautious about this sounds like a
fun idea let's just spin a group up um
and when people don't have all of those
things just like one note there's often
this temptation be like well we'll it'll
just be experimental we'll just start
this thing in the corner and we'll just
experiment we we have a hard no to that
because that results in people doing a
whole bunch of work and then coming back
with their experiment and we're like no
we're not accepting this experiment
because you know it was not enough
people involved so that's that's how you
actually do it and that's how we we make
the
decisions um I think I can maybe speak
for the
GC and say it's actually like there's I
think there's a perception that we have
we have accepted too much
work um and I think that's partially
true but I also think that it's a
there's a little bit of recency bias
because a lot of the stuff
that just to be like pretty blunt like
people will I've I've heard from people
that are like very heavy tracing like
primarily using otel for tracing and
they're like why are you doing profiling
why are you doing events this doesn't
make sense to me profiling and events
have are literally like 2-year-old plus
projects right we've been working on
those for quite a while and in a lot of
cases those came effectively pre-
staffed right like the people that are
working on profiling as a signal are
people that work on profiling it is a
kind of a distinct um sub specialty the
people that are working on events are
web performance Engineers that you know
are very deeply committed to it and I
think they've brought like this amazing
level of like depth and care and it's
required us you know and this is not an
easy process to kind of EX to comport
those together together right now the
end result is I think a good one where
we've kind of are going of achieved for
the most part uh we will have a
compromise that makes everyone equally
unhappy
but you don't get there by like without
taking that time and putting in that
effort I do think in general we're
probably going to be like the the bar
for like a net new thing to come in is
going to get higher
but I also feel like as the spec matures
and becomes more stable you know you
don't have
to be a Sig to build on otel right like
as the spec stabilizes and the API I the
API stabiliz and all this stuff becomes
more stable I actually expect to see
like the ecosystem around otel getting
larger and there's already some examples
of this there's a project um that I
think is a Sandbox project now called
open
lemetry which is a which is basically
like o which is built on otel apis and
sdks and it's for like hey this is how
you observe these various AI components
right so you know people should feel
free to do that if you have look at the
kubernetes ecosystem there's a boatload
of projects in the cncf that are like
kubernetes plus something or some
kubernetes component right they don't
all live under the kubernetes project
you know this is this is how these
things mature and improve so people
should feel free to if they have like a
really great Hotel idea to like go build
that really great idea um and if it is
sustainable then it can become you know
a separate project as part of the you
know the
foundation perhaps just a a small
clarification before um you asked what
when we Define when we decide for a new
Sig and we here would be like open time
as a project or the GC the GC doesn't
decide when a new Sig is for right so we
can uh sometimes yeah but I mean we
merge the pr we yeah but we um it
doesn't we wear different heads um it
doesn't have to come from us and we
don't do the the the the sourcing for
that project right so we we expect the
the the people U proposing the Sig to go
and find all of the information they
need and then we just say okay that's
that's okay or that's uh we're not going
to accept it um but it's mostly formal
right I mean if if the a proposal has
enough like people and interest then
we're not going to say no um likely but
yeah I think there was a question
question yeah
okay hi how's it going uh my name is
Mike so I caught something there that
was kind of fascinating to me um
you're talking about how while
fundamentally all of these signals are
probably like a time stamp and a number
or a time stamp and a string you are uh
as it evolves you're finding that there
are specific objects you want to be able
to capture because those are the objects
we work with so like an event for
example you mentioned or a metric I
think again fundamentally time stamp in
a string time stamp in a number but
you're recognizing those as specific
objects that should be
captured uh
separately uh you also mentioned
lemetry that's an interesting word um
and I believe one of you mentioned that
you are uh responsible for observability
in AI at Apple um so this is all kind of
leading me to be very curious what are
some of these newer signals that we're
discovering that should be captured as
their own first class object in our
Telemetry in the AI world
you want to go first sure okay so I I
think you know um uh it's a good
question and I think we are still in
that uh process of discovery if you will
but uh the foundational approach uh for
even new types of objects or new types
of uh data that are related to these
objects you know mod models is a good
example um models have existed in ml for
a long time um they are becoming more
and more sophisticated larger and also
you know a lot faster but uh I think
that foundationally what we the approach
we have taken in observability
especially is to align with the existing
signals and the existing data types as
much as possible because it's always
easy to go and spin off and say Hey you
know this is another type of data uh and
let's go and support this new data
stream right and and but um it's also
very hard to roll out instrumentation
across the industry and the different
layers um of support whether that's uh
you know Hardware instrumentation or Os
instrumentation can you know or your
orchestration layer or your application
layer where you have to have full
support for that whole stack right so we
are very considerate in looking at how
that you know what these new data types
are can be aligned with existing um data
signals in observability and if you know
uh there is a need to Define additional
areas uh we are doing some work in
semantic conventions as uh Austin was
mentioning earlier we do have an llm uh
semantic convention Sig in on the
project and again also the community he
was referring to from llm metri and
others have also been working in that
Sig so we do try to kind of invite
everybody to work together and
collaborate so that we still maintain a
baseline on the protocol
standardization uh that said things will
evolve we may have new data types you
know as um the uh generation of AI
implementations gets more complex and uh
you'll see that Evolution
occur uh the
sorry no you
can uh so the the question was about
adding new signals and what I wanted to
say is if we've done a good job
designing the existing signals they
should
cover many use cases into the future the
idea is not that we're going to add a
new signal every time that technology
changes because that would be uh a
NeverEnding uh misery probably um but I
mean the the the way that I think about
the signals is that kind of at its at
its base layer everything is either an
event or some compression layer on top
of events so when you have metrics uh
you're aggregating events together
you're counting events you're you're
doing something along those lines traces
is fundamentally a start event and an
end event that are correlated together
um so the time to add a new signal is
when you see something that is so widely
adopted uh that it makes sense to to
promote it up to a higher level uh to
enable like more strict guarantees about
the structure of the data when it comes
into your backend for analysis uh or if
it's in the case of like metrics you
just can't send every event so you have
to aggregate it as almost for
compression reasons uh over the wire so
in terms of new signals that's the way
that I tend to think about it um and
then uh what you're talking about I
think is more new use cases on top of
the existing signals uh and a lot of
that will happen in uh the semantic
conventions for example so you have uh
like a definition for this event should
have these attributes and this name uh
and as new use cases come in that can
evolve much more quickly without having
to Define new top level
signals yeah
um I think like we talk a lot about
semantic inventions but it's also
probably worth considering that like
otel Telemetry itself is semantic
Telemetry right a trace is you know a
semantic conven a semantically durable
way to express like what is an what is
the series of events that happen between
a start time and an end time and d d a
metric is you know a like that's why the
otel you know otel metric spec is like
super in-depth right why is that because
you need all these semantics um for
storage reasons for query reasons and
also for like end user reasons right
like there's no reason if you know that
a you know particular instrument is of
Type X then why let people do the wrong
kind of math on it um now a lot of the
tooling hasn't caught up to where all
these semantics are and I think that is
you know to my earlier point about like
people that are in kind of Class A
versus Class B of how they look at o
like we're not going to get to that
class B part until there is this the
tooling kind of catches up to where
these underlying um standards are
going so perhaps um specifically to the
um AI or or um that kind of new signal I
would say that it's too early for us to
know whether that warrs a new signal uh
we we had proposals from allita actually
for rum signal uh in the past when we
discussed you know do we need a rum
signal or do we not need it um people
still don't know whether we need that um
so it might we might have we might have
a new signal in the future um the same
for profiling so profiling even though
we can see we can debate whether it is
an event or not um when we look at the
details for theab for the LLP profiling
um specification we can see that a lot
of it is um optimization to make
profiling usable as a signal you know so
it's so much data that is that is
generated for one specific sample that
we have to do a lot of
optimizations um and I guess my point is
because we don't know which
optimizations we need in the future for
this new um new whole new area we don't
know whether we need or not a new signal
in the future even though it might look
like what we that we have the basics
covered already
today yeah
sure uh well I just wanted to say like
the most most important signal we need
to integrate is the bat signal like if
every time a system goes down if like
that thing could go off
so hi um I'm interested in the sort of
language question um you know when you
think about the balance
between um supporting different
languages and trying to stand I versus
idiom in a language kind of where do you
think the lines should be drawn how do
you how do you conal a spec that's as
broad as this across so many different
uh language environments in
particular
ionna yeah so I I I think that's
dangerously close to a TC question but
as a language maintainer and a GC member
I guess I'll try um the spec is written
very generally and leaves a lot of a lot
to the interpretation of the language
maintainers um we we really do want
everything to feel the same if you're
switching between languages as a lot of
Engineers do uh you shouldn't have to
completely relearn how to use otel uh
but at the same time if you've been
using some language like Java your
entire life you shouldn't install otel
and then just be like what is this API
looks so weird so it's it's kind of a
balance between the two uh and from a
specification level uh it's generally
left quite open to interpretation by
maintainers in order for them to make
decisions about their own uh language
implementations um that said we do have
some very specific requirements in the
spec uh around certain things that
either come up in multiple languages and
and multiple maintainers have asked the
same question that'll be added to the
spec as a clarification a lot of those
are like should requirements not not
must
requirements um but yeah in in general
we we like to uh like to think that we
can trust our maintainers to to make
those decisions in a way that makes
sense uh and occasionally you know
people make mistakes it happens and
things bubble up uh but that's why we
have uh kind of a long uh
experimental uh type of uh cycle longer
than I think most people would hope
sometimes but uh it is you know for good
reason and it does help a lot in that
regard I the the developer experience
Sig is designed to kind of tackle this
question I think just speaking
spiritually I do feel like we've
probably ered on the side of not being
explicit enough that languages
should be air on the side of like idiom
um existing language idioms
maybe you know maybe we should have been
more explicit about that from the jump
that has always been the intention of
the project right because one of the big
things we learned from open tracing was
that uh spec
adherence to the API spec um made the
spec really made the apis really painful
in langu in
it made it suck equally for everyone
which wasn't a great outcome I don't
think um and we really wanted to avoid
that in open
Telemetry
one just just to add one quick thing as
a a language maintainer I receive
complaints opposite complaints for this
question all the time so I get issues
that are open that are like this feels
like Java why would you implement this
way like you're a JS developer I'm like
well I don't know like and then I get
the exact opposite uh complaint that's
like you made this way too JS specific
and I'm coming from Ruby and this feels
super weird I've been using open
Telemetry Ruby for years and and now I
can't figure out how to use your API
I'll I'll get exactly opposite
conflicting issues opened and like this
is something that I think will never be
completely solved uh and like I said
it's just a balance that we have to
strike yeah um so
one point I've noticed people uh haven't
realized so I've been saying more
recently is the spec actually doesn't
matter the details don't matter we
actually don't care what we care is that
the Black Box whatever it is
participates in tracing and emits OTP
and semantic conventions that is what we
care about if abandoning the spec and
everything else uh gets you to that
point then uh that's fine like that's
that is my hot take like now I'm not
saying now I'm not saying if we I'm not
saying if we spun up a group within open
Telemetry for a new language we wouldn't
say follow the spec and do all of these
things like for sure that is how we do
it but I'm saying for somebody out there
for some project out there if a language
decides they're going to fully integrate
tracing and observability into the
language and they're uh
you know um contous language developers
so they want to do everything different
like as long as the thing that's the
same is semantic conventions and OTP and
participating in tracing like that is
actually as far as like the the true
spec it's it's the data the the data is
more important than how the data is
generated and this does come up when
people are like what if I use this other
API to to do it we're like well that's
fine if whatever works for you
I I want to be very specific
participating in open Telemetry
context not just participating in
tracing uh API the API interop has to
work and the context propagation and
assigning
yeah I guess the idea and going into the
same spirit is we are all about vendor
neutrality so the ability of changing
Solutions if you want to change if you
want to use another API that's fine um
as long as you're still able to move uh
to different Solutions as As You Wish as
you as you
need I think we're one more question and
an important way for that Sig to be
successful are for those subject matter
experts to have direct Liaisons with uh
people who've understand the open
Telemetry architecture so those two
groups working together for a while to
develop their proposals we found that is
way way way uh has a much higher success
rate um and then on the other side of
that our hope is that some of those
subject matter experts will stay
involved and as they expand their
understanding of open slamry in general
become spec approvers and then uh
eventually uh we would like to make sure
that the TC contains subject matter
experts for every signal um but but we
kind of need it to go in that that level
of graduation
basically I think we're at a at a time
you want to no I think yeah um perhaps
just one thing to add there is we know
that we have gaps there uh on the TC
like profiling uh right now we don't
have anyone with profiling um background
there or some um but uh the other one
that I can think of is like client
instrumentation right so we we could
make use of people with better client
instrumentation background on a
TC um so but we don't have a a process
defined right now to cover those gaps uh
we we are thinking about how to make the
path to the TC more clear uh and we have
that in our heads uh it just needs to be
documented I I also just want to point
out like this has been a topic
for at least a year and a half or so um
certainly ever since I started like yeah
the the question not question but like
just this entire like top level TC um
maintenance and the role of the TC you
know I think we've actually had a lot of
really good progress this year in terms
of trying
to you know help the TC grow organically
um I know we actually have one of the
newest TC members in the room you know
so there's always more work to be done
um and but we need to make sure that
we're doing it in kind of a responsible
way for both the health of the project
um and also the health of the TC so
feedback is always of course
welcome yeah as as far as adding TC
members for for like you said new
signals but it doesn't have to be a new
signal like a new area of the project uh
people coming in from the outside are
very unlikely to be immediately added to
the TC and so kind of you know the the
these things that run for long periods
of time like a new signal is not going
to be added overnight so subject matter
experts that are that are working in
those signals should
hopefully evolve into the project and
and continue to stay in the project uh
to me the question is not like how do we
get uh you know a TC member with uh
experience in this specific area that we
just started working on but it's how do
we uh get that person's opinion
effectively integrated into the project
whether they're a TC member or not if a
subject matter expert is is working in
the project they shouldn't feel like
they have to to be like Oh I'm not on
the TC so I don't get a say here or I'm
not a maintainer so I don't get a say if
you know what you're talking about on a
particular subject uh having your that
that opinion heard and and integrated
into the project in a way that makes
sense uh to me is more important than
what that person's actual title
is yeah just yeah I think there's a
really important point to me made there
the thing we are looking for in TC
members is not necessarily being a
subject matter expert it's about being
able to work with subject matter experts
and evaluate designs against broader
requirements so someone who is say a
super expert at profiling or something
but has like a very high-handed opinion
about how to do it and doesn't really
want to listen to anyone else they might
be right it's not that they they would
necessarily be wrong but that is not the
kind of person we're looking for on the
TC we're looking for someone who's able
to to maybe be a little more diplomatic
and a little more interested in in
gathering requirements and and hearing
from everyone uh rather than being like
my way or the highway so that that's a
really important
quality yeah and and just to close that
out I think the ultimate goal is to what
Daniel said right in a perfect
world the pro you know the the the
processes and the ways that we work are
are established enough that you don't
that there is both a the perception and
the reality a line up that you don't
have to be like on the TC to be able to
actually have impact you know and I I
actually think that if there is a
perception that you have to be on the TC
to have impact then that is like a
failing of the project that we should
address um and you you hear all sorts of
things at this level um but I think
we're always working to make sure that
there is more autonomy and that people
can kind of come in and you know do do
their best work and as long as it's
aligned with what we're trying to do at
the project level that they will be
successful so I think with that we are
over
time
um yeah so we don't want to cut too much
into everyone's else's um time but you
know we'll be here all day so if you
want to grab someone in the GC um
there's actually a few other GC members
raise your hand yeah GC or TC GC or TC
yeah TC members also raise your hand TC
M yeah anybody yeah so come find us
during a break or at lunch um if you
have four questions and there will be a
panel with our TC members this afternoon
so thank you all very
much and thank you Reese
Browse More Related Video
PANEL: OpenTelemetry Technical Committee Panel - Josh Suereth, Reiley Yang, Rynn Mancuso, Jack Berg
AI and the Workforce
Welcome + Opening Remarks - Austin Parker, OTel Community Day Event Chair
Heihachi, Guest Characters, and the Future of Tekken 8 - EVO 2024 Interview
What's new in Angular
Michael Shellenberger on Why California Is Such a Mess
5.0 / 5 (0 votes)