rfc(feature): Support Linking Traces

[Lms24]

This RFC specifies our plans to support linked traces via span links.

Since this became quite a lot of text, I recommend especially reading the Preferred Option section which specifies SDK, Relay and Storage changes. The other sections are supposed to provide context and additional information.

Rendered Document

ref: https://github.com/getsentry/projects/issues/268

[andreiborza]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

[Lms24]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

I would argue that this is up to users. If users decide to drop a span link, we just won't be able to link the respective spans together. Meaning, unless there's a strong case for consistent sampling across traces (which is still TBD) I think we can permit mutation of links just like any other span property.

[philipphofmann]

Thanks for all the deep context and background. I added a few comments.

[mjq]

Sorry, looks like "starting a review" with both new comments and replies got weird and duplicated a bunch of stuff in the timeline above. Sorry 😬

[cleptric]

Let's do it.

[smeubank]

I would go with your preferred option. Great write up, thorough and detail. Glad the session topic was address as well.

One option that is not here, is the "do nothing". i suppose, but maybe that is pedantic.

[philipphofmann]

LTGM

While this doesn't solve all of our long-lived trace problems, and I would love to see traces linked to sessions, this is a significant step in the right direction. This concept will solve multiple issues you thoroughly described in the RFC. It's also excellent that OTel already has this concept

Thanks for the detailed write-up. I'm looking forward to implementing this.

[philipphofmann]

Most customers don't sample at 100%. They will frequently experience previously negatively sampled traces. I can imagine that some customers would like to navigate to the last positively sampled trace when most of their previous traces are negatively sampled. But it might be OK to wait for customers to ask for that feature. The current proposal would allow multiple options to achieve this without changing the APIs and the protocol.

[Lms24]

Yup. I sketched out a solution a while ago how we could add multiple links to indicate missing traces due to sampling:

I didn't include this in the RFC because as I wrote, it's something we can tackle if there's demand for it.

[Lms24]

This RFC specifies our plans to support linked traces via span links.

Since this became quite a lot of text, I recommend especially reading the Preferred Option section which specifies SDK, Relay and Storage changes. The other sections are supposed to provide context and additional information.

Rendered Document

ref: https://github.com/getsentry/projects/issues/268

[andreiborza]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

[Lms24]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

I would argue that this is up to users. If users decide to drop a span link, we just won't be able to link the respective spans together. Meaning, unless there's a strong case for consistent sampling across traces (which is still TBD) I think we can permit mutation of links just like any other span property.

[philipphofmann]

Thanks for all the deep context and background. I added a few comments.

[mjq]

Sorry, looks like "starting a review" with both new comments and replies got weird and duplicated a bunch of stuff in the timeline above. Sorry 😬

[cleptric]

Let's do it.

[smeubank]

I would go with your preferred option. Great write up, thorough and detail. Glad the session topic was address as well.

One option that is not here, is the "do nothing". i suppose, but maybe that is pedantic.

[philipphofmann]

LTGM

While this doesn't solve all of our long-lived trace problems, and I would love to see traces linked to sessions, this is a significant step in the right direction. This concept will solve multiple issues you thoroughly described in the RFC. It's also excellent that OTel already has this concept

Thanks for the detailed write-up. I'm looking forward to implementing this.

[philipphofmann]

Most customers don't sample at 100%. They will frequently experience previously negatively sampled traces. I can imagine that some customers would like to navigate to the last positively sampled trace when most of their previous traces are negatively sampled. But it might be OK to wait for customers to ask for that feature. The current proposal would allow multiple options to achieve this without changing the APIs and the protocol.

[Lms24]

Yup. I sketched out a solution a while ago how we could add multiple links to indicate missing traces due to sampling:

I didn't include this in the RFC because as I wrote, it's something we can tackle if there's demand for it.

[Lms24]

This RFC specifies our plans to support linked traces via span links.

Since this became quite a lot of text, I recommend especially reading the Preferred Option section which specifies SDK, Relay and Storage changes. The other sections are supposed to provide context and additional information.

Rendered Document

ref: https://github.com/getsentry/projects/issues/268

[andreiborza]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

[Lms24]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

I would argue that this is up to users. If users decide to drop a span link, we just won't be able to link the respective spans together. Meaning, unless there's a strong case for consistent sampling across traces (which is still TBD) I think we can permit mutation of links just like any other span property.

[philipphofmann]

Thanks for all the deep context and background. I added a few comments.

[mjq]

Sorry, looks like "starting a review" with both new comments and replies got weird and duplicated a bunch of stuff in the timeline above. Sorry 😬

[cleptric]

Let's do it.

[smeubank]

I would go with your preferred option. Great write up, thorough and detail. Glad the session topic was address as well.

One option that is not here, is the "do nothing". i suppose, but maybe that is pedantic.

[philipphofmann]

LTGM

While this doesn't solve all of our long-lived trace problems, and I would love to see traces linked to sessions, this is a significant step in the right direction. This concept will solve multiple issues you thoroughly described in the RFC. It's also excellent that OTel already has this concept

Thanks for the detailed write-up. I'm looking forward to implementing this.

[philipphofmann]

Most customers don't sample at 100%. They will frequently experience previously negatively sampled traces. I can imagine that some customers would like to navigate to the last positively sampled trace when most of their previous traces are negatively sampled. But it might be OK to wait for customers to ask for that feature. The current proposal would allow multiple options to achieve this without changing the APIs and the protocol.

[Lms24]

Yup. I sketched out a solution a while ago how we could add multiple links to indicate missing traces due to sampling:

I didn't include this in the RFC because as I wrote, it's something we can tackle if there's demand for it.

[Lms24]

This RFC specifies our plans to support linked traces via span links.

Since this became quite a lot of text, I recommend especially reading the Preferred Option section which specifies SDK, Relay and Storage changes. The other sections are supposed to provide context and additional information.

Rendered Document

ref: https://github.com/getsentry/projects/issues/268

[andreiborza]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

[Lms24]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

I would argue that this is up to users. If users decide to drop a span link, we just won't be able to link the respective spans together. Meaning, unless there's a strong case for consistent sampling across traces (which is still TBD) I think we can permit mutation of links just like any other span property.

[philipphofmann]

Thanks for all the deep context and background. I added a few comments.

[mjq]

Sorry, looks like "starting a review" with both new comments and replies got weird and duplicated a bunch of stuff in the timeline above. Sorry 😬

[cleptric]

Let's do it.

[smeubank]

I would go with your preferred option. Great write up, thorough and detail. Glad the session topic was address as well.

One option that is not here, is the "do nothing". i suppose, but maybe that is pedantic.

[philipphofmann]

LTGM

While this doesn't solve all of our long-lived trace problems, and I would love to see traces linked to sessions, this is a significant step in the right direction. This concept will solve multiple issues you thoroughly described in the RFC. It's also excellent that OTel already has this concept

Thanks for the detailed write-up. I'm looking forward to implementing this.

[philipphofmann]

Most customers don't sample at 100%. They will frequently experience previously negatively sampled traces. I can imagine that some customers would like to navigate to the last positively sampled trace when most of their previous traces are negatively sampled. But it might be OK to wait for customers to ask for that feature. The current proposal would allow multiple options to achieve this without changing the APIs and the protocol.

[Lms24]

Yup. I sketched out a solution a while ago how we could add multiple links to indicate missing traces due to sampling:

I didn't include this in the RFC because as I wrote, it's something we can tackle if there's demand for it.

[Lms24]

This RFC specifies our plans to support linked traces via span links.

Since this became quite a lot of text, I recommend especially reading the Preferred Option section which specifies SDK, Relay and Storage changes. The other sections are supposed to provide context and additional information.

Rendered Document

ref: https://github.com/getsentry/projects/issues/268

[andreiborza]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

[Lms24]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

I would argue that this is up to users. If users decide to drop a span link, we just won't be able to link the respective spans together. Meaning, unless there's a strong case for consistent sampling across traces (which is still TBD) I think we can permit mutation of links just like any other span property.

[philipphofmann]

Thanks for all the deep context and background. I added a few comments.

[mjq]

Sorry, looks like "starting a review" with both new comments and replies got weird and duplicated a bunch of stuff in the timeline above. Sorry 😬

[cleptric]

Let's do it.

[smeubank]

I would go with your preferred option. Great write up, thorough and detail. Glad the session topic was address as well.

One option that is not here, is the "do nothing". i suppose, but maybe that is pedantic.

[philipphofmann]

LTGM

While this doesn't solve all of our long-lived trace problems, and I would love to see traces linked to sessions, this is a significant step in the right direction. This concept will solve multiple issues you thoroughly described in the RFC. It's also excellent that OTel already has this concept

Thanks for the detailed write-up. I'm looking forward to implementing this.

[philipphofmann]

Most customers don't sample at 100%. They will frequently experience previously negatively sampled traces. I can imagine that some customers would like to navigate to the last positively sampled trace when most of their previous traces are negatively sampled. But it might be OK to wait for customers to ask for that feature. The current proposal would allow multiple options to achieve this without changing the APIs and the protocol.

[Lms24]

Yup. I sketched out a solution a while ago how we could add multiple links to indicate missing traces due to sampling:

I didn't include this in the RFC because as I wrote, it's something we can tackle if there's demand for it.

[Lms24]

This RFC specifies our plans to support linked traces via span links.

Since this became quite a lot of text, I recommend especially reading the Preferred Option section which specifies SDK, Relay and Storage changes. The other sections are supposed to provide context and additional information.

Rendered Document

ref: https://github.com/getsentry/projects/issues/268

[andreiborza]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

[Lms24]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

I would argue that this is up to users. If users decide to drop a span link, we just won't be able to link the respective spans together. Meaning, unless there's a strong case for consistent sampling across traces (which is still TBD) I think we can permit mutation of links just like any other span property.

[philipphofmann]

Thanks for all the deep context and background. I added a few comments.

[mjq]

Sorry, looks like "starting a review" with both new comments and replies got weird and duplicated a bunch of stuff in the timeline above. Sorry 😬

[cleptric]

Let's do it.

[smeubank]

I would go with your preferred option. Great write up, thorough and detail. Glad the session topic was address as well.

One option that is not here, is the "do nothing". i suppose, but maybe that is pedantic.

[philipphofmann]

LTGM

While this doesn't solve all of our long-lived trace problems, and I would love to see traces linked to sessions, this is a significant step in the right direction. This concept will solve multiple issues you thoroughly described in the RFC. It's also excellent that OTel already has this concept

Thanks for the detailed write-up. I'm looking forward to implementing this.

[philipphofmann]

Most customers don't sample at 100%. They will frequently experience previously negatively sampled traces. I can imagine that some customers would like to navigate to the last positively sampled trace when most of their previous traces are negatively sampled. But it might be OK to wait for customers to ask for that feature. The current proposal would allow multiple options to achieve this without changing the APIs and the protocol.

[Lms24]

Yup. I sketched out a solution a while ago how we could add multiple links to indicate missing traces due to sampling:

I didn't include this in the RFC because as I wrote, it's something we can tackle if there's demand for it.

[Lms24]

This RFC specifies our plans to support linked traces via span links.

Since this became quite a lot of text, I recommend especially reading the Preferred Option section which specifies SDK, Relay and Storage changes. The other sections are supposed to provide context and additional information.

Rendered Document

ref: https://github.com/getsentry/projects/issues/268

[andreiborza]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

[Lms24]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

I would argue that this is up to users. If users decide to drop a span link, we just won't be able to link the respective spans together. Meaning, unless there's a strong case for consistent sampling across traces (which is still TBD) I think we can permit mutation of links just like any other span property.

[philipphofmann]

Thanks for all the deep context and background. I added a few comments.

[mjq]

Sorry, looks like "starting a review" with both new comments and replies got weird and duplicated a bunch of stuff in the timeline above. Sorry 😬

[cleptric]

Let's do it.

[smeubank]

I would go with your preferred option. Great write up, thorough and detail. Glad the session topic was address as well.

One option that is not here, is the "do nothing". i suppose, but maybe that is pedantic.

[philipphofmann]

LTGM

While this doesn't solve all of our long-lived trace problems, and I would love to see traces linked to sessions, this is a significant step in the right direction. This concept will solve multiple issues you thoroughly described in the RFC. It's also excellent that OTel already has this concept

Thanks for the detailed write-up. I'm looking forward to implementing this.

[philipphofmann]

Most customers don't sample at 100%. They will frequently experience previously negatively sampled traces. I can imagine that some customers would like to navigate to the last positively sampled trace when most of their previous traces are negatively sampled. But it might be OK to wait for customers to ask for that feature. The current proposal would allow multiple options to achieve this without changing the APIs and the protocol.

[Lms24]

Yup. I sketched out a solution a while ago how we could add multiple links to indicate missing traces due to sampling:

I didn't include this in the RFC because as I wrote, it's something we can tackle if there's demand for it.

[Lms24]

This RFC specifies our plans to support linked traces via span links.

Since this became quite a lot of text, I recommend especially reading the Preferred Option section which specifies SDK, Relay and Storage changes. The other sections are supposed to provide context and additional information.

Rendered Document

ref: https://github.com/getsentry/projects/issues/268

[andreiborza]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

[Lms24]

Great stuff! One thing I was wondering about is how links are affected by span filtering. What if links are actively dropped by users? Is that a concern?

I would argue that this is up to users. If users decide to drop a span link, we just won't be able to link the respective spans together. Meaning, unless there's a strong case for consistent sampling across traces (which is still TBD) I think we can permit mutation of links just like any other span property.

[philipphofmann]

Thanks for all the deep context and background. I added a few comments.

[mjq]

Sorry, looks like "starting a review" with both new comments and replies got weird and duplicated a bunch of stuff in the timeline above. Sorry 😬

[cleptric]

Let's do it.

[smeubank]

I would go with your preferred option. Great write up, thorough and detail. Glad the session topic was address as well.

One option that is not here, is the "do nothing". i suppose, but maybe that is pedantic.

[philipphofmann]

LTGM

While this doesn't solve all of our long-lived trace problems, and I would love to see traces linked to sessions, this is a significant step in the right direction. This concept will solve multiple issues you thoroughly described in the RFC. It's also excellent that OTel already has this concept

Thanks for the detailed write-up. I'm looking forward to implementing this.

[philipphofmann]

Most customers don't sample at 100%. They will frequently experience previously negatively sampled traces. I can imagine that some customers would like to navigate to the last positively sampled trace when most of their previous traces are negatively sampled. But it might be OK to wait for customers to ask for that feature. The current proposal would allow multiple options to achieve this without changing the APIs and the protocol.

[Lms24]

Yup. I sketched out a solution a while ago how we could add multiple links to indicate missing traces due to sampling:

I didn't include this in the RFC because as I wrote, it's something we can tackle if there's demand for it.