Conversation
This defines trace and span link storage, ingestion and search behavior, and the initial integration scope. It also aligns pull and digest semantics with RFC-0008. Provenance is dropped because all values are client asserted, so they cannot be used to weigh trust and would add unnecessary complexity. Signed-off-by: mprahl <mprahl@users.noreply.github.com>
e26fe75 to
f2e96ae
Compare
| skill_workspace VARCHAR(63) NOT NULL, | ||
| skill_organization VARCHAR(64) NOT NULL DEFAULT '', |
There was a problem hiding this comment.
Do we have downstream use case for querying traces with extra metadata like organization/workspace/role/digest? For example, I don't imagine ppl ask "I want to know all traces that uses whichever skills in my org", and even if there is few ppl want that, it's possible to do through join as we don't need great performance for niche use cases. Since there can be millions of traces that refers to a single same skill, denormalizing them feels like unnecessary storage cost.
There was a problem hiding this comment.
Organization and workspace are required to uniquely identify the skill version linked to the trace.
Role is not intended for querying. It explains why the span was linked: either the framework activated the skill or the span used it. This information is displayed in the UI.
Digest allows querying traces for the same skill content regardless of how it was packaged. When a packaged agent plugin is re-imported, RFC-0008 creates a new skill version for every discovered member, even when its content is unchanged, so each version retains its package-derived source. If only one of two bundled skills changed, the unchanged skill’s new version shares the previous version’s digest. That digest lets a user query traces for the same content across both agent plugin versions.
There was a problem hiding this comment.
Organization and workspace are required to uniquely identify the skill version linked to the trace.
Ah right, sorry I was under the mental model that skill version has an unique ID.
That digest lets a user query traces for the same content across both agent plugin versions.
Hmm not convinced why would users want to do that, and even if so, they can search traces for multiple skill versions and join them, unless a single skill is shared across tens of plugins (which I think is extreme niche case)?
Signed-off-by: mprahl <mprahl@users.noreply.github.com>
2 participants
This defines trace and span link storage, ingestion and search behavior, and the initial integration scope. It also aligns pull and digest semantics with RFC-0008.
Provenance is dropped because all values are client asserted, so they cannot be used to weigh trust and would add unnecessary complexity.