docs: add noIndex:true docs for experimental query planning mode #5883
Conversation
|
@lrlna, please consider creating a changeset entry in |
| unsupported features are detected, the router falls back to legacy with an | ||
| `info` log. | ||
| * `legacy`. Enables only the legacy query planner. | ||
| * `both`. Enables comparison between legacy and new native query planners. |
There was a problem hiding this comment.
The difference between both and both_best_effort is not clear to me? Why even have the both option around?
There was a problem hiding this comment.
Is it that both won't fall back to legacy-only if unsupported features are detected?
There was a problem hiding this comment.
Yes, the both won't fall back. We only just internally introduced both_best_effort in the last week specifically for us to be able to run this canary testing for all deployments. I can remove both_best_effort (or I guess both) from this list if you think it's too confusing.
There was a problem hiding this comment.
I think we should remove both. I don't see why anybody would care about falling back to legacy_only if the native planner is running in shadow mode.
| unsupported features are detected, the router falls back to legacy with an | ||
| `info` log. | ||
| * `legacy`. Enables only the legacy query planner. | ||
| * `both`. Enables comparison between legacy and new native query planners. |
There was a problem hiding this comment.
Is it that both won't fall back to legacy-only if unsupported features are detected?
|
Thank you both! All suggestions have been applied. If you think it's less confusing to have mention of only one of the |
lrlna
requested review from
a team,
dariuszkuc,
sachindshinde,
goto-bus-stop,
SimonSapin,
TylerBloom and
duckki
as code owners
When adding `__typename` field for abstract types, we need to check for inline fragment type condition and not the parent selection set. If we have an inline fragment without type condition and it is pointing to and abstract type, then the required `__typename` field should have already been added to its parent field.
We continue adding `__typename` for fragments with a type condition as their type condition _may_ be different than the parent type.
```graphql
foo {
# if this is abstract type we will add __typename
a
... @Skip(if: false) {
# if this fragment is on abstract type we don't need to add anything as
# we already added the __typename to the parent selection
b
}
}
```
…fter modification (#5870)
Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Iryna Shestak <[email protected]>
Co-authored-by: Edward Huang <[email protected]> Co-authored-by: Chandrika Srinivasan <[email protected]>
Co-authored-by: Edward Huang <[email protected]>
Signed-off-by: Benjamin Coenen <[email protected]> Co-authored-by: Edward Huang <[email protected]>
Signed-off-by: Benjamin <[email protected]> Co-authored-by: bryn <[email protected]> Co-authored-by: Benjamin <[email protected]>
Co-authored-by: Simon Sapin <[email protected]>
Signed-off-by: Benjamin <[email protected]>
Co-authored-by: Edward Huang <[email protected]> Co-authored-by: Coenen Benjamin <[email protected]>
lrlna
force-pushed
the
lrlna/both-qp-docs
branch
from
e64c941 to
63fbf8a
Compare
|
i messed up the changing the base branch, and don't want to spend too much time on trying to fix it, so this PR has moved to #5904 |
Draft of the docs for experimental query planning mode. @chandrikas, @shorgi let me know what you two think!