CP-52708: Avoid making Unix read/write calls for internal API calls: …#6178
CP-52708: Avoid making Unix read/write calls for internal API calls: …#6178edwintorok wants to merge 2 commits intoxapi-project:masterfrom
Conversation
f5d6094 to
5e1d2db
Compare
5e1d2db to
78a86c6
Compare
This will enable short-circuiting internal API calls. When the FD is missing mark the call as Internal. Signed-off-by: Edwin Török <edwin.torok@cloud.com>
…forward the API call directly like we do with the CLI for calls to the coordinator when we are the coordinator Signed-off-by: Edwin Török <edwin.torok@cloud.com>
78a86c6 to
4d275d6
Compare
Pull request was converted to draft
|
Moving back to draft, no feature flag on this one. |
|
This seems quite safe and contained to merge, are you sure this should still be a draft? |
|
Unrelated: why is a parameter that is used everywhere using a long and unusual name like |
|
@lindig I think you need a time machine set to 2006 in order to answer that question. |
|
I can try a quick grep + sed if it makes you happy, I would be painful to rebase all work being done though 😬 |
|
Can we merge this? |
|
Looking at this again, I wonder if/how it affects tracing and also logging. Would we lose info like the source IP from the log lines? |
GabrielBuica
left a comment
There was a problem hiding this comment.
I don't think it is affecting distributed tracing. Might be worth a check.
It seems to me that this is pretty straight forward and has been lingering for a while.
Maybe run a few tests with dt enabled, check logs and merge it.
Or alternatively closed if it goes nowhere.
…forward the API call directly like we do with the CLI for calls to the coordinator when we are the coordinator
We might want to check that tracing still works with this, e.g. we might have to create a new span.
This avoids a lot of needless internal serialization, and a similar mechanism is already used by the CLI.