server : refactor middleware and /health endpoint #9056
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
ggerganov
approved these changes
Aug 16, 2024
Co-authored-by: Georgi Gerganov <[email protected]>
|
I started a discussion thread related to this issue, please take a look: #9276 |
ngxson
added
the
breaking change
arthw
pushed a commit
to arthw/llama.cpp
that referenced
this pull request
Nov 15, 2024
* server : refactor middleware and /health endpoint * move "fail_on_no_slot" to /slots * Update examples/server/server.cpp Co-authored-by: Georgi Gerganov <[email protected]> * fix server tests * fix CI * update server docs --------- Co-authored-by: Georgi Gerganov <[email protected]>
arthw
pushed a commit
to arthw/llama.cpp
that referenced
this pull request
Nov 18, 2024
* server : refactor middleware and /health endpoint * move "fail_on_no_slot" to /slots * Update examples/server/server.cpp Co-authored-by: Georgi Gerganov <[email protected]> * fix server tests * fix CI * update server docs --------- Co-authored-by: Georgi Gerganov <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
/healthendpointIn the beginning,
/healthendpoint was used to retrieve slots state. That was because at the time,/completionsendpoint returns an error if there is no slot available. Therefore,/healthwas used to allow the application to wait until one slot is available.Nowadays, the server now can queue (defer) the request if no slots is available.
/healthis used by docker for health checking. This is now become a problem when the server is busy doing a long task,/healthcan timeout. On HF inference endpoint, this causes the container to be in unhealthy state, which triggers a force restart.Therefore, I propose a cleaner usage:
GET /healthis now purely used to report actual healthGET /slotscan be used as a replacement to get slot stateAs a consequence,
/health?fail_on_no_slot=1is also moved to/slots?fail_on_no_slot=1(for compatibility, we keep this option)Refactor middleware
Some repeated code blocks, for example setting
Access-Control-Allow-Origin, is now moved to middleware.Middleware now also responsible to return error if the server is not yet ready:
When the server starts, if the model is being loaded, accessing to any endpoint will result in 503 error code:
Behavior on loading model failed
If model fails to load (for example, file does not exist), the server will simply exit with status code 1. This resolves #7787 where user reports that loading invalid model causes the server to crash.