[Bugfix][Frontend] Fix Issues Under High Load With `zeromq` Frontend #7394

robertgshaw2-redhat · 2024-08-10T14:51:08Z

BUG REPO:

On v0.5.4 due to the mp frontend, we fail under heavy load:
server script

vllm serve meta-llama/Meta-Llama-3-8B-Instruct

client script:

python benchmarks/benchmark_serving.py --model meta-llama/Meta-Llama-3-8B-Instruct --dataset-name sonnet --sonnet-input-len 1024 --sonnet-output-len 128 --dataset-path benchmarks/sonnet.txt --port 8000 --num-prompts 10000 --request-rate 1000

The server dies when we get ~1000 active requests in the system

SUMMARY:

Multiprocessing OpenAI Frontend uses icp for the zeromq-based RPC connection. Each new request connects to this socket on the RPC Server side, meaning for N active requests, we will have N connections.
zmq.MAX_SOCKETS defaults to 1023. If more than this number of sockets is opened we fail Too many open files error. This puts a hard cap on the number of active requests in vLLM and causes the bugs lists below.
Setting zmq.MAX_SOCKETS to a high number resolves our issue. However, lsof -U | wc -l shows we are using unix sockets under the hood, the number of active requests will be limited by system ulimit. This is a bad UX and again puts a hard cap on the number of active requests running in vLLM.
This PR introduces a proxy on the RPCClient side. The proxy creates a single ipc connection from the RPCClient to the RPCServer. Each API server thread handing a request connects to the proxy via inproc protocol and the message is forwarded over the since ipc connection with proper identity tags. This means we only ever create 1 unix socket.
As a bonus, this architecture is more performant by ~5% E2E, since we avoid so many syscalls to open unix sockets

NOTES:

Note: zmq has a variable called zmq.MAX_SOCKETS which applies at the Context level. This variable can be set up to a maximum of zmq.SOCKET_LIMIT=65536. This imposes a hard limit on the number of concurrent connections (see comments below for more details on this topic).

PERFORMANCE:

Serving benchmark H100 for Llama-3-8B-Instruct at QPS=10:

server launch:

vllm serve meta-llama/Meta-Llama-3-8B-Instruct --enable-chunked-prefill --disable-log-requests

benchmark script:

MODEL="meta-llama/Meta-Llama-3-8B-Instruct"
TOTAL_SECONDS=120
QPS_RATES=("10" "10" "10")
PORT=8000

for QPS in ${QPS_RATES[@]}; do
    NUM_PROMPTS=$((TOTAL_SECONDS * QPS))
    echo "===== RUNNING NUM_PROMPTS = $NUM_PROMPTS QPS = $QPS ====="

    python3 benchmarks/benchmark_serving.py \
        --model $MODEL \
        --dataset-name sonnet --sonnet-input-len 550 --sonnet-output-len 150 --dataset-path benchmarks/sonnet.txt \
        --num-prompts $NUM_PROMPTS --request-rate $QPS --port $PORT
done

results:

`branch`	`run 1`	`run 2`	`run 3`	mean
`main`	51.7	50.7	51.8	51.4
`pr`	48.5	49.5	49.0	49.0
`--disable-frontend-multiprocessing`	72.8	71.1	72.9	72.3

FIX

UPDATE 8/20 POST OFFLINE DISCUSSION

Per discussion offline with @njhill @simon-mo and @youkaichao

After much further investigation, we ran into a separate issue with the proxy design after enabling the proxy. Specifically, we encountered a situation where under high load, zeromq can DROP MESSAGES 😢 , due to a concept called high-water mark (https://zeromq.org/socket-api/#high-water-mark), which is designed to protect servers running zeromq from slow clients. Specifically:

DEALER sockets go mute when the HWM is reached
ROUTER sockets drop messages when the HWM is reached

This prevents issues where a server running ZMQ can run OOM due to a slow reciever which is not able to keep up with message passing. The HWM is defaulted to 1000. Its difficult to track down exactly the sequence of events that causes the message dropping under load, but the HWM seems to be the culprit

In our case, we disable the HWM 😨 . This is generally not advisable, however:

We are in control of both the RPCServer + RPCClient
Either the server OR the client dying is catastrophic for vllm + the whole system should restart in this case
Any CPU OOM that occurs in the mp design will also occur in the non-mp design (+ we measured that overall memory usage is close)

Anyways:

added a test_load.py to simulate load on the client and try to detect errors like this.
added a test_accuracy.py to run an lm-eval-harness with concurrent requests via the openai API, which simulates real usage and makes sure we get the correct answer out of the server

BEFORE SUBMITTING, PLEASE READ THE CHECKLIST BELOW AND FILL IN THE DESCRIPTION ABOVE

PR Checklist (Click to Expand)

Thank you for your contribution to vLLM! Before submitting the pull request, please ensure the PR meets the following criteria. This helps vLLM maintain the code quality and improve the efficiency of the review process.

PR Title and Classification

Only specific types of PRs will be reviewed. The PR title is prefixed appropriately to indicate the type of change. Please use one of the following:

[Bugfix] for bug fixes.
[CI/Build] for build or continuous integration improvements.
[Doc] for documentation fixes and improvements.
[Model] for adding a new model or improving an existing model. Model name should appear in the title.
[Frontend] For changes on the vLLM frontend (e.g., OpenAI API server, LLM class, etc.)
[Kernel] for changes affecting CUDA kernels or other compute kernels.
[Core] for changes in the core vLLM logic (e.g., LLMEngine, AsyncLLMEngine, Scheduler, etc.)
[Hardware][Vendor] for hardware-specific changes. Vendor name should appear in the prefix (e.g., [Hardware][AMD]).
[Misc] for PRs that do not fit the above categories. Please use this sparingly.

Note: If the PR spans more than one category, please include all relevant prefixes.

Code Quality

The PR need to meet the following code quality standards:

We adhere to Google Python style guide and Google C++ style guide.
Pass all linter checks. Please use format.sh to format your code.
The code need to be well-documented to ensure future contributors can easily understand the code.
Include sufficient tests to ensure the project to stay correct and robust. This includes both unit tests and integration tests.
Please add documentation to docs/source/ if the PR modifies the user-facing behaviors of vLLM. It helps vLLM user understand and utilize the new features or changes.

Notes for Large Changes

Please keep the changes as concise as possible. For major architectural changes (>500 LOC excluding kernel/data/config/test), we would expect a GitHub issue (RFC) discussing the technical design and justification. Otherwise, we will tag it with rfc-required and might not go through the PR.

What to Expect for the Reviews

The goal of the vLLM team is to be a transparent reviewing machine. We would like to make the review process transparent and efficient and make sure no contributor feel confused or frustrated. However, the vLLM team is small, so we need to prioritize some PRs over others. Here is what you can expect from the review process:

After the PR is submitted, the PR will be assigned to a reviewer. Every reviewer will pick up the PRs based on their expertise and availability.
After the PR is assigned, the reviewer will provide status update every 2-3 days. If the PR is not reviewed within 7 days, please feel free to ping the reviewer or the vLLM team.
After the review, the reviewer will put an action-required label on the PR if there are changes required. The contributor should address the comments and ping the reviewer to re-review the PR.
Please respond to all comments within a reasonable time frame. If a comment isn't clear or you disagree with a suggestion, feel free to ask for clarification or discuss the suggestion.

Thank You

Finally, thank you for taking the time to read these guidelines and for your interest in contributing to vLLM. Your contributions make vLLM a great tool for everyone!

github-actions · 2024-08-10T14:51:21Z

👋 Hi! Thank you for contributing to the vLLM project.
Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which consists a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of default ones by unblocking the steps in your fast-check build on Buildkite UI.

Once the PR is approved and ready to go, please make sure to run full CI as it is required to merge (or just use auto-merge).

To run full CI, you can do one of these:

Comment /ready on the PR
Add ready label to the PR
Enable auto-merge.

🚀

robertgshaw2-redhat · 2024-08-10T18:21:56Z

After further investigation, I using zmq sockets will impose a hard limit on the number of active request.

While it is possible to change the value of ZMQ_MAX_SOCKET on a Context by Context basis, the via zmq_ctx_set, ZMQ_SOCKET_LIMIT, which puts an upper bound on this. We are unable to change the value of ZMQ_SOCKET_LIMIT (see: https://libzmq.readthedocs.io/en/latest/zmq_ctx_get.html)

We are unable to change the value:

import zmq
ctx = zmq.Context()
ctx.get(zmq.SOCKET_LIMIT)
# >>> 65535
ctx.set(zmq.SOCKET_LIMIT, 10000000)
ctx.get(zmq.SOCKET_LIMIT)
# >>> 65535

I believe this will put a hard limit on ZMQ at 65k concurrent connections.

This is obviously a lot of requests, but I am worried about an offline batch use case where someone sends 1M requests to
the server at once.

Other Options

Set --limit-concurrency in uvicorn to explicitly prohibit too many requests from being active in the server. This will give 503s if the Server is getting too many requests. (https://www.uvicorn.org/settings/#resource-limits)
Try to handle this in the OAI server (by buffering)

WDYT?

vllm/entrypoints/openai/rpc/client.py

Co-authored-by: Nick Hill <[email protected]>

youkaichao · 2024-08-11T18:29:51Z

@robertgshaw2-neuralmagic thanks for the great work!

I think limiting the total number of pending requests make sense to me. Many web servers should have similar constraints.

As for offline batching inference, we don't need to surface this to users. We can accept arbitrary number of requests, but send part of them to the engine every time.

BTW, offline batch inference does not use this api server. So it should not be a concern there.

For the technical review, I would like to hand it over to @njhill who has better expertise here.

robertgshaw2-redhat · 2024-08-11T18:32:14Z

Also note --- I tried running with v0.5.3 and we seem to be dropping requests as is if we get >20k connections. I don't quite see why / what causes this without zeromq, but it means that our existing state did not support this either

robertgshaw2-redhat · 2024-08-11T18:38:59Z

tests/tracing/test_tracing.py

@@ -114,5 +114,5 @@ def test_traces(trace_service):
        SpanAttributes.LLM_LATENCY_TIME_TO_FIRST_TOKEN) == ttft
    e2e_time = metrics.finished_time - metrics.arrival_time
    assert attributes.get(SpanAttributes.LLM_LATENCY_E2E) == e2e_time
-    assert attributes.get(SpanAttributes.LLM_LATENCY_TIME_IN_SCHEDULER
-                          ) == metrics.scheduler_time
+    assert attributes.get(


make ./format happy

joerunde · 2024-08-12T16:22:38Z

Set --limit-concurrency in uvicorn to explicitly prohibit too many requests from being active in the server. This will give 503s if the Server is getting too many requests. (https://www.uvicorn.org/settings/#resource-limits)

@robertgshaw2-neuralmagic That would be great, the only issue I'd see is if the /health endpoint is used for a liveness probe, that could cause pod restarts under load. So I think we'd just want to also add a few words in the documentation recommending not to do that. (Looking at all the IBM deployment configs 😉 )

joerunde · 2024-08-12T16:31:06Z

vllm/entrypoints/openai/rpc/server.py

@@ -177,11 +174,11 @@ async def run_server_loop(self):
        running_tasks = set()
        while True:
            # Wait for a request.
-            identity, message = await self.socket.recv_multipart()
+            identity, part2, message = await self.socket.recv_multipart()


For future readers it'd be nice to add a link to some zmq docs here or give this a descriptive name to say what part2 is. From context here I'm guessing this is routing information for the client-side proxy?

Its related to the use of ROUTER, will do

robertgshaw2-redhat · 2024-08-12T16:41:12Z

Set --limit-concurrency in uvicorn to explicitly prohibit too many requests from being active in the server. This will give 503s if the Server is getting too many requests. (https://www.uvicorn.org/settings/#resource-limits)

@robertgshaw2-neuralmagic That would be great, the only issue I'd see is if the /health endpoint is used for a liveness probe, that could cause pod restarts under load. So I think we'd just want to also add a few words in the documentation recommending not to do that. (Looking at all the IBM deployment configs 😉 )

hmmmm - can you explain more?

joerunde · 2024-08-12T16:57:37Z

Set --limit-concurrency in uvicorn to explicitly prohibit too many requests from being active in the server. This will give 503s if the Server is getting too many requests. (https://www.uvicorn.org/settings/#resource-limits)

@robertgshaw2-neuralmagic That would be great, the only issue I'd see is if the /health endpoint is used for a liveness probe, that could cause pod restarts under load. So I think we'd just want to also add a few words in the documentation recommending not to do that. (Looking at all the IBM deployment configs 😉 )

hmmmm - can you explain more?

It's probably not a huge deal since vllm itself is a library and doesn't ship k8s configs but, in general readiness probes are used to back off traffic from pods experiencing issues accepting new requests, while liveness probes are used to kill pods that have crashed or experienced some unrecoverable error. If we were to set --limit-concurrency, then the /health endpoint would also start returning 503 when the pod is busy processing the maximum number of requests. That would be great to use as a readiness probe, k8s would automatically route traffic away while the pod is busy. But, if somebody accidentally used the /health endpoint as a liveness probe, then k8s would reboot the pod when it's busy. This would then likely cause that traffic to hit other pods, causing them to reach the maximum request limit and restart, and you'd end up in a cascading failure.

I think we'd want to suggest something different to use as a liveness probe in that case, maybe something as simple as checking that the frontend and backend processes are running.

robertgshaw2-redhat · 2024-08-12T17:04:19Z

Set --limit-concurrency in uvicorn to explicitly prohibit too many requests from being active in the server. This will give 503s if the Server is getting too many requests. (https://www.uvicorn.org/settings/#resource-limits)

@robertgshaw2-neuralmagic That would be great, the only issue I'd see is if the /health endpoint is used for a liveness probe, that could cause pod restarts under load. So I think we'd just want to also add a few words in the documentation recommending not to do that. (Looking at all the IBM deployment configs 😉 )

hmmmm - can you explain more?

It's probably not a huge deal since vllm itself is a library and doesn't ship k8s configs but, in general readiness probes are used to back off traffic from pods experiencing issues accepting new requests, while liveness probes are used to kill pods that have crashed or experienced some unrecoverable error. If we were to set --limit-concurrency, then the /health endpoint would also start returning 503 when the pod is busy processing the maximum number of requests. That would be great to use as a readiness probe, k8s would automatically route traffic away while the pod is busy. But, if somebody accidentally used the /health endpoint as a liveness probe, then k8s would reboot the pod when it's busy. This would then likely cause that traffic to hit other pods, causing them to reach the maximum request limit and restart, and you'd end up in a cascading failure.

I think we'd want to suggest something different to use as a liveness probe in that case, maybe something as simple as checking that the frontend and backend processes are running.

Thanks - this is clear

robertgshaw2-redhat · 2024-08-21T02:49:53Z

.buildkite/test-pipeline.yaml

@@ -86,6 +86,7 @@ steps:
  - vllm/
  commands:
  - pip install -e ./plugins/vllm_add_dummy_model
+  - pip install git+https://github.com/EleutherAI/lm-evaluation-harness.git@a4987bba6e9e9b3f22bd3a6c1ecf0abd04fd5622#egg=lm_eval[api]


need to install from source since local-completions api with support for concurrent requests is not yet in release of lm_eval

robertgshaw2-redhat · 2024-08-21T02:51:18Z

vllm/entrypoints/openai/rpc/client.py

+        self.from_api_server.bind(INPROC_PROXY_PATH)
+
+        # Asyncio background task for the proxy.
+        self.proxy_task = asyncio.create_task(


@njhill does this need to be explicitly canceled somewhere?

(e.g. in close())

@robertgshaw2-neuralmagic yes, we should cancel it there.

robertgshaw2-redhat · 2024-08-21T02:53:15Z

tests/entrypoints/openai/test_load.py

@@ -0,0 +1,105 @@
+"""
+This file tests significant load on the vLLM server.


cc @simon-mo --- this test takes ~3 minutes on H100

Will likely take >10 min on L4 ... are you okay with this?

robertgshaw2-redhat · 2024-08-21T12:46:35Z

test_load.py is failing in the CI due to Too Many Open Files ---> It seems like this happens at the very beginning which is odd ... looking into it

…vllm into fix-zmq-max-sockets

robertgshaw2-redhat · 2024-08-21T14:30:24Z

Should be merged after: #7698

njhill

Thanks @robertgshaw2-neuralmagic!

…llm-project#7394) Co-authored-by: Nick Hill <[email protected]>

…llm-project#7394) Co-authored-by: Nick Hill <[email protected]> Signed-off-by: Alvant <[email protected]>

…llm-project#7394) Co-authored-by: Nick Hill <[email protected]>

added proxy to limit use of uniz sockets

b2e29a5

robertgshaw2-redhat changed the title ~~[Bugfix][Frontend] Fix zeromq Frontend At High QPS~~ [Bugfix][Frontend][Performance] Reduce Socket Usage for zeromq Frontend Aug 10, 2024

Merge branch 'main' into fix-zmq-max-sockets

8d31115

robertgshaw2-redhat changed the title ~~[Bugfix][Frontend][Performance] Reduce Socket Usage for zeromq Frontend~~ [Frontend][Performance] Reduce Unix Socket Usage for zeromq Frontend Aug 10, 2024

robertgshaw2-redhat changed the title ~~[Frontend][Performance] Reduce Unix Socket Usage for zeromq Frontend~~ [Frontend][Performance] Reduce Unix Sockets for zeromq Frontend Aug 10, 2024

robertgshaw2-redhat added 2 commits August 10, 2024 16:02

comment

6d2b3df

use random inproc path

c73e943

robertgshaw2-redhat changed the title ~~[Frontend][Performance] Reduce Unix Sockets for zeromq Frontend~~ [Bugfix][Frontend] Fix Issues Under High Load With zeromq Frontend Aug 10, 2024

format

f1768fb

robertgshaw2-redhat requested review from njhill, simon-mo and youkaichao August 10, 2024 16:08

foamt

601a461

format

1a47d94

njhill reviewed Aug 10, 2024

View reviewed changes

vllm/entrypoints/openai/rpc/client.py Outdated Show resolved Hide resolved

Update vllm/entrypoints/openai/rpc/client.py

eeecb09

Co-authored-by: Nick Hill <[email protected]>

robertgshaw2-redhat commented Aug 11, 2024

View reviewed changes

joerunde reviewed Aug 12, 2024

View reviewed changes

njhill mentioned this pull request Aug 12, 2024

[RFC]: Isolate OpenAI Server Into Separate Process #6797

Closed

joerunde mentioned this pull request Aug 12, 2024

[Bug][Frontend] Improve ZMQ client robustness #7443

Merged

robertgshaw2-redhat mentioned this pull request Aug 13, 2024

Release v0.5.5 #7481

Closed

cleaning

2770e40

robertgshaw2-redhat added 4 commits August 21, 2024 02:42

added test accuracy

a9ecaa9

format

6f8d5e8

updated test pipeline

bab177f

fix lm eval

7b58281

robertgshaw2-redhat commented Aug 21, 2024

View reviewed changes

robertgshaw2-redhat added 6 commits August 21, 2024 04:13

cleanup

adf45d1

updated

9e827b0

Merge branch 'main' into fix-zmq-max-sockets

47dca36

added sleep time

f84c341

actually sleep

0ce78f8

formatting

8054348

robertgshaw2-redhat added 8 commits August 21, 2024 12:47

format

5ddbdab

mypy

1ebbe9e

mypy

53d639b

format

a36b381

remove test load

415ee39

stash

26440e6

Merge branch 'fix-zmq-max-sockets' of https://github.com/neuralmagic/…

2442a9d

…vllm into fix-zmq-max-sockets

Merge branch 'fix-raise-cancelled' into fix-zmq-max-sockets

b72f84f

njhill approved these changes Aug 21, 2024

View reviewed changes

robertgshaw2-redhat merged commit f7e3b0c into vllm-project:main Aug 21, 2024
64 checks passed

omrishiv pushed a commit to omrishiv/vllm that referenced this pull request Aug 26, 2024

[Bugfix][Frontend] Fix Issues Under High Load With zeromq Frontend (v…

e03135e

…llm-project#7394) Co-authored-by: Nick Hill <[email protected]>

zifeitong mentioned this pull request Aug 27, 2024

[Bug]: OpenAI server errors out with "ZMQError Too many open files" under heavy load #7920

Closed

1 task

Alvant pushed a commit to compressa-ai/vllm that referenced this pull request Oct 26, 2024

[Bugfix][Frontend] Fix Issues Under High Load With zeromq Frontend (v…

8446f48

…llm-project#7394) Co-authored-by: Nick Hill <[email protected]> Signed-off-by: Alvant <[email protected]>

KuntaiDu pushed a commit to KuntaiDu/vllm that referenced this pull request Nov 20, 2024

[Bugfix][Frontend] Fix Issues Under High Load With zeromq Frontend (v…

ddc86fa

…llm-project#7394) Co-authored-by: Nick Hill <[email protected]>

AlpinDale mentioned this pull request Dec 12, 2024

api: fix crashes under very high loads aphrodite-engine/aphrodite-engine#878

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Bugfix][Frontend] Fix Issues Under High Load With `zeromq` Frontend #7394

[Bugfix][Frontend] Fix Issues Under High Load With `zeromq` Frontend #7394

robertgshaw2-redhat commented Aug 10, 2024 •

edited

Loading

github-actions bot commented Aug 10, 2024

robertgshaw2-redhat commented Aug 10, 2024 •

edited

Loading

youkaichao commented Aug 11, 2024

robertgshaw2-redhat commented Aug 11, 2024

robertgshaw2-redhat Aug 11, 2024 •

edited

Loading

joerunde commented Aug 12, 2024

joerunde Aug 12, 2024

robertgshaw2-redhat Aug 12, 2024

robertgshaw2-redhat commented Aug 12, 2024

joerunde commented Aug 12, 2024

robertgshaw2-redhat commented Aug 12, 2024

robertgshaw2-redhat Aug 21, 2024 •

edited

Loading

robertgshaw2-redhat Aug 21, 2024

njhill Aug 21, 2024

robertgshaw2-redhat Aug 21, 2024

robertgshaw2-redhat commented Aug 21, 2024

robertgshaw2-redhat commented Aug 21, 2024

njhill left a comment

		@@ -0,0 +1,105 @@
		"""
		This file tests significant load on the vLLM server.

[Bugfix][Frontend] Fix Issues Under High Load With zeromq Frontend #7394

[Bugfix][Frontend] Fix Issues Under High Load With zeromq Frontend #7394

Conversation

robertgshaw2-redhat commented Aug 10, 2024 • edited Loading

BUG REPO:

SUMMARY:

NOTES:

PERFORMANCE:

UPDATE 8/20 POST OFFLINE DISCUSSION

PR Title and Classification

Code Quality

Notes for Large Changes

What to Expect for the Reviews

Thank You

github-actions bot commented Aug 10, 2024

robertgshaw2-redhat commented Aug 10, 2024 • edited Loading

youkaichao commented Aug 11, 2024

robertgshaw2-redhat commented Aug 11, 2024

robertgshaw2-redhat Aug 11, 2024 • edited Loading

Choose a reason for hiding this comment

joerunde commented Aug 12, 2024

joerunde Aug 12, 2024

Choose a reason for hiding this comment

robertgshaw2-redhat Aug 12, 2024

Choose a reason for hiding this comment

robertgshaw2-redhat commented Aug 12, 2024

joerunde commented Aug 12, 2024

robertgshaw2-redhat commented Aug 12, 2024

robertgshaw2-redhat Aug 21, 2024 • edited Loading

Choose a reason for hiding this comment

robertgshaw2-redhat Aug 21, 2024

Choose a reason for hiding this comment

njhill Aug 21, 2024

Choose a reason for hiding this comment

robertgshaw2-redhat Aug 21, 2024

Choose a reason for hiding this comment

robertgshaw2-redhat commented Aug 21, 2024

robertgshaw2-redhat commented Aug 21, 2024

njhill left a comment

Choose a reason for hiding this comment

[Bugfix][Frontend] Fix Issues Under High Load With `zeromq` Frontend #7394

[Bugfix][Frontend] Fix Issues Under High Load With `zeromq` Frontend #7394

robertgshaw2-redhat commented Aug 10, 2024 •

edited

Loading

robertgshaw2-redhat commented Aug 10, 2024 •

edited

Loading

robertgshaw2-redhat Aug 11, 2024 •

edited

Loading

robertgshaw2-redhat Aug 21, 2024 •

edited

Loading