[TRTLLM-7846][feat] Http disagg-cluster management implemention #7869
Conversation
reasonsolo
force-pushed
the
tllm7846_sdhttp
branch
6 times, most recently
from
97215ab to
93ab23c
Compare
📝 WalkthroughWalkthroughIntroduces a new autoscaling subsystem and a cluster storage layer with HTTP server/client and watch semantics; adds tests. Also changes ServerRole to inherit from IntEnum. New components include ClusterManager, ClusterWorker, storage abstractions and implementations, and comprehensive unit tests for storage and autoscaling behavior. Changes
Sequence Diagram(s)sequenceDiagram
autonumber
actor Worker as ClusterWorker
participant Storage as ClusterStorage
participant Manager as ClusterManager
Note over Worker,Storage: Registration
Worker->>Storage: set(worker_key, worker_info, ttl)
Storage-->>Worker: ack
Note over Manager,Storage: Watch worker registry
Manager->>Storage: watch(worker_key_prefix)
Storage-->>Manager: WatchEventQueue
loop Heartbeat (periodic)
Worker->>Storage: expire(worker_key, ttl)
Storage-->>Worker: ack
end
Note over Storage,Manager: Change notifications
Storage-->>Manager: WatchEvent(SET/DELETE)
Manager->>Manager: update internal worker maps and counts
sequenceDiagram
autonumber
participant Test as Test Suite
participant HTTP as HttpClusterStorageServer
participant Client as HttpClusterStorageClient
Test->>HTTP: start()
Test->>Client: set(key, value, ttl)
Client->>HTTP: POST /set
HTTP-->>Client: ok
Test->>Client: watch(prefix)
Client->>HTTP: GET /watch (stub/queue)
HTTP-->>Client: WatchEventQueue
Test->>Client: delete(key)
Client->>HTTP: POST /delete
HTTP-->>Client: ok
HTTP-->>Client: notify(DELETE)
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60–90 minutes Pre-merge checks and finishing touches❌ Failed checks (2 warnings)
✅ Passed checks (1 passed)
✨ Finishing touches
🧪 Generate unit tests
Tip 👮 Agentic pre-merge checks are now available in preview!Pro plan users can now enable pre-merge checks in their settings to enforce checklists before merging PRs.
Please see the documentation for more information. Example: reviews:
pre_merge_checks:
custom_checks:
- name: "Undocumented Breaking Changes"
mode: "warning"
instructions: |
Pass/fail criteria: All breaking changes to public APIs, CLI flags, environment variables, configuration keys, database schemas, or HTTP/GraphQL endpoints must be documented in the "Breaking Change" section of the PR description and in CHANGELOG.md. Exclude purely internal or private changes (e.g., code not exported from package entry points or explicitly marked as internal).Please share your feedback with us on this Discord post. Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 3
🧹 Nitpick comments (2)
tests/unittest/serve/test_cluster_storage.py (1)
92-95: Swap the blocking sleep forasyncio.sleep.This coroutine is running on the event loop; the synchronous
time.sleepcall blocks the loop and can delay other async tasks (like the expiry check) for the full duration. Replace it withawait asyncio.sleep(...)so the loop stays cooperative.- time.sleep(1) + await asyncio.sleep(1) - time.sleep(2) + await asyncio.sleep(2)tensorrt_llm/serve/cluster_storage.py (1)
293-297: Make the session cleanup resilient when no event loop is running.
asyncio.get_event_loop()raisesRuntimeErroronce the loop is closed (common at interpreter shutdown), so the destructor will emit unhandled exceptions instead of closing the session. Wrap the lookup and only schedule a close when a running loop is available:- if asyncio.get_event_loop(): - asyncio.run_coroutine_threadsafe(self._session.close(), - asyncio.get_event_loop()) + try: + loop = asyncio.get_running_loop() + except RuntimeError: + return + asyncio.run_coroutine_threadsafe(self._session.close(), loop)This keeps teardown quiet and still closes the client session.
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (5)
tensorrt_llm/llmapi/disagg_utils.py(2 hunks)tensorrt_llm/serve/auto_scaling.py(1 hunks)tensorrt_llm/serve/cluster_storage.py(1 hunks)tests/unittest/serve/test_cluster_manager_worker.py(1 hunks)tests/unittest/serve/test_cluster_storage.py(1 hunks)
🧰 Additional context used
📓 Path-based instructions (3)
**/*.{h,hpp,hh,hxx,cpp,cxx,cc,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Use only spaces, no tabs; indent with 4 spaces.
Files:
tensorrt_llm/llmapi/disagg_utils.pytensorrt_llm/serve/auto_scaling.pytests/unittest/serve/test_cluster_storage.pytests/unittest/serve/test_cluster_manager_worker.pytensorrt_llm/serve/cluster_storage.py
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: Python code must target Python 3.8+.
Indent Python code with 4 spaces; do not use tabs.
Maintain module namespace when importing; prefer 'from package.subpackage import foo' then 'foo.SomeClass()' instead of importing the class directly.
Python filenames should be snake_case (e.g., some_file.py).
Python classes use PascalCase names.
Functions and methods use snake_case names.
Local variables use snake_case; prefix 'k' for variables that start with a number (e.g., k_99th_percentile).
Global variables use upper SNAKE_CASE prefixed with 'G' (e.g., G_MY_GLOBAL).
Constants use upper SNAKE_CASE (e.g., MY_CONSTANT).
Avoid shadowing variables from an outer scope.
Initialize all externally visible members of a class in the constructor.
Prefer docstrings for interfaces that may be used outside a file; comments for in-function or file-local interfaces.
Use Google-style docstrings for classes and functions (Sphinx-parsable).
Document attributes and variables inline so they render under the class/function docstring.
Avoid reflection when a simpler, explicit approach suffices (e.g., avoid dict(**locals()) patterns).
In try/except, catch the most specific exceptions possible.
For duck-typing try/except, keep the try body minimal and use else for the main logic.
Files:
tensorrt_llm/llmapi/disagg_utils.pytensorrt_llm/serve/auto_scaling.pytests/unittest/serve/test_cluster_storage.pytests/unittest/serve/test_cluster_manager_worker.pytensorrt_llm/serve/cluster_storage.py
**/*.{cpp,cxx,cc,h,hpp,hh,hxx,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Prepend the NVIDIA Apache-2.0 copyright header with current year to the top of all source files (e.g., .cpp, .h, .cu, .py).
Files:
tensorrt_llm/llmapi/disagg_utils.pytensorrt_llm/serve/auto_scaling.pytests/unittest/serve/test_cluster_storage.pytests/unittest/serve/test_cluster_manager_worker.pytensorrt_llm/serve/cluster_storage.py
🧬 Code graph analysis (4)
tensorrt_llm/serve/auto_scaling.py (2)
tensorrt_llm/llmapi/disagg_utils.py (1)
ServerRole(19-22)tensorrt_llm/serve/cluster_storage.py (25)
ClusterStorage(57-96)StorageItem(18-23)WatchEvent(32-34)WatchEventType(26-28)key_time(130-131)start(63-64)start(156-159)stop(67-68)stop(161-164)get_prefix(93-96)get_prefix(217-224)get_prefix(354-359)watch(87-88)watch(226-235)watch(370-372)unwatch(90-91)unwatch(237-244)unwatch(374-376)drain(44-54)delete(84-85)delete(207-215)delete(361-368)expire(78-79)expire(200-205)expire(345-349)
tests/unittest/serve/test_cluster_storage.py (1)
tensorrt_llm/serve/cluster_storage.py (24)
HttpClusterStorageServer(134-281)StorageItem(18-23)WatchEvent(32-34)WatchEventType(26-28)create_cluster_storage(99-105)create_cluster_storage_client(108-114)start(63-64)start(156-159)get(81-82)get(188-198)get(351-352)get_prefix(93-96)get_prefix(217-224)get_prefix(354-359)watch(87-88)watch(226-235)watch(370-372)drain(44-54)delete(84-85)delete(207-215)delete(361-368)stop(67-68)stop(161-164)add_routes(145-154)
tests/unittest/serve/test_cluster_manager_worker.py (4)
tensorrt_llm/llmapi/disagg_utils.py (1)
ServerRole(19-22)tensorrt_llm/serve/auto_scaling.py (13)
ClusterManager(33-178)ClusterWorker(181-283)start(45-46)stop(48-49)register_worker(212-247)watch_workers(82-96)worker_id(196-197)is_ready(173-174)current_ctx_worker_num(71-72)current_gen_worker_num(75-76)get_worker_events(102-118)worker_info(200-205)deregister_worker(249-258)tensorrt_llm/serve/cluster_storage.py (7)
WatchEventType(26-28)create_cluster_storage(99-105)create_cluster_storage_client(108-114)start(63-64)start(156-159)stop(67-68)stop(161-164)tests/unittest/serve/test_cluster_storage.py (5)
http_server_storage(174-187)storage_server(194-198)storage_server(205-212)run_in_thread(36-45)storage_client(52-54)
tensorrt_llm/serve/cluster_storage.py (1)
tensorrt_llm/serve/auto_scaling.py (2)
start(45-46)stop(48-49)
🪛 Ruff (0.13.1)
tensorrt_llm/serve/auto_scaling.py
29-29: Unused function argument: role
(ARG001)
113-113: Do not catch blind exception: Exception
(BLE001)
131-131: Avoid specifying long messages outside the exception class
(TRY003)
139-139: Avoid specifying long messages outside the exception class
(TRY003)
161-161: Do not catch blind exception: Exception
(BLE001)
167-167: Avoid specifying long messages outside the exception class
(TRY003)
193-193: Standard pseudo-random generators are not suitable for cryptographic purposes
(S311)
tests/unittest/serve/test_cluster_storage.py
70-70: Unused method argument: storage_server
(ARG002)
78-78: Unused method argument: storage_server
(ARG002)
86-86: Unused method argument: storage_server
(ARG002)
99-99: Unused method argument: storage_server
(ARG002)
116-116: Unused method argument: storage_client
(ARG002)
178-178: Unused function argument: app
(ARG001)
207-207: subprocess call: check for execution of untrusted input
(S603)
208-208: Starting a process with a partial executable path
(S607)
tests/unittest/serve/test_cluster_manager_worker.py
33-33: f-string without any placeholders
Remove extraneous f prefix
(F541)
35-35: f-string without any placeholders
Remove extraneous f prefix
(F541)
37-37: Avoid specifying long messages outside the exception class
(TRY003)
60-60: subprocess call: check for execution of untrusted input
(S603)
61-61: Starting a process with a partial executable path
(S607)
68-68: Avoid specifying long messages outside the exception class
(TRY003)
73-73: Unpacked variable cluster_uri is never used
Prefix it with an underscore or any other dummy variable pattern
(RUF059)
104-104: Avoid equality comparisons to True; use await cluster_manager.is_ready(): for truth checks
Replace with await cluster_manager.is_ready()
(E712)
113-113: Pointless comparison. Did you mean to assign a value? Otherwise, prepend assert or remove it.
(B015)
114-114: Pointless comparison. Did you mean to assign a value? Otherwise, prepend assert or remove it.
(B015)
118-118: Avoid equality comparisons to False; use not await cluster_manager.is_ready(): for false checks
Replace with not await cluster_manager.is_ready()
(E712)
128-128: Avoid equality comparisons to False; use not await cluster_manager.is_ready(): for false checks
Replace with not await cluster_manager.is_ready()
(E712)
137-137: Avoid equality comparisons to True; use await cluster_manager.is_ready(): for truth checks
Replace with await cluster_manager.is_ready()
(E712)
144-144: Avoid equality comparisons to False; use not await cluster_manager.is_ready(): for false checks
Replace with not await cluster_manager.is_ready()
(E712)
151-151: Avoid equality comparisons to False; use not await cluster_manager.is_ready(): for false checks
Replace with not await cluster_manager.is_ready()
(E712)
200-200: Avoid equality comparisons to True; use await cluster_manager.is_ready(): for truth checks
Replace with await cluster_manager.is_ready()
(E712)
203-203: Avoid equality comparisons to True; use await cluster_manager.is_ready(): for truth checks
Replace with await cluster_manager.is_ready()
(E712)
211-211: Avoid equality comparisons to False; use not await cluster_manager.is_ready(): for false checks
Replace with not await cluster_manager.is_ready()
(E712)
tensorrt_llm/serve/cluster_storage.py
57-57: ClusterStorage is an abstract base class, but it has no abstract methods or properties
(B024)
59-60: ClusterStorage.__init__ is an empty method in an abstract base class, but has no abstract decorator
(B027)
63-64: ClusterStorage.start is an empty method in an abstract base class, but has no abstract decorator
(B027)
67-68: ClusterStorage.stop is an empty method in an abstract base class, but has no abstract decorator
(B027)
70-75: ClusterStorage.set is an empty method in an abstract base class, but has no abstract decorator
(B027)
77-77: Comment contains ambiguous ’ (RIGHT SINGLE QUOTATION MARK). Did you mean ``` (GRAVE ACCENT)?
(RUF003)
78-79: ClusterStorage.expire is an empty method in an abstract base class, but has no abstract decorator
(B027)
81-82: ClusterStorage.get is an empty method in an abstract base class, but has no abstract decorator
(B027)
84-85: ClusterStorage.delete is an empty method in an abstract base class, but has no abstract decorator
(B027)
87-88: ClusterStorage.watch is an empty method in an abstract base class, but has no abstract decorator
(B027)
90-91: ClusterStorage.unwatch is an empty method in an abstract base class, but has no abstract decorator
(B027)
93-96: ClusterStorage.get_prefix is an empty method in an abstract base class, but has no abstract decorator
(B027)
105-105: Avoid specifying long messages outside the exception class
(TRY003)
114-114: Avoid specifying long messages outside the exception class
(TRY003)
136-136: Unused method argument: cluster_uri
(ARG002)
136-136: Unused method argument: cluster_name
(ARG002)
243-244: Avoid specifying long messages outside the exception class
(TRY003)
280-280: Do not catch blind exception: Exception
(BLE001)
281-281: Use logging.exception instead of logging.error
Replace with exception
(TRY400)
304-304: Do not use mutable data structures for argument defaults
Replace with None; initialize within function
(B006)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)
- GitHub Check: Pre-commit Check
🔇 Additional comments (1)
tensorrt_llm/llmapi/disagg_utils.py (1)
3-23: IntEnum switch looks good.Moving
ServerRoleto inherit fromIntEnumkeeps existing semantics but now plays nicely with JSON serialization in the autoscaler/storage paths. ✅
|
/bot run |
|
PR_Github #20060 [ run ] triggered by Bot |
|
PR_Github #20060 [ run ] completed with state |
|
/bot run |
|
PR_Github #20692 [ run ] triggered by Bot |
reasonsolo
force-pushed
the
tllm7846_sdhttp
branch
2 times, most recently
from
71be18b to
bcfcd6d
Compare
|
/bot run |
|
PR_Github #20698 [ run ] triggered by Bot |
|
PR_Github #20692 [ run ] completed with state |
|
PR_Github #20698 [ run ] completed with state |
reasonsolo
force-pushed
the
tllm7846_sdhttp
branch
from
bcfcd6d to
d37f425
Compare
|
/bot run |
|
PR_Github #20708 [ run ] triggered by Bot |
|
PR_Github #20708 [ run ] completed with state |
reasonsolo
force-pushed
the
tllm7846_sdhttp
branch
from
d37f425 to
fc43659
Compare
|
/bot run |
|
PR_Github #20718 [ run ] triggered by Bot |
|
PR_Github #20718 [ run ] completed with state |
Signed-off-by: Lizhi Zhou <1432185+reasonsolo@users.noreply.github.com>
reasonsolo
force-pushed
the
tllm7846_sdhttp
branch
2 times, most recently
from
71f6dc5 to
e5271c6
Compare
|
/bot run |
|
PR_Github #20766 [ run ] triggered by Bot |
|
PR_Github #20766 [ run ] completed with state |
Signed-off-by: Lizhi Zhou <1432185+reasonsolo@users.noreply.github.com>
reasonsolo
force-pushed
the
tllm7846_sdhttp
branch
from
e5271c6 to
62c250d
Compare
|
/bot run |
|
PR_Github #20795 Bot args parsing error: usage: /bot [-h] |
|
PR_Github #20796 [ run ] triggered by Bot |
|
PR_Github #20796 [ run ] completed with state |
…IA#7869) Signed-off-by: Lizhi Zhou <1432185+reasonsolo@users.noreply.github.com>
4 participants
This is the first PR for the whole epic, only including the basic tools for service discovery.
Summary by CodeRabbit
New Features
Refactor
Tests
Description
Test Coverage
PR Checklist
Please review the following before submitting your PR:
PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.
PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.
Test cases are provided for new code paths (see test instructions)
Any new dependencies have been scanned for license and vulnerabilities
CODEOWNERS updated if ownership changes
Documentation updated as needed
The reviewers assigned automatically/manually are appropriate for the PR.
Please check this after reviewing the above items as appropriate for this PR.
GitHub Bot Help
/bot [-h] ['run', 'kill', 'skip', 'reuse-pipeline'] ...Provide a user friendly way for developers to interact with a Jenkins server.
Run
/bot [-h|--help]to print this help message.See details below for each supported subcommand.
run [--reuse-test (optional)pipeline-id --disable-fail-fast --skip-test --stage-list "A10-PyTorch-1, xxx" --gpu-type "A30, H100_PCIe" --test-backend "pytorch, cpp" --add-multi-gpu-test --only-multi-gpu-test --disable-multi-gpu-test --post-merge --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" --detailed-log --debug(experimental)]Launch build/test pipelines. All previously running jobs will be killed.
--reuse-test (optional)pipeline-id(OPTIONAL) : Allow the new pipeline to reuse build artifacts and skip successful test stages from a specified pipeline or the last pipeline if no pipeline-id is indicated. If the Git commit ID has changed, this option will be always ignored. The DEFAULT behavior of the bot is to reuse build artifacts and successful test results from the last pipeline.--disable-reuse-test(OPTIONAL) : Explicitly prevent the pipeline from reusing build artifacts and skipping successful test stages from a previous pipeline. Ensure that all builds and tests are run regardless of previous successes.--disable-fail-fast(OPTIONAL) : Disable fail fast on build/tests/infra failures.--skip-test(OPTIONAL) : Skip all test stages, but still run build stages, package stages and sanity check stages. Note: Does NOT update GitHub check status.--stage-list "A10-PyTorch-1, xxx"(OPTIONAL) : Only run the specified test stages. Examples: "A10-PyTorch-1, xxx". Note: Does NOT update GitHub check status.--gpu-type "A30, H100_PCIe"(OPTIONAL) : Only run the test stages on the specified GPU types. Examples: "A30, H100_PCIe". Note: Does NOT update GitHub check status.--test-backend "pytorch, cpp"(OPTIONAL) : Skip test stages which don't match the specified backends. Only support [pytorch, cpp, tensorrt, triton]. Examples: "pytorch, cpp" (does not run test stages with tensorrt or triton backend). Note: Does NOT update GitHub pipeline status.--only-multi-gpu-test(OPTIONAL) : Only run the multi-GPU tests. Note: Does NOT update GitHub check status.--disable-multi-gpu-test(OPTIONAL) : Disable the multi-GPU tests. Note: Does NOT update GitHub check status.--add-multi-gpu-test(OPTIONAL) : Force run the multi-GPU tests in addition to running L0 pre-merge pipeline.--post-merge(OPTIONAL) : Run the L0 post-merge pipeline instead of the ordinary L0 pre-merge pipeline.--extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx"(OPTIONAL) : Run the ordinary L0 pre-merge pipeline and specified test stages. Examples: --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx".--detailed-log(OPTIONAL) : Enable flushing out all logs to the Jenkins console. This will significantly increase the log volume and may slow down the job.--debug(OPTIONAL) : Experimental feature. Enable access to the CI container for debugging purpose. Note: Specify exactly one stage in thestage-listparameter to access the appropriate container environment. Note: Does NOT update GitHub check status.For guidance on mapping tests to stage names, see
docs/source/reference/ci-overview.mdand the
scripts/test_to_stage_mapping.pyhelper.kill
killKill all running builds associated with pull request.
skip
skip --comment COMMENTSkip testing for latest commit on pull request.
--comment "Reason for skipping build/test"is required. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.reuse-pipeline
reuse-pipelineReuse a previous pipeline to validate current commit. This action will also kill all currently running builds associated with the pull request. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.