package xpu_timer.server

Get desktop application:
View/edit binary Protocol Buffers messages

• rpc RegisterPrometheus (RegisterPrometheusRequest, RegisterPrometheusResponse)

  hosting_service.proto:242

  message RegisterPrometheusRequest

  hosting_service.proto:178

  • string gauge_prefix = 1

  • map<string, string> labels = 2

  • int32 rank = 3

  • string kernel_name = 4

  • int32 local_rank = 5

• rpc DeRegisterPrometheus (DeRegisterPrometheusRequest, RegisterPrometheusResponse)

  hosting_service.proto:244

  message DeRegisterPrometheusRequest

  hosting_service.proto:215

  • string name = 1

  • int32 local_rank = 2

  • int32 rank = 3

• rpc PushPrometheus (BrpcMetrics, google.protobuf.Empty)

  hosting_service.proto:246

  message BrpcMetrics

  hosting_service.proto:170

  • string name = 1

  • int32 local_rank = 2

  • optional CommonBrpcMetrics common_metrics = 3

  • optional KernelBrpcMetrics kernel_metrics = 4

  • optional MemBrpcMetrics mem_metrics = 5

• rpc DumpStringStacktrace (StacktraceRequest, StacktraceResponse)

  hosting_service.proto:247

  message StacktraceRequest

  hosting_service.proto:129

  • int32 pid = 1

  • int32 rank = 2

  • int32 world_size = 3

  • string dump_path = 4

  • string gdb_bin = 5

  • string pyspy_bin = 6

  • string pstack_bin = 7

  • optional HangKernels hang_kernels = 8

  message StacktraceResponse

  hosting_service.proto:140

  • optional Stacktrace structure_stack = 1

    server call dump_driver in new process, we use file to shared the serialized message, maybe use other ipc methods later.

  • string stacktrace = 2

• rpc DumpKernelTrace (DumpKernelTraceRequest, DumpKernelTraceResponse)

  hosting_service.proto:248

  message DumpKernelTraceRequest

  hosting_service.proto:191

  • string dump_path = 1

  • uint32 dump_count = 2

  • int64 dump_time = 3

  • bool reset = 4

  • uint32 dump_kernel_type = 5

    bits, 1 is matmul, 2 is comm

  • optional DumpKernelTraceRequest.OssArgs oss_args = 6

  message DumpKernelTraceResponse

  hosting_service.proto:221

  • string node = 1

  • string msg = 2

• rpc PushSignalFrameInfo (SignalFrameRequest, google.protobuf.Empty)

  hosting_service.proto:249

  message SignalFrameRequest

  hosting_service.proto:226

  • int32 signal = 1

  • int32 rank = 2

  • string ip = 3

  • string pod_name = 4

  • string job_name = 5

  • uint64 timestamp = 6

  • repeated SignalFrameRequest.FrameInfo frame_infos = 7

Used in: BrpcMetrics

• repeated int64 metrics = 1

  arrays for int64 hang = 1; int64 start_dump = 2; int64 end_dump = 3; int64 pool_queue_size = 4; int64 work_queue_size = 5; int64 gc_count = 6;

Used in: DeviceFrame

• string device_func = 1

• string curr_location = 2

• string curr_symbol = 3

• string inlined_location = 4

• string block = 5

• string thread = 6

• string kernel_args = 7

• string sass = 8

Used in: DeviceStacktrace

• oneof frame

  • CudaStackFrame cuda_frame = 1

• string stderr = 2

• string origin = 3

Used in: Stacktrace

• repeated DeviceFrame devices_frames = 1

Used in: Stacktrace

• string block_range = 1

  Block range, e.g., "0-0"

• string thread_range = 2

  Thread range, e.g., "0-0"

• repeated string stack_trace = 3

  Stack trace information

• string sass = 4

  SASS instruction

• bool has_next = 5

  Indicator for next block

• int32 hang_step = 6

  Step where hang occurred

• repeated string hang_sass = 7

  List of hang SASS instructions

• map<string, int32> registers = 8

  Registers as key-value pairs

Used in: DumpKernelTraceRequest

• string oss_ak = 1

• string oss_sk = 2

• string oss_endpoint = 3

• string oss_path = 4

Used in: StacktraceRequest

• repeated string hang_kernels = 1

Used in: BrpcMetrics

• uint64 avg_latency = 1

• uint64 max_latency = 2

• uint64 p99_latency = 3

• uint64 min_latency = 4

• double performance = 5

Used in: BrpcMetrics

• uint64 counter = 1

pstack frame Thread 24 (Thread 0x7eff1f6b4000 (LWP 3920)): #0 0x00007effca957ce5 in pthread_cond_wait@@GLIBC_2.3.2 () from /usr/lib64/libpthread.so.0 #1 0x00007eff328cddeb in blas_thread_server () from /opt/conda/lib/python3.8/site-packages/numpy/core/../../numpy.libs/libopenblas64_p-r0-742d56dc.3.20.so #2 0x00007effca9514e2 in start_thread () from /usr/lib64/libpthread.so.0 #3 0x00007effca72e5b3 in clone () from /usr/lib64/libc.so.6

Used in: PstackStacktrace

• string func_name = 1

• string file_name = 2

• string origin = 3

gdb info threads * 1 process 1 "bash" 0x00007f3dca76093a in waitpid () from /lib64/libc.so.6

Used in: Stacktrace

• uint32 pid = 1

• string thread_name = 2

• repeated PstackFrame frames = 3

Used in: PySpyFrameList, PySpyStacktrace

• string func_name = 1

• string file_name = 2

• string module = 3

Used in: PythonStackInTimeline

• repeated PySpyFrame frames = 1

"pid": 1142, "thread_id": 139620998182656, "thread_name": "Thread-4", "os_thread_id": 4127, "active": false, "owns_gil": false,

Used in: Stacktrace

• uint32 pid = 1

• uint64 thread_id = 2

• string thread_name = 3

• uint32 os_thread_id = 4

• bool active = 5

• bool owns_gil = 6

• repeated PySpyFrame frames = 7

• map<string, PySpyFrameList> named_frames = 1

• int32 rank = 2

Used as response type in: HostingService.DeRegisterPrometheus, HostingService.RegisterPrometheus

• string name = 1

• int32 ret_code = 2

Used in: SignalFrameRequest

• uint64 address = 1

• uint64 offset = 2

• string function_name = 3

Used in: StacktraceResponse

• repeated PstackStacktrace stacktrace = 1

• repeated PySpyStacktrace py_stacktrace = 2

• int32 pid = 3

• int32 rank = 4

• string pstack_stdout = 5

• string pstack_stderr = 6

• string pyspy_stdout = 7

• string pyspy_stderr = 8

• repeated DeviceStacktrace device_stacktrace = 9

• repeated DeviceStatus device_status = 10

• string process_state = 11