Get desktop application:
View/edit binary Protocol Buffers messages
A periodic reporting of the functions being executed by each executor, and how many resources each function is consuming.
The list of functions on which statistics are being reported in this message.
The list of DAGs on which statistics are being reported in this message.
Statistics regarding an entire DAG request, including call frequency, arrival rates, and end-to-end runtime.
Used in:
The name of the DAG for which statistics are reported.
A list of how long each request to this DAG took to run (end-to-end, including the schedule).
A list of the interval between the arrival times of each sequential pair of request.
The number of calls to this DAG received by a scheduler.
Statistics regarding an individual function -- specifically, call frequency and aggregated runtime.
Used in:
The name of the function being tracked.
The number of calls this function received.
A list of how long each request to this function took to run.
An enum representing the types of resources this executor thread has access to. Currently, we only support CPUs and GPUs.
Used in:
The default ExecutorType for all threads, which says that this executor thread only has a CPU available.
Indicates that this executor thread has access to a GPU.
A message sent by the scheduler to tell an executor thread to pin a function locally.
The name of the function to pin.
The address to which the executor should accept or reject the pin request.
Whethher or not this function supports batching.
An update shared between schedulers about what DAGs they are aware of and what functions are being run by what executors.
A list of the names of DAGs that have been registered with this scheduler.
A list of function location information that this scheduler is aware of.
The location of an individual function, along with the IP address and thread ID it's running on.
Used in:
The name of the function.
The IP address at which it is running.
The ID of the thread on which it is running.
A status update sent from executors to schedulers, giving the schedulers useful information about what resources are available in the system.
The IP address of the executor sending the update.
The thread ID of the scheduler sending the update.
The list of function(s) that are pinned on this executor.
The average utilization of this executor over the last epoch -- this is used to avoid sending work to executors that are overloaded.
The status of this executor -- if this is set to false, it means the executor is clearing its queues and leaving the system, and the schedulers should no longer send it any work.
The type of resources this executor has access to (see ExecutorType definition for more details).