Skip to content

Latest commit

 

History

History
155 lines (126 loc) · 5.59 KB

README.md

File metadata and controls

155 lines (126 loc) · 5.59 KB
Raw

O² Control and Configuration

Build

Starting from the occ directory.

$ mkdir build && cd build
$ cmake .. -DCMAKE_BUILD_TYPE=Debug -DFAIRMQPATH=<path to FairMQ prefix> -DFAIRLOGGERPATH=<path to FairLogger prefix>
$ make

It can also be built via aliBuild (package Control-OCCPlugin).

Run example

From build dir:

$ occlib/examples/dummy-process/occexample-dummy-process

or

$ occlib/examples/dummy-process/occexample-dummy-process --control-port <some port>

or

$ OCC_CONTROL_PORT=<some port> occlib/examples/dummy-process/occexample-dummy-process

The dummy process now waits for control commands.

The OCC state machine

The figure below describes the OCC state machine, as implemented in the OCC library and exposed by the OCC API for non-FairMQ devices (controlmode.DIRECT).

The OCC state machine

Note: a PAUSED state with events PAUSE/RESUME is foreseen but not used yet.

Single process control with peanut

peanut is the Process Execution And coNtrol UTility for OCClib-based O² processes. Its purpose is to be a debugging and development aid for non-FairMQ O² devices, where FairMQ's interactive controller is not available.

In aliBuild it is part of the coconut package.

peanut can connect to a running OCClib-based process, query its status, drive its state machine and push runtime configuration data.

Peanut is an interactive tool, the only information it picks up from its environment is the OCC_CONTROL_PORT variable, which is used to connect to a running OCClib-based process.

$ OCC_CONTROL_PORT=<some port> peanut

Screenshot of peanut

Peanut commands are documented inline. Each transition is applied immediately and the state is updated in real time.

Compared to the raw gRPC API, the following limitations apply:

  • It is not possible to perform a GO_ERROR transition, as this transition is only triggered from user code.
  • The CONFIGURE transition may be triggered both with and without runtime configuration data, which may or may not be suitable depending on user code. All other transitions send no payload.

The last two commands are not transitions:

  • Load configuration allows the user to read in a JSON or YAML file containing sample configuration data that is then available to be pushed to the controlled process during a future CONFIGURE transition. On startup, there is no file loaded, so a CONFIGURE transition will push an empty payload. Once a runtime configuration file is loaded, its title bar reports NOT PUSHED until the next CONFIGURE transition, at which point it becomes PUSHED.
  • Quit disconnects from the controlled process and quits peanut, but it performs no transitions or other data exchange with the controlled process. A future instance of peanut may reattach itself to the same process and continue from there.

OCC API debugging with grpcc

We can send gRPC-based OCC commands manually with an interactive gRPC client like grpcc:

$ sudo yum install http-parser nodejs npm
$ npm install -g grpcc

In a new terminal, we go to the occ directory (not the build dir) and connect via gRPC:

$ grpcc -i --proto protos/occ.proto --address 127.0.0.1:47100

If all went well, we get an interactive environment like so:


Connecting to occ_pb.Occ on 127.0.0.1:47100. Available globals:

  client - the client connection to Occ
    stateStream (StateStreamRequest, callback) returns StateStreamReply
    getState (GetStateRequest, callback) returns GetStateReply
    transition (TransitionRequest, callback) returns TransitionReply

  printReply - function to easily print a unary call reply (alias: pr)
  streamReply - function to easily print stream call replies (alias: sr)
  createMetadata - convert JS objects into grpc metadata instances (alias: cm)
  printMetadata - function to easily print a unary call's metadata (alias: pm)

[email protected]:47100>

Let's try to send some commands. State changes will be reported in the standard output of the process.

[email protected]:47100> client.getState({}, pr)
{
  "state": "STANDBY"
}
[email protected]:47100> client.transition({srcState:"STANDBY", transitionEvent:"CONFIGURE", arguments:[]}, pr)
{
  "trigger": "EXECUTOR",
  "state": "CONFIGURED",
  "transitionEvent": "CONFIGURE",
  "ok": true
}
[email protected]:47100> client.getState({}, pr)
{
  "state": "CONFIGURED"
}
[email protected]:47100> client.transition({srcState:"CONFIGURED", transitionEvent:"START", arguments:[]}, pr)
{
  "trigger": "EXECUTOR",
  "state": "RUNNING",
  "transitionEvent": "START",
  "ok": true
}
[email protected]:47100> client.transition({srcState:"RUNNING", transitionEvent:"STOP", arguments:[]}, pr)
{
  "trigger": "EXECUTOR",
  "state": "CONFIGURED",
  "transitionEvent": "STOP",
  "ok": true
}
[email protected]:47100> client.transition({srcState:"CONFIGURED", transitionEvent:"EXIT", arguments:[]}, pr)
{
  "trigger": "EXECUTOR",
  "state": "DONE",
  "transitionEvent": "EXIT",
  "ok": true
}
# no further commands possible, EXIT stops the process

Developer reference

  1. Build & install the OCC library either manually or via aliBuild (Control-OCCPlugin);
  2. check out the dummy process example and its entry point and to see how to instantiate OCC;
  3. implement interface at occlib/RuntimeControlledObject.h,
  4. link your non-FairMQ O² process against the target AliceO2::Occ as described in the dummy process README.