Composing microblx systems

Building a microblx application typically involves instantiating blocks, configuring and interconnecting their ports and finally starting all blocks. The recommended way to do this is by specifying the system using the microblx composition DSL.

Microblx System Composition DSL (usc files)

usc are declarative descriptions of microblx systems that can be validated and instantiated using the ubx-launch tool. A usc model describes one microblx system, as illustrated by the following minimal example:

local bd = require("blockdiagram")

return bd.system
{
   -- import microblx modules
   imports = {
      "stdtypes", "ptrig", "lfrb", "myblocks",
   },

   -- describe which blocks to instantiate
   blocks = {
      { name="x1", type="myblocks/x" },
      { name="y1", type="myblocks/y" },
      { name="ptrig1", type="ubx/ptrig" },
      ...
   },

   -- connect blocks
   connections = {
      { src="x1.out", tgt="y1.in" },
      { src="y1.out", tgt="x1.in", buffer_len=16 },
   },

   -- configure blocks
   configurations = {
      { name="x1", config = { cfg1="foo", cfg2=33.4 } },
      { name="y1", config = { cfgA={ p=1,z=22.3 }, cfg2=33.4 } },

      -- configure a trigger
      { name="trig1", config = { period = {sec=0, usec=100000 },
                                 sched_policy="SCHED_OTHER",
                                 sched_priority=0,
                                 chain0={
                                    -- the #<blockname> directive will
                                    -- be resolved to an actual
                                    -- reference to the respective
                                    -- block once instantiated.
                                    -- every=N triggers the block only
                                    -- every N-th step (default: 1).
                                    { b="#x1", num_steps=1, every=1, measure=0 },
                                    { b="#y1", num_steps=1, every=2, measure=0 } } } }
   },
}

ptrig additionally supports affinity (CPU list), stacksize, thread_name, autostop_steps, and sleep_mode (0=OS sleep, 1=busy-wait). The sched_policy field accepts SCHED_OTHER (default), SCHED_FIFO, SCHED_RR, and SCHED_DEADLINE (Linux ≥ 3.14); for SCHED_DEADLINE a sched_deadline config of type struct ptrig_deadline is required (see the README for details). See the trig/ptrig block README for the full configuration reference.

Lua blocks

Blocks implemented in Lua using the luablock module can be instantiated directly in a composition using the luablock: type prefix:

blocks = {
    { name="myblk", type="luablock:myluablock" },
},

When the blockdiagram module encounters a block type with the luablock: prefix, it uses the luablock-util module to locate and instantiate the Lua block. The name after the colon (e.g. myluablock) is resolved by searching for the corresponding .lua file in the standard microblx block prefixes. This means there is no need to manually import the luablock module, create a ubx/luablock instance, or configure lua_file with an absolute path.

Alternatively, short blocks can be defined inline using the lua_str config instead of lua_file.

Luablocks support built-in self-triggering via the thread and period configs — no separate ptrig is required for non-RT use-cases:

{ name="myblk", type="luablock:myluablock",
  config = { thread=1, period=100 } }  -- period in milliseconds

Launching

usc files like the above example can be launched using ubx-launch tool. Run with -h for further information. The following example

$ cd /usr/local/share/ubx/examples/usc/pid/
$ ubx-launch -webgraph -c pid_test.usc,ptrig_nrt.usc
...

will launch the given system composition and additionally start the webgraph block. Browse to http://localhost:8888 to see a live graph of blocks and connections.

Unless the -nostart option is provided, all blocks will be initialized, configured and started. ubx-launch handles this in safe way by starting up active blocks after all other blocks (In earlier versions, there was start directive to list the blocks to be started, however now this information is obtained by means of the block attributes BLOCK_ATTR_ACTIVE and BLOCK_ATTR_TRIGGER.)

Node configs

Node configs allow to assign the same configuration to multiple blocks. This is useful to avoid repeating global configuration values that are identical for multiple blocks.

The node_configurations keyword allows to define one or more named node configurations.

node_configurations = {
    global_rnd_conf = {
        type = "struct random_config",
        config = { min=333, max=999 },
    }
}

These configurations can then be assigned to multiple blocks:

{ name="b1", config = { min_max_config = "&global_rnd_conf"} },
{ name="b2", config = { min_max_config = "&global_rnd_conf"} },

Please refer to examples/systemmodels/node_config_demo.usc for a full example.

Connections

The powerful connections keyword supports connecting blocks in multiple ways:

• cblocks to cblocks

• cblocks to iblocks

• cblocks to non-existing iblocks (the latter are created on the fly)

The syntax for these variants is discussed below.

cblock to cblock connections

The following example shows how to create ports among cblock ports:

{ src="blkA.portX", tgt="blkB.portY", type="lfrb", config = { ... }

• both src and tgt are of the form CBLOCK.PORT. Both blocks and ports must exist.

• type specifies the type of iblock to create for the connection. If unset it defaults to ubx/lfrb

• config is the optional configuration to apply to the newly created iblock. The configs type_name and data_len are set automatically unless specified.

cblock to iblock

The following examples illustrates creating connections to/from an existing iblock myMq:

{ src="blkX.portZ", tgt="myMQ" }

-- or

{ src="myMQ", tgt="blkX.portZ" }

• the iblock must exist and be of the form IBLOCK (i.e. no port).

• the cblock must exist and be of the form CBLOCK.PORT

• type and config must not be set (they will be ignored with a warning).

cblock to non-existing iblock

The following example creates a new mqueue with an automatic, unique name, configures it with config and connect blkX.portZ to it:

{ src="blkX.portZ", type="ubx/mqueue", config={ buffer_len=32 } }

• type must be set to desired iblock type and one of src or tgt must be unset

• type_name, data_len and buffer_len are set automatically unless defined in config.

• for type ubx/mqueue: if no mq_id is set in config, then mq_id is set to the corresponding peer “BLOCK.PORT”, e.g. to blkX.portZ in the example above.

This form is useful to create one-line connections via mqueues or similar.

External blocks

The extern_blocks keyword declares blocks that are expected to already exist in the node when the composition is launched. This is useful when loading an auxiliary composition into a running node (e.g. via the LoadUSCLua or LoadUSCJSON lsdb interface methods) that needs to connect to blocks from the already running core composition.

return bd.system
{
    imports = { "stdtypes", "lfrb", "myblocks" },

    extern_blocks = { "core_sensor", "core_actuator" },

    blocks = {
        { name="controller", type="myblocks/ctrl" },
    },

    connections = {
        { src="core_sensor.out", tgt="controller.in" },
        { src="controller.out", tgt="core_actuator.in" },
    },

    configurations = {
        { name="controller", config = { gain=1.5 } },
    },
}

Blocks listed in extern_blocks can be referenced in connections just like blocks defined in blocks. Validation will reject references to blocks that are in neither blocks nor extern_blocks, so typos are still caught at model checking time.

Hierarchical compositions

Using hierarchical composition [1] an application can be composed from other compositions. The motivation is to permit reuse of the individual compositions.

The subsystems keyword accepts a list of namespace-subsystem entries:

return bd.system {
    import = ...
    subsystems = {
        subsys1 = bd.load("subsys1.usc"),
        subsys2 = bd.load("subsys1.usc"),
    }
}

Subsystem elements like configs can be accessed by higher levels by adding the subsystem namespace. For example, the following lines override a configuration value of the blk block in subsystems sub11 and sub11/sub21:

configurations = {
    { name="sub11/blk",       config = { cfgA=1, cfgB=2 } },
    { name="sub11/sub21/blk", config = { cfgA=5, cfgB=6 } },
}

Note how the subsystem namespaces prevent name collisions of the two identically names blocks. Similar to configurations, connections can be added among subsystems blocks:

connections = {
    { src="sub11/sub21/blk.portX", tgt="sub11/blk.portY" },
},

When launched, a hierarchical system is instantiated in a similar way to a non-hierarchical one, however:

• modules are only imported once

• blocks from all hierarchy levels are instantiated, configured and started together, i.e. the hierarchy has no implications on the startup sequence.

• microblx block names use the fully qualified name including the namespace. Therefore, the #blockname syntax for resolving block pointers works just the same.

• if multiple configs for the same block exist, only the highest one in the hierarchy will be applied.

• node configs are always global, hence no prefix is required. In case of multiple identically named node configs, the one at the highest level will be selected.

Merging subsystems

It is possible to add a subsystem without a namespace, as shown by the following snippet:

return bd.system {
    subsystems = {
        bd.load("subsys1.usc"),
    }
}

In this case, the subsys1.usc system will be merged directly into the parent system. Note that entries of the parent system take precedence, so in case of conflicts elements of the subsystem will be skipped.

This feature is useful to avoid an extra hierarchy level.

Model mixins

To obtain a reusable composition, it is important to avoid introducing platform specifics such as ptrig blocks and their configurations. Instead, passive trig blocks can be used to encapsulate the trigger schedule. ptrig or similar active blocks can then be added at launch time by merging them (encapsulated in an usc file) into the primary model by specifying both on the ubx-launch command line.

For example, consider the example in examples/systemmodels/composition:

ubx-launch -webgraph -c deep_composition.usc,ptrig.usc

Note: unlike merging from within the usc using an unnamed subsystems entry (see Merging subsystems), models merged on the command line will override existing entries.

Alternatives

Although using usc model is the preferred approach, there are others way to launch a microblx application:

Launching in C

It is possible to avoid the Lua scripting layer entirely and launch an application in C/C++. A small self-contained example c-launch.c is available under examples/C/ (see the README for further details).

For a more complete example, checkout the respective tutorial section Deployment via C program. Please note that such launching code is a likely candidate for code generation and there are plans for a usc-to-C compiler. Please ask on the mailing if you are interested.

Lua scripts

One can write a Lua “deployment script” similar to the ubx-launch. Checkout the scripts in the tools section. This approach not recommended under normally, but can be useful in specific cases such as for building dedicated test tools.

Footnotes

[1]

This feature was introduced in the context of the COCORF RobMoSys Integrated Technical Project. Please see docs/dev/001-blockdiagram-composition.md for background information.