stepfunc / dnp3 Goto Github PK

We should limit the number of class type resources in our APIs whenever possible as they have to be destroyed in C and handled specially in C#/Java.

An association is really just an address token and as far as the ffi is concerned can just be:

struct Association {
   address: u16
}

All of the methods on Association can be moved to Master and take an additional Association argument. A rust association is just a MasterHandle paired with an EndpointAddress anyway:

#[derive(Clone, Debug)]
pub struct AssociationHandle {
    address: EndpointAddress,
    master: MasterHandle,
}

This change will eliminate resource leaks related to not freeing associations. It's much easier for the user to just have to deal with freeing a master or outstation.

I'd like to use this in a few places in the docs, namely master startup, diagrams of the runtime, etc.

https://mermaid-js.github.io/mermaid/#/

Similar to dnp3/opendnp3#277.

Instead of abandonning an AutoTask if it failed, we should provide a retry strategy with a delay. OpenDNP3 has a exponential backoff strategy that can be configured with https://github.com/dnp3/opendnp3/blob/097b6ddcb012a54c9e8e10db03fec98ddc898341/cpp/lib/include/opendnp3/master/MasterParams.h#L65-L69

I found myself wanting to pull in the License.txt into the documentation without having anchors.

Tracing can provide additional context, e.g. each connection can be identified in some way, whereas the log crate has no such info and will make it impossible to correlate logs with sessions without injecting identifiers down into the log calls themselves.

• link-layer
• transport

Do we need to the raw hexadecimal?

• Ignore restart IIN
• Define which classes should be requested in an event scan when detecting corresponding events available IIN
• Define the time delay before failing a non-recurring task (e.g. commands) that cannot start
• Index qualifier code optimization (one-byte vs two-byte)

Update to tokio 2.23 when it's released. We currently use a git dependency in order to have this change: tokio-rs/tokio#2649

We need both APIs to give full control over how objects/headers are constructed.

It should be possible to have multiple objects per object header.

e.g. at the moment we have runtime_config_t but logging_configuration_t

With tracing integrated, it's worth going through each log statement in the library to determine if context in the log statement should be moved into a parent span instead.

See https://github.com/stepfunc/dnp3/blob/master/dnp3/src/outstation/database/read.rs#L448-L456

• See https://github.com/dnp3/opendnp3/blob/097b6ddcb012a54c9e8e10db03fec98ddc898341/cpp/lib/include/opendnp3/link/ILinkListener.h#L32-L52
• Callbacks must be able to collect these statistics: https://github.com/dnp3/opendnp3/blob/097b6ddcb012a54c9e8e10db03fec98ddc898341/cpp/lib/include/opendnp3/StackStatistics.h#L33-L79 but in a more versatile callback based approach.

• Expose AssociationHandler::get_system_time in the bindings
• Modify AssociationHandler::get_system_time to return an Option<SystemTime>. If it is None, then no time synchronization should happen. See dnp3/opendnp3#397

Setting the log handler more than once panics with unable to install tracing subscriber here:

The tracing docs of tracing::subscriber::set_global_default says:

Libraries should NOT call set_global_default()! That will cause conflicts when executables try to set them later.

This group is supported by OpenDNP3, but it's very rarely used.

The outstation follows this pattern but the master creation methods are on the Runtime.

This is required for conformance testing, although rarely used.

See also: dnp3/opendnp3#405 (comment)

Let's customize the syntax colors to be more on-brand:

The default background appears to be the same purplish thing used in the footer.

• the comment color could stand out a bit more
• we don't need a separate color for bool/literal, at least we can get rid of the red

I have reserved the io.stepfunc groupId on Maven central.

https://issues.sonatype.org/browse/OSSRH-63043

Add the ability to have a custom reconnection strategy, instead of the hard-coded exponential backoff.

Allowing single port to map to multiple outstation sessions, based on IP address filter or wildcard.

This ability is needed to pull in the CHANGLOG.md. I prototyped how this would look with a chunk of the opendnp3 changelog,
and it looks great. It even creates links to the versions in the right sidebar!

In the FFI, the command callback yields the following enum:

/// @brief Result of a command
typedef enum command_result_t
{
    /// @brief Command was a success
    CommandResult_Success = 0,
    /// @brief Failed b/c of a generic task execution error
    CommandResult_TaskError = 1,
    /// @brief Outstation indicated that a command was not SUCCESS
    CommandResult_BadStatus = 2,
    /// @brief Number of headers in the response doesn't match the number in the request
    CommandResult_HeaderCountMismatch = 3,
    /// @brief Header in the response doesn't match the request
    CommandResult_HeaderTypeMismatch = 4,
    /// @brief Number of objects in one of the headers doesn't match the request
    CommandResult_ObjectCountMismatch = 5,
    /// @brief Value in one of the objects in the response doesn't match the request
    CommandResult_ObjectValueMismatch = 6,
} command_result_t;

Its a little strange that we have 4 different kinds of details about why the device might not respond to a select/operate correctly, i.e. all the header count mismatches, yet there's just a single TaskError. We might consider rolling up all the "mismatch" errors into something like "BadReply".

The TaskError should at least be able to differentiate between "response timeout" and "disconnected".

Uncomment // missing_docs in lib.rs and add missing rustdoc

The octet string read handler method is currently doing nothing. See https://github.com/grid-io/dnp3/blob/1739253b8afdbbef262fc68f1fdf5bb7756bf12b/ffi/dnp3-ffi/src/handler.rs#L76-L82

It should accept just a function code and request headers. It could be used for class assignement or to freeze counters for example.

See https://github.com/dnp3/opendnp3/blob/097b6ddcb012a54c9e8e10db03fec98ddc898341/cpp/lib/include/opendnp3/master/IMasterOperations.h#L150-L153

Preludes:

use dnp3::prelude::*;
use dnp3::prelude::master::*;
use dnp3::prelude::outstation::*;

Also, move entry functions to master and outstation modules.

Also add all missing public Rustdoc

See https://github.com/dnp3/opendnp3/blob/097b6ddcb012a54c9e8e10db03fec98ddc898341/cpp/lib/include/opendnp3/master/MasterParams.h#L74-L78

• Should be configurable
• Be careful with this: dnp3/opendnp3#393

The callbacks should return the time-to-complete value.

We don't need the separate light/dark toggle. Let's just focus on a single light theme to match the website.

Parameterize the link layer parser so that it bubbles up or swallows framing errors. TCP/TLS should bubble up errors and close the stream by default, but it shouldn't be hard coded by channel type (except for perhaps serial) b/c of the use of terminal servers which can convert serial to IP.

This was noted in #35 but wasn't implemented in it.

Add performance benchmark where a set of masters communicate with an outstation as fast as possible using criterion to collect statistics.

• See https://github.com/dnp3/opendnp3-dnp4s-driver for inspiration

• Add an optional delay before attempting a new connection. See dnp3/opendnp3#354
• Add the ability to have a custom reconnection strategy, instead of the hard-coded exponential backoff.

• Periodic keep-alives (with timer restarted every time bytes are received)
• Demand a keep-alive manually

When connecting the dnp3rs master to the OpenDNP3 outstation example, the startup integrity poll fails because it does not expect g50v4 (see #10) in the response, and then retries it immediatly. It creates a loop of request/response. Perhaps it should wait a couple of seconds before retrying that integrity poll.

Once stepfunc/oo_bindgen#42 is implemented, the Poll and Association should used an opaque struct instead of a class. This would prevent memory leaks and remove the misunderstanding of users for non-working polls if they let them be garbage-collected.

Optionally, specify the class assignements at startup (whenever a restart is detected).

Also, parameterize the link layer parser so that it bubbles up or swallows framing errors. TCP/TLS should bubble up errors and close the stream by default, but it shouldn't be hard coded by channel type (except for perhaps serial) b/c of the use of terminal servers which can convert serial to IP.

Let's get rid of the purple background and replace with something on-brand. I think the footer is an opportunity to integrate one of our nice illustration, perhaps with animation.