[PATCH v4 1/5] graph: add support for node specific errors

[Robin Jarry]

Hi Pavan,

, Aug 16, 2024 at 17:09:
> From: Pavan Nikhilesh <pbhagavatula at marvell.com>
>
> Add ability for Nodes to advertise error counters
> during registration.
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula at marvell.com>
> ---
> v2 Changes:
> - Fix compilation.
> v3 Changes:
> - Resend as 1/5 didn't make it through.
> v4 Changes:
> - Address review comments.
> - Rebase on main branch.
>
>  doc/guides/prog_guide/graph_lib.rst           |  22 +-
>  .../prog_guide/img/anatomy_of_a_node.svg      | 329 +++++--
>  .../prog_guide/img/graph_mem_layout.svg       | 921 +++++++++++++-----
>  lib/graph/graph_private.h                     |   1 +
>  lib/graph/node.c                              |  37 +-
>  lib/graph/rte_graph.h                         |   7 +
>  6 files changed, 1016 insertions(+), 301 deletions(-)
>
> diff --git a/doc/guides/prog_guide/graph_lib.rst b/doc/guides/prog_guide/graph_lib.rst
> index ad09bdfe26..018900caea 100644
> --- a/doc/guides/prog_guide/graph_lib.rst
> +++ b/doc/guides/prog_guide/graph_lib.rst
> @@ -21,6 +21,7 @@ Features of the Graph library are:
>  - Nodes as plugins.
>  - Support for out of tree nodes.
>  - Inbuilt nodes for packet processing.
> +- Node specific error counts.
>  - Multi-process support.
>  - Low overhead graph walk and node enqueue.
>  - Low overhead statistics collection infrastructure.
> @@ -124,6 +125,18 @@ Source nodes are static nodes created using ``RTE_NODE_REGISTER`` by passing
>  While performing the graph walk, the ``process()`` function of all the source
>  nodes will be called first. So that these nodes can be used as input nodes for a graph.
>
> +nb_errors:
> +^^^^^^^^^^
> +
> +The number of errors that this node can report. The ``err_desc[]`` stores the error
> +descriptions which will later be propagated to stats.
> +
> +err_desc[]:
> +^^^^^^^^^^^
> +
> +The dynamic array to store the error descriptions that will be reported by this
> +node.

This feature could be useful to store detailed statistics per node, not 
only for errors. It would be better to rename "errors" to "stats" or 
"xstats".

See below for a concrete suggestion.

> diff --git a/lib/graph/graph_private.h b/lib/graph/graph_private.h
> index d557d55f2d..e663b04d8b 100644
> --- a/lib/graph/graph_private.h
> +++ b/lib/graph/graph_private.h
> @@ -61,6 +61,7 @@ struct node {
>  	rte_node_t id;		      /**< Allocated identifier for the node. */
>  	rte_node_t parent_id;	      /**< Parent node identifier. */
>  	rte_edge_t nb_edges;	      /**< Number of edges from this node. */
> +	struct rte_node_errors *errs; /**< Node specific errors. */

How about:

    struct rte_node_xstats *xstats; /**< Node specific extra statistics. */

And maybe const? I didn't check if you make a full copy, or if you only 
copy the pointer.

> diff --git a/lib/graph/rte_graph.h b/lib/graph/rte_graph.h
> index ecfec2068a..b28143d737 100644
> --- a/lib/graph/rte_graph.h
> +++ b/lib/graph/rte_graph.h
> @@ -29,6 +29,7 @@ extern "C" {
>
>  #define RTE_GRAPH_NAMESIZE 64 /**< Max length of graph name. */
>  #define RTE_NODE_NAMESIZE 64  /**< Max length of node name. */
> +#define RTE_NODE_ERROR_DESC_SIZE 64  /**< Max length of node name. */

#define RTE_NODE_XSTATS_DESC_SIZE 64 /**< Max length of node xstats names. */

>  #define RTE_GRAPH_PCAP_FILE_SZ 64 /**< Max length of pcap file name. */
>  #define RTE_GRAPH_OFF_INVALID UINT32_MAX /**< Invalid graph offset. */
>  #define RTE_NODE_ID_INVALID UINT32_MAX   /**< Invalid node id. */
> @@ -460,6 +461,11 @@ void rte_graph_cluster_stats_get(struct rte_graph_cluster_stats *stat,
>   */
>  void rte_graph_cluster_stats_reset(struct rte_graph_cluster_stats *stat);
>
> +struct rte_node_errors {
> +	uint16_t nb_errors;				 /**< Number of errors. */
> +	char err_desc[][RTE_NODE_ERROR_DESC_SIZE];	 /**< Names of errors. */
> +};


struct rte_node_xstats {
    uint16_t xstats_num; /**< Number of xstats. */
    char xstats_desc[RTE_NODE_XSTATS_DESC_SIZE];  /**< Names of xstats. */
};

> +
>  /**
>   * Structure defines the node registration parameters.
>   *
> @@ -472,6 +478,7 @@ struct rte_node_register {
>  	rte_node_process_t process; /**< Node process function. */
>  	rte_node_init_t init;       /**< Node init function. */
>  	rte_node_fini_t fini;       /**< Node fini function. */
> +	struct rte_node_errors *errs; /**< Node specific errors. */

    struct rte_node_xstats *xstats; /**< Node specific extra statistics. */

Cheers!